Search

Real-time Hand Gesture Recognition (HGR) has emerged as a vital technology in human-computer interaction, offering intuitive and natural ways for users to interact with computer-vision systems. This comprehensive review explores the advancements, challenges, and future directions in real-time HGR. Various HGR-related technologies have also been investigated, including sensors and vision technologies, which are utilized as a preliminary step in acquiring data in HGR systems. This paper discusses different recognition approaches, from traditional handcrafted feature methods to state-of-the-art deep learning techniques. Learning paradigms have been analyzed such as supervised, unsupervised, transfer, and adaptive learning in the context of HGR. A wide range of applications has been covered, from sign language recognition to healthcare and security systems. Despite significant developments in the computer vision domain, challenges remain in areas such as environmental robustness, gesture complexity, computational efficiency, and user adaptability. Lastly, this paper concludes by highlighting potential solutions and future research directions trying to develop more robust, efficient, and user-friendly real-time HGR systems. [ABSTRACT FROM AUTHOR]

Recommended in this paper, because the existing single objective experience is poor, and the recommended model in a large difference of targets under the complex relationship of joint optimization and the conflict caused by faults, this paper proposes a personalized recommendation based on the deep learning multi-objective optimization algorithm, the estimated probability of users on the individual behavior as a model to study target, Multiple objectives are integrated into a model for learning. Firstly, the embedding layer is used to change the feature vectors, so that the bottom layer of the model shares the same feature embedding. Secondly, the factorization machine and deep learning are used to construct high-low order feature interaction. Then, the gating network and multilevel expert network constructed by a fully connected neural network are used to learn the characteristic relationship of user behavior. Finally, make connections between goals. Through experiments on public and real datasets, The results show that the multi-objective model proposed in this paper has better co-optimization performance and increases the AUC value by 0.1% compared with advanced personalized recommendation models such as MMoE and ESMM, to achieve the ultimate goal of increasing the prediction accuracy and improving user satisfaction. [ABSTRACT FROM AUTHOR]

Demand-Driven Material Requirements Planning (DDMRP) is an emerging inventory management approach that has garnered significant attention from academia and industry. Numerous recent studies have highlighted the advantages of DDMRP compared to traditional methods such as material requirement planning (MRP), Theory of constraint (TOC), and Kanban. However, the performance of DDMRP relies on several parameters that affect its effectiveness. Parameterization models and the optimization of control variables have significantly contributed to the field of inventory management and have proven to be effective and practical in addressing challenges by providing a structured approach to handling complex variables and constraints. This paper introduces an innovative parameterization model that leverages deep reinforcement learning (DRL) to parameterize a DDMRP system in the face of uncertain demand. The main objective is to dynamically determine the optimal values for the variability and lead time factors within the DDMRP framework, to maximize customer service levels and optimize inventory efficiency. The results of this study emphasize the effectiveness of DRL as an automated decision-making approach for controlling DDMRP parameters. Additionally, the findings highlight the potential for enhancing the performance of the DDMRP approach, particularly in terms of on-time delivery (OTD) and average on-hand inventory (AOHI) by adjusting the variability and lead-time factors. [ABSTRACT FROM AUTHOR]

On-road car accidents are immensely unfortunate but quite common occurrences worldwide. Instant data-centric and informed decisions of crisis management are rarely experienced due to the absence of real-time car accident detection and severity analysis mechanisms. On this background, the current paper presents a deep learning model for car accident detection and analysis of its severity so that the crisis management activities might follow without any delay saving invaluable human lives. The existing works lack in using time-series data, the proper learning model for accurate prediction, and minimizing the time taken in post-accident scenarios for the victims to receive immediate medical help. This paper introduces the Long Short Term Memory (LSTM) model in conjunction with the Gradient Boosted Regression Trees (GBRT) technique for the determination of car accidents with different levels of severity. The proposed model works with the accelerometer and gyroscopic data collected through an application installed in the smartphones of the users inside the car. The LSTM-GBRT hybrid model is proposed to achieve higher accuracy than LSTM which deals with time-variant data. The satisfactory performance of the proposed technique has been reported and the results are extensively investigated in comparison with another hybrid technique such as LSTM with Random Forest (RF) as well. The statistics confirm the superiority of the proposed model over other parallel models in terms of several performance metrics, like Accuracy, Precision, etc. [ABSTRACT FROM AUTHOR]

The widespread adoption of Electronic Healthcare Records has resulted in an abundance of healthcare data. This data holds significant potential for improving healthcare services by providing valuable clinical insights and enhancing clinical decision-making. This paper presents a patient classification methodology that utilizes a multiclass and multilabel diagnostic approach to predict the patient's clinical class. The proposed model effectively handles comorbidities while maintaining a high level of accuracy. The implementation leverages the MIMIC III database as a data source to create a phenotyping dataset and train the models. Various machine learning models are employed in this study. Notably, the natural language processing-based One-Vs-Rest classifier achieves the best classification results, maintaining accuracy and F1 scores even with a large number of classes. The patient diagnostic class prediction model, based on the International Classification of Diseases 9, showcased in this paper, has broad applications in diagnostic support, treatment prediction, clinical assistance, recommender systems, clinical decision support systems, and clinical knowledge discovery engines. [ABSTRACT FROM AUTHOR]

Early kidney stone detection is essential for the diagnosis and treatment of people who have kidney stones. The objective of this study is to employ deep learning algorithms for renal stone detection, addressing the critical need for early, accurate diagnosis, which can significantly improve patient outcomes and reduce healthcare costs. The paper thoroughly assesses a variety of models, including ResNet, DenseNet, and EfficientNet, for CT images. The limitations of manual identification procedures highlight the urgent need for a more effective automated approach, making this research necessary. Notably, the painstakingly improved DenseNet model achieves a peak accuracy of 0.86, demonstrating its potential superiority. These results convincingly demonstrate the revolutionary power of deep learning, which is poised to revolutionise the detection of renal stones. This fast, trustworthy, and non-invasive method has the potential to advance clinical procedures and significantly improve patient care. [ABSTRACT FROM AUTHOR]

In this study, we obtain a sex prediction algorithm based on CNN in two ways - building a red Convolutional Neural Network (CNN) model from scratch and via transfer learning. We built a model from scratch and compared it with fine-tuned EfficientNetB1. We use these models for gender determination using periocular images and compare the two models depending on the accuracy of the models. The CNN model proposed from scratch yields an accuracy of 94.46% while the fine-tuned EfficientNetB1 yields an accuracy of 97.94%. This paper is one of the first works in determining gender from periocular images in the visible spectrum using a CNN model built from the outset. [ABSTRACT FROM AUTHOR]

Many science, technology and innovation (STI) resources are attached with several different labels. To assign automatically the resulting labels to an interested instance, many approaches with good performance on the benchmark datasets have been proposed for multilabel classification task in the literature. Furthermore, several open-source tools implementing these approaches have also been developed. However, the characteristics of real-world multilabel patent and publication datasets are not completely in line with those of benchmark ones. Therefore, the main purpose of this paper is to evaluate comprehensively seven multi-label classification methods on real-world datasets. Three real-world datasets (Biological-Sciences, Health-Sciences, and USPTO) from SciGraph and USPTO database are constructed. Seven multilabel classification methods with tuned parameters (dependency-LDA, MLkNN, LabelPowerset, RAkEL, TextCNN, TexRNN, and TextRCNN) are comprehensively compared on these three real-world datasets. To evaluate the performance, the study adopts three classification-based metrics: Macro-F1, Micro-F1, and Hamming Loss. The TextCNN and TextRCNN models show obvious superiority on small-scale datasets with more complex hierarchical structure of labels and more balanced documentlabel distribution in terms of macro-F1, micro-F1 and Hamming Loss. The MLkNN method works better on the larger-scale dataset with more unbalanced document-label distribution. Three real-world datasets differ in the following aspects: statement, data quality, and purposes. Additionally, open-source tools designed for multi-label classification also have intrinsic differences in their approaches for data processing and feature selection, which in turn impacts the performance of a multi-label classification approach. In the near future, we will enhance experimental precision and reinforce the validity of conclusions by employing more rigorous control over variables through introducing expanded parameter settings. The observed Macro F1 and Micro F1 scores on real-world datasets typically fall short of those achieved on benchmark datasets, underscoring the complexity of real-world multi-label classification tasks. Approaches leveraging deep learning techniques offer promising solutions by accommodating the hierarchical relationships and interdependencies among labels. With ongoing enhancements in deep learning algorithms and large-scale models, it is expected that the efficacy of multi-label classification tasks will be significantly improved, reaching a level of practical utility in the foreseeable future. (1) Seven multi-label classification methods are comprehensively compared on three real-world datasets. (2) The TextCNN and TextRCNN models perform better on small-scale datasets with more complex hierarchical structure of labels and more balanced document-label distribution. (3) The MLkNN method works better on the larger-scale dataset with more unbalanced document-label distribution. [ABSTRACT FROM AUTHOR]

Rotor-bearing systems are important components of rotating machinery and transmission systems, and imbalance and misalignment are inevitable in such systems. At present, the main challenges faced by state-of-the-art fault diagnosis methods involve the extraction of fault features under strong background noise and the classification of different fault modes. In this paper, a fault diagnosis method based on an improved deep residual shrinkage network (IDRSN) is proposed with the aim of achieving end-to-end fault diagnosis of a rotor-bearing system. First, a method called wavelet threshold denoising and variational mode decomposition (WTD-VMD) is proposed, which can process original noisy signals into intrinsic mode functions (IMFs) with a salient feature. These one-dimensional IMFs are then transformed into two-dimensional images using a Gramian angular field (GAF) to give datasets for the deep residual shrinkage network (DRSN), which can achieve high levels of accuracy under strong background noise. Finally, a comprehensive test platform for a rotor-bearing system is built to verify the effectiveness of the proposed method in the field. The true test accuracy of the model at a 95% confidence interval is found to range from 84.09% to 86.51%. The proposed model exhibits good robustness when dealing with noisy samples and gives the best classification results for fault diagnosis under misalignment, with a test accuracy of 100%. It also achieves a higher testing accuracy compared to fault diagnosis methods based on convolutional neural networks and deep residual networks without improvement. In summary, IDRSN has significant value for deep learning engineering applications involving the fault diagnosis of rotor-bearing systems. [ABSTRACT FROM AUTHOR]

Automatic crack detection in construction facilities is a challenging yet crucial task. However, existing deep learning (DL)-based semantic segmentation methods for this field are based on fully supervised learning models and pixel-level manual annotation, which are time-consuming and labor-intensive. To solve this problem, this paper proposes a novel crack semantic segmentation network using weakly supervised approach and mixed-label training strategy. Firstly, an image patch-level classifier of crack is trained to generate a coarse localization map for automatic pseudo-labeling of cracks combined with a thresholding-based method. Then, we integrated the pseudo-annotated with manual-annotated samples with a ratio of 4:1 to train the crack segmentation network with a mixed-label training strategy, in which the manual labels were assigned with a higher weight value. The experimental data on two public datasets demonstrate that our proposed method achieves a comparable accuracy with the fully supervised methods, reducing over 65% of the manual annotation workload. [ABSTRACT FROM AUTHOR]

In the context of rapid developments in drone technology, the significance of recognizing and detecting low-altitude unmanned aerial vehicles (UAVs) has grown. Although conventional algorithmic enhancements have increased the detection rate of low-altitude UAV targets, they tend to neglect the intricate nature and computational demands of the algorithms. This paper introduces ATD-YOLO, an enhanced target detection model based on the YOLOv5s architecture, aimed at tackling this issue. Firstly, a realistic low-altitude UAV dataset is fashioned by amalgamating various publicly available datasets. Secondly, a C3F module grounded in FasterNet, incorporating Partial Convolution (PConv), is introduced to decrease model parameters while upholding detection accuracy. Furthermore, the backbone network incorporates an Efficient Multi-Scale Attention (EMA) module to extract essential image information while filtering out irrelevant details, facilitating adaptive feature fusion. Additionally, the universal upsampling operator CARAFE (Content-aware reassembly of features) is utilized instead of nearest-neighbor upsampling. This enhancement boosts the performance of the feature pyramid network by expanding the receptive field for data feature fusion. Lastly, the Slim-Neck network is introduced to fine-tune the feature fusion network, thereby reducing the model's floating-point calculations and parameters. Experimental findings demonstrate that the improved ATD-YOLO model achieves an accuracy of 92.8%, with a 31.4% decrease in parameters and a 28.7% decrease in floating-point calculations compared to the original model. The detection speed reaches 75.37 frames per second (FPS). These experiments affirm that the proposed enhancement method meets the deployment requirements for low computational power while maintaining high precision. [ABSTRACT FROM AUTHOR]

With the rapid development of computer vision and artificial intelligence technologies, indoor scene reconstruction has been more and more widely used in the fields of virtual reality, augmented reality and architectural design. In this paper, we study an indoor scene reconstruction method based on the 3DGS model, which has been widely used in computer graphics and vision processing with powerful scene representation and rendering capabilities. In this study, we optimize the 3DGS model to enhance the detail preservation and realism of the reconstruction results by adjusting the opacity of the Gaussian function. We used the Replica dataset and the self-harvested dataset for model training. Through experimental validation, the peak signal-tonoise ratio as well as the structural similarity ratio of the reconstruction results of the optimized model have an improvement effect of more than 1%, which indicates that the optimized model has a significant improvement in detail retention and realism, and the reconstructed scene performs more realistically in terms of texture details and light and shadow effects. [ABSTRACT FROM AUTHOR]

In the infrared imaging scene where the target is at a long distance and the background is cluttered, due to the interference of noise and background texture information, the infrared image is prone to problems such as low contrast between the target and the background, and feature confusion, which makes it difficult to accurately extract and detect the target. To solve this problem, firstly, the infrared image is enhanced by combining DDE and MSR algorithm to improve the contrast and detail visibility of the image. For the RT-DETR network structure, the EMA attention mechanism is introduced into the backbone to enhance the feature extraction ability of the model by extracting context information. The CAMixing convolutional attention module is introduced into CCFM, and the multi-scale convolutional self-attention mechanism is introduced to focus on local information and enhance the detection ability of small targets. The filtering rules of the prediction box are improved, combined with Shape-IoU, and the convergence speed of the loss function in the detection and the detection accuracy of small targets are improved by paying attention to the influence of the intrinsic properties of the bounding box itself on the regression. In the experiment, the infrared weak target image dataset of the National University of Defense Technology was selected, labeled and trained. Experimental results show that compared with the original DETR algorithm, the average precision of the improved algorithm (mAP) is increased by 3.2%, and it can effectively detect infrared weak and small targets in different complex backgrounds, which reflects good robustness and adaptability, and can be effectively applied to infrared weak and small target detection in complex backgrounds. [ABSTRACT FROM AUTHOR]

Urine sediment examination (USE) is an essential aspect in detecting urinary system diseases, and it is a prerequisite for diagnostic procedures. Urine images are complex, containing numerous particles, which makes a detailed analysis and interpretation challenging. It is crucial for both patients and medical professionals to conduct urine analysis automatically, quickly and inexpensively, without compromising reliability. In this paper, we present a deep multi-modal fusion system, commonly employed in artificial intelligence, capable of automatically distinguishing particles in urine sediment. To achieve this objective, we first created a new dataset comprising erythrocytes, leukocytes, yeast, epithelium, bacteria, crystals, cylinders, and other particles (such as sperm). The data were gathered from urinalysis requests made between July 2022 and September 2022 at the biochemistry laboratory of Fethi Sekin Medical Center Hospital. A dataset containing 8509 images was compiled using the Optika B293PLi microscope with trinocular brightfield. We propose a 5-step process for detecting particles in the dataset using a multi-modal fusion deep learning model: i) The obtained images were augmented by applying affine transformation. ii) To distinguish images, we opted for ResNet18 and ResNet50 models, which yielded high performance in medical data. iii) Feature vectors from both models were fused to generate more consistent, accurate, and useful particle features. iv) We employed ReliefF, Neighborhood Component Analysis (NCA), and Minimum-Redundancy Maximum-Relevancy (mRMR) feature selection methods, widely used to determine features that maximise particle discrimination success. v) In the final step, Support Vector Machine (SVM) was utilised to distinguish the particles. The results demonstrate that the highest accuracy value achieved is 98.54 % when employing the ReliefF algorithm. Contributions of the study include eliminating standardisation differences in manual microscopy, achieving high accuracy in particle discrimination, offering an artificial intelligence-based system applicable in laboratory environments, and providing the dataset as educational and practical material for biochemistry professionals. [ABSTRACT FROM AUTHOR]

An efficient face recognition system is essential for security and authentication-based applications. However, real-time face recognition systems have a few significant concerns, including face pose orientations. In the last decade, numerous solutions have been introduced to estimate distinct face pose orientations. Nevertheless, these solutions must be adequately addressed for the three main face pose orientations: Yaw, Pitch, and Roll. This paper proposed a novel deep transfer learning-based multitasking approach for solving three integrated tasks, i.e., face detection, landmarks detection, and face pose estimation. The face pose variation vulnerability has been intensely investigated here underlying three modules: image preprocessing, feature extraction module through deep transfer learning, and regression module for estimating the face poses. The experiments are performed on the well-known benchmark dataset Annotated Faces in the Wild (AFW). We evaluate the outcomes of the experiments to reveal that our proposed approach is superior to other recently available solutions. [ABSTRACT FROM AUTHOR]

The current research is dedicated to harnessing cutting-edge technologies within the paradigm of Industry 5.0. The objective is to capitalize on advancements in Machine and Deep Learning techniques. This research endeavors to construct robust predictive models, utilizing historical data, for precise real-time predictions in estimating material quantities within a cement workshop. Machine Learning regressors evaluated based on several metrics, SVR (R-squared 0.9739, MAE 0.0403), Random Forest (R-squared 0.9990, MAE 0.0026), MLP (R-squared 0.9890, MAE 0.0255), Gradient Boosting (R-squared 0.9989, MAE 0.0042). The time series models LSTM and GRU yielded R-squared 0.9978, MAE 0.0100, and R-squared 0.9980, MAE 0.0099, respectively. The ultimate outcomes include improved and efficient production, optimization of production processes, streamlined operations, reduced downtime, mitigation of potential disruptions, and the facilitation of the factory's evolution towards intelligent manufacturing processes embedded within the framework of Industry 5.0. These achievements underscore the potential impact of leveraging advanced machine learning techniques for enhancing the operational dynamics and overall efficiency of manufacturing facilities [ABSTRACT FROM AUTHOR]

Self-regulated learning is a multifaceted process that involves cognitive, metacognitive, motivational, affective, and behavioural components. These components are interconnected and influenced by contextual factors that shape the overall learning experience. Self-regulated learning empowers students to take control of their learning by utilising effective learning techniques, engaging in metacognitive processes, and fostering motivation and self-efficacy beliefs. The goal of the article is to explore the importance of self-regulated learning in enhancing students' academic performance and fostering a deep understanding of the learning material. Specifically, this research article explores the utilisation of learning techniques and strategies (LTSs) among pedagogy students, investigating their patterns and preferences for self-studying. The analysis of survey responses from a sample of 76 pedagogy students revealed that the most commonly used LTSs are rereading, note-taking, memorising, and highlighting. The utilisation of LTSs that are empirically considered effective, such as distributed practice, practice testing, and elaborative interrogation, was found to be extremely low. Moreover, the study revealed a lack of comprehensive training in learning techniques and strategies in pedagogy programmes, suggesting that current teachers are not adequately trained to create optimal classroom contexts for developing flexible self-regulatory skills in students. The findings highlight that by incorporating these practices into curriculum, future educators can empower their students to become active learners who select the appropriate strategies and reflect on their learning process. This paper contributes to an understanding of self-regulated learning and the implications of self-regulated learning for educational practices. [ABSTRACT FROM AUTHOR]

In a Multiple-Input Multiple-Output Orthogonal Frequency Division Multiplexing (MIMO-OFDM) method, multiple antennas can be used on either the transmitter or receiver end to improve the system capacity, data throughput, and robustness. OFDM has been used as the modulation system that divides the data stream into multiple parallel low-rate subcarriers. MIMO enhances the system by utilizing spatial diversity and multiplexing abilities. Modulation classification in the MIMO-OFDM systems describes the process of recognizing the modulation scheme used by the communicated signals in a MIMO-OFDM communication system. This is a vital step in receiver design as it enables proper demodulation of the received signals. In this paper, an Enhanced Modulation Classification Approach using an Arithmetic Optimization Algorithm with Deep Learning (EMCA-AOADL) is developed for MIMO-OFDM systems. The goal of the presented EMCAAOADL technique is to detect and classify different types of modulation signals that exist in MIMO-OFDM systems. To accomplish this, the EMCA-AOADL technique performs a feature extraction process based on the Sevcik Fractal Dimension (SFD). For modulation classification, the EMCA-AOADL technique uses a Convolution Neural Network with Long Short-Term Memory (CNN-LSTM) approach. Finally, the hyperparameter values of the CNN-LSTM algorithm can be chosen by using AOA. To highlight the better recognition result of the EMCA-AOADL approach, a comprehensive range of simulations was performed. The simulation values illustrate the better results of the EMCA-AOADL algorithm. [ABSTRACT FROM AUTHOR]

Human activity recognition (HAR) by deep learning is a challenging and interesting topic. Although there are robust models, there is also a bunch of parameters and variables, which affect the performance such as the number of layers, pooling type. This study presents a new deep learning architecture that is obtained by fine-tuning of the conventional CNN-LSTM model, namely, CNN (+3)-LSTM. Three changes are made to the conventional model to increase the accuracy. Firstly, kernel size is set to 1×1 to extract more information. Secondly, three convolutional layers are added to the model. Lastly, average pooling is used instead of max-pooling. Performance analysis of the proposed model is conducted on the KTH dataset and implemented on Keras. In addition to the overall accuracy of the proposed model, the contribution of each change is observed individually. Results show that adding layers made the highest contribution followed by kernel size and pooling, respectively. The proposed model is compared with state-of-art and outperformed some of the recent studies with a 94.1% recognition rate. [ABSTRACT FROM AUTHOR]

Jangal contains the rich literary value and aesthetic value, which has important research significance. Through the application of deep learning technology in graphic semantic analysis, this paper specifically describes the extraction method of graphic features in the children’s picture books of “Jianggeer”. Then, the BERT framework is used as the basis to construct the graphic semantic analysis model JCDB-BRET in this paper, and it is applied to the actual analysis of the graphic semantics of the children’s picture books of “Jianggeer” after training. In this paper, the model is used to recognize all the images in the picture book and branch out their semantics to be expressed, and the proportion of geographic images in the picture book is the highest at 36.55%. The rationality and pursuit of a harmonious home in “Jianggeer” are shown through the visualization of Benba Ledu, where Jianggeer lives. The textual semantics focuses on analyzing the praise words in the picture book of Jianggeer, and the description of the “fragrance” on Jianggeer’s body and the “warm fragrance” on his neck are used as examples to elucidate the aesthetic tendency of the Mongolian people embodied in the text.

In the context of the increasingly severe world climate form, how to rationally arrange and dispatch energy has become an urgent need. This paper proposes a deep learning-based power system scheduling model based on the concept of perfect scheduling, using GRU to learn scheduling data. A different training set is constructed to train the model according to the load characteristics at different moments, and the model is updated in real time based on the data at the current moment. The analysis of the algorithms reveals that the scheduling error rate of this model ranges from −-3% to 2%, and the average RMSE of the scheduling scheme is 2.72, placing it in close proximity to the optimal scheduling strategy. Due to a 6.5% reduction in scheduling cost compared to the average cost of the two analyzed algorithms, the average time reduction is 76.3%. The scheduling optimization model proposed in this paper exhibits excellent performance.

Load forecasting is a critical component of time series simulation in power systems, essential for the reliability and accuracy of simulations. With the integration of renewable energy sources such as photovoltaics, power systems face increasingly complex load forecasting challenges. This paper introduces a deep learning approach that combines Long Short-Term Memory networks (LSTM) and Attention Mechanisms (AM) to enhance the precision and reliability of load forecasting in power distribution networks. Utilizing electric load data from a specific region in China, the LSTM-AM model captures long-term dependencies in time-series data and highlights the impact of critical periods on forecasting accuracy. Experimental results demonstrate that the LSTM-AM model surpasses traditional Back Propagation neural networks, CNNs, and standard LSTM models in terms of prediction precision, affirming the potential application of the proposed method in the field of electric load forecasting. Moreover, the paper introduces a similar day selection strategy to distinguish between weekdays and weekends, reducing RMSE and MAE from 22.6 MW and 15.1 MW to 20.1 MW and 13.9 MW, respectively, thereby further optimizing the accuracy of the model

With the reform and development of modern education curriculum, the focus of preschool education has changed from focusing on the “learning results” to the “learning process” of young children, and deep learning, as a hot trend of education reform, is an important way to implement education reform in pre-school education, which is based on problem-solving, emphasizing on the application of migration, and focusing on the reflective construction. Deep learning, a trend of educational reform that is based on problem-solving, emphasizes transfer and application, and focuses on reflection and construction, is an important way to implement educational reform in preschool education. This study first explores the connection between preschool education and deep learning, collects the problems of deep learning in preschool education through a questionnaire survey, and builds a personalized preschool education model based on deep learning in combination with DELC theory. The teaching effect of the deep learning preschool education model is explored through educational comparison experiments in this paper. The mean value of the preschool education model of this paper compared with the traditional preschool education in the four dimensions of learning motivation, learning input, learning strategy, and learning outcome were increased by 2.00~4.00 points, respectively, and the Sig values were all less than 0.05. Meanwhile, the average grade of this paper’s teaching model was 65.124 points, which was increased by 3.112 points compared with the average grade of the traditional cumulative preschool education model, proving that this paper’s effect of the application of deep learning personalized preschool education mode, which has a significant advantage in the degree of deep learning and learning achievement. In addition, this study provides suggestions for the development of deep learning in preschool education from the levels of teachers, teaching materials, teaching curricula, and external support so as to promote the process of deep learning in preschool education.

In this paper, a credit scoring integration model incorporating BRNN is used to study the credit scoring problem in automobile finance. Aiming at the problems of existing credit scoring models constructed with shallow architecture and the unidirectional limitation of RNN itself, this paper introduces a BRNN model that superimposes RNN models in two directions. The potential relationship between each credit feature is mined through logistic regression, extreme gradient boosting tree, and bidirectional recurrent neural network algorithms, and the final prediction output is linked to the customer’s overall credit to improve the prediction accuracy. In this paper, we study the application of a credit scoring model based on the improved BRNN model for an auto finance company. Data preprocessing techniques and feature screening methods are used to improve the BRNN model and construct the credit scoring model for auto finance at Company A. The BRNN model is the basis for Company A’s credit scoring model. Based on the comparison with other models, it is concluded that the automobile finance credit scoring IBRNN model constructed based on the improved BRNN model in this paper has an accuracy of 89.6% in classifying the user finance data of Company A on different datasets, which is a significant improvement compared with the other five models.

In recent years, as deep learning has demonstrated powerful characterization capabilities in the fields of speech, image, and text, researchers have begun to apply it to the field of prediction, i.e., predicting customer behaviors through current interaction records and features. This paper proposes a deep wandering-based customer behavior prediction model that combines deep learning techniques to forecast customer behavioral trends in the cultural industry. The model randomly wanders from the social network graph structure of the customer’s purchase of goods to generate a new behavioral sequence. We regard the user’s behavioral sequence as a word, and we pre-train all the behavioral sequence documents using the Word2vec algorithm model. The experimental comparison revealed that the model, which incorporates the depth-wandering technique, outperforms other models on the test set in terms of predictiveness. The website uses the deep wandering user behavior prediction model to forecast sales and adapts its sales strategy based on the customer’s behavior. 31% of customers were content with the books they bought from the website, while 52% were extremely content. By comparing the book sales before and after applying the model, it was found that the book sales increased significantly after adjusting the sales strategy, indicating that the customer behavior prediction model constructed in this paper can be used practically.

In recent years, the development of Guangxi’s national folk dance has been on the rise and has gained much attention. The research of Guangxi’s national folk dance is currently in a booming period. The research is based on deep learning theory, using stack denoising autoencoder and convolutional depth Boltzmann mechanism to build a SdAE-CDBM model for dance movement classification. The dance movements are recognized and detected by using feature mining and extraction of dance movements in Guangxi folk dance videos. The SdAE-CDBM model of this paper is compared with other classification models in terms of semantic classification accuracy of dance movements to explore the classification performance of the SdAE-CDBM model proposed in this paper. The average F1 values of the SdAE-CDBM model in the classification of the seven types of dance movements are 86.77%, 88.54%, and 90.18%, respectively, which are the maximum values among the movement classification models. The SdAE-CDBM model was able to achieve the highest classification accuracy and the fastest classification convergence speed among all classification models. When it comes to classifying dance movements semantically, the SdAE-CDBM model achieves a classification accuracy of 70.28%, which is significantly superior to other classification models. The SdAE-CDBM model in this paper is highly effective in the semantic classification of dance movements, as evidenced by this.

This paper proposes the integration method of athlete training volume monitoring information based on deep learning, preprocessing the monitoring information using wavelet transform, obtaining the athlete training volume information after denoising and dimensionality reduction, combining with the theory of deep learning, utilizing convolutional neural network for feature extraction and integration of the processed information, and performing simulation test and analysis. The average Gini coefficient is 84.926, which proves the effectiveness of the method used in this paper. After exercising, the athletes’ weight and body fat rate were monitored to reduce a certain degree, and their lung capacity was improved. A deep learning algorithm generated personalized exercise program training achieved a heart rate interval of 94% for safe and effective workouts. The three stages of the test subjects’ training were tracked and analyzed, and the use of training monitoring can effectively promote the efficiency of training. The physical fitness program (P < 0.01) and physical fitness test (P < 0.05) in the first stage compared with the third stage showed significant improvement.

With the deepening of higher vocational English teaching research and the continuous advancement of reform, this paper combines in-depth learning with higher vocational English teaching and makes an in-depth study of higher vocational English teaching practice based on in-depth learning theory and learning theory. First of all, the paper expounds on the concept of higher vocational English teaching and the research of English teaching. Explain the related concepts of psychological factors and the composition of psychological factors, then describe the model of in-depth learning, elaborate the related algorithms of in-depth learning optimization methods and the functions of various data of in-depth learning-related research, and finally, based on in-depth learning, explain the psychological factors of teachers and students in higher vocational English teaching. It reflects the importance of psychological factors in higher vocational English teaching. With the development of in-depth learning, we should apply in-depth learning to higher vocational English teaching and optimize the teaching mode of in-depth learning reasonably, including formulating comprehensive higher vocational English teaching objectives, adjusting students' psychological status reasonably, and carrying out cooperative learning.

In the context of practical education, deepening the reform of quality education and implementing the fundamental task of cultivating moral integrity have become the guidelines and the basis for the development of curriculum reform in colleges and universities. This study conducts data mining based on the association rule model on the influencing factors of the reform of the Civic and Political Theory Course in colleges and universities to obtain the key influencing factors in the process of the reform of the theory course. Then, under the perspective of deep learning, the idea of practical parenting is integrated, and the structural framework of Civic and Political Theory course reform is designed from four directions: learning theme, learning objectives, learning activities, and evaluation feedback, and finally, the course reform method of this paper is examined through teaching practice. The results show that the influence weights of self-factors, school education factors, family factors, and social factors are 0.3022, 0.3521, 0.1381, and 0.2076, respectively, with self-factors and school education as the key influencing factors. After the teaching practice, the P-value of each dimension of the students under the curriculum reform based on this paper is less than 0.05, and the grade of the Civics and Political Science Theory course is improved from 81.47 to 90.45, which is an increase of 7.49 points compared with the class of the regular course. The reform method based on deep learning in this study can promote students’ development and improve their academic performance, which provides a reference for the research on the reform and development of the Civics and Political Science Theory course.

In recent years, with the continuous development of the tourism industry, changing the education model has become a fundamental problem that must be solved to cultivate the quality of professional talents, and the OBE education model has become the trend of higher education development. In this paper, the program design of tourism management mode based on OBE education is carried out in two aspects of personality characteristics cognition and personality learning resources, then constructs the learner characteristics model from several aspects of style similarity, behavioral similarity, learner influence relationship, and proposes the personality recommendation method based on deep learning after the selected experimental objects are experimental class and control class respectively, and through the analysis of the cognitive level of the students, learning style, interaction behavior characteristics were analyzed to verify the validity of the learning characteristic model and program design constructed in this paper. After the experiment, the average post-test score of the experimental group was 84.91, which is higher than the average post-test score of the C1 group, which is 83.63, indicating that the OBE education model based on the deep learning recommendation can improve the student’s learning performance.

Deep learning technology has the characteristics of automatic learning, nonlinear modeling, etc., which makes the machine algorithm better adapt to the data so as to better complete the task of sentiment analysis. The purpose of this paper is to explore the application of deep learning technology in the emotional analysis of art design works. After combining the emotion construction mechanism of art design works, based on the superior performance of convolutional neural network image recognition, it is introduced into the emotion analysis of art design works, combined with the attention mechanism model, to construct the emotion analysis model of art design works based on deep learning. Art design works are collected as the experimental dataset. The emotion analysis effect of this paper’s model for art design works is evaluated through the comparison of emotion classification indexes. The comparison of ROC curves of different models, and the emotion instance analysis of art design works is carried out. The experiments show that the experimental accuracy, precision, recall, and F1 value of this paper’s model range from 81% to 86%, and the F1 and ROC values are improved by 2.3% to 7.0% and 1.7% to 4.2% compared with the comparison model, which reflects the high accuracy and precision of this paper’s model for the sentiment analysis of art design works. Seven emotional patterns were found in tea art design works, and the model can be used to analyze the emotions of art design works.

Movement disorders are the main symptoms of neurological diseases such as Parkinson’s disease and deep learning-related methods can provide some intelligent solutions for the assessment and diagnosis of Parkinson’s movement disorders. In this paper, we propose a Kinect-based movement disorder assessment and analysis method, which uses the Kinect algorithm to capture and inverse kinematics analysis of human skeletal points, and further suggests the study of movement disorder assessment method based on dynamic time regularization algorithm so as to further achieve the effect of movement disorder assessment. Through the clinical experimental research on Parkinson’s disease patients and healthy subjects of the same age group, the use of the algorithm proposed in this paper is 15.18% higher than the GaitSet method in the CL state. The error of the algorithm proposed in this paper in the experiments comparing the gait parameter with the gold-standard motion capture system is close to 0.03s, which is a better improvement and upgrade compared with the advanced skeleton-based methods. In summary, the algorithm proposed in this paper is valuable and feasible for use in the assessment of Parkinson’s dyskinesia.

Traditional basketball free-throw hitting probability prediction mainly relies on naked eye vision to analyze basketball players, which makes it difficult to achieve stable and accurate free-throw hitting probability prediction. In this study, we propose a deep learning-based algorithm for recognizing basketball free throw poses and predicting the probability of hitting free throws, and we build a corresponding system framework. DetectNet is used to screen free throw motion detectors, real-time estimate and recognition of free throw poses, and code rewriting. At the same time, Open Pose is used as a human skeletal joint point detector to extract the joint point information of the free throw shooter. The joint points with wrong information are repaired, and the classification prediction of the free throw hitting result is realized based on the support vector machine. In the performance simulation experiments, the average accuracy of the free throw pose recognition method proposed in this paper is as high as 98.55% on the MSR Action 3D dataset, and the recognition rate of the elbow lift pose, which is the most difficult to recognize, is also higher than that of other comparative algorithms. The hit probability prediction method is also 12.64% and 6.25% more accurate than the OpenPose algorithm in hit-and-miss prediction, and the accuracy has also improved by 4.47% and 5.41%, respectively. The free throw pose recognition and hit probability prediction method in this paper has excellent performance.

The continuous updating of deep learning algorithms and theories has laid a solid foundation for the development of the machine vision field. Automated detection technology has become a hot topic of research in the field of machine vision in recent years. In this paper, we first introduce the traditional one-dimensional and two-dimensional image segmentation algorithm and then optimize the image segmentation algorithm by combining the conventional pigeon flocking algorithm and chaotic search algorithm so as to obtain a more accurate detection target image. Then, on the basis of a deep learning network, an attention mechanism is introduced to construct a Swin-Transformer image detection model to realize automatic detection of machine vision. Finally, the performance of the model is tested, and it is applied to watermelon seedling quality detection to explore its application value. The results show that in the performance test experiment of the image segmentation algorithm of this paper, the three indexes of F1, IoU, and accuracy of the image segmentation algorithm designed in this paper on the ISIC-2020 dataset are 93.90%, 93.74%, and 98.37%, which are ranked the first among the algorithms participating in the experiment. The precision, recall, and mAP values of the image detection model designed in this paper are 92.87%, 77.13%, and 83.21% on the experimental data test set, which are higher than those of other models participating in the experiment. The image detection model designed in this paper was practically applied to watermelon seedling quality detection. The accuracy of the model in detecting the presence or absence of diseased spots and cotyledon area, two key characteristic parameters of seedling quality, reached 100%. The model showed high reliability in practical application. The image segmentation algorithm and image detection model developed in this paper are highly useful in automated detection.

The era of big data produces massive data, and carrying out data mining can effectively obtain effective information in huge data, which provides support for efficient decision-making and intelligent optimization. The purpose of this paper is to establish a digital twin system, preprocess massive data using random matrix theory, and design the knowledge graph construction process based on digital twin technology. The BERT model, attention mechanism, BiLSTM model, and conditional random field of the joint deep learning technology are used to identify the knowledge entities in the digital twin system, extract the knowledge relations through the Transformer model, and utilize the TransE model for the knowledge representation in order to construct the knowledge graph. Then, the constructed knowledge graph is combined with the multi-feature attention mechanism to build an anomaly data prediction model in the digital twin system. Finally, the effectiveness of the methods in this paper is validated through corresponding experiments. The TransE model is used for knowledge representation. The accuracy of ternary classification is higher than 80% in all cases, and the MR value decreases by up to 64 compared to the TransR model. The F1 composite score of the anomaly data prediction model is 0.911, and the AUC value of the validation of knowledge graph effectiveness is 0.702. Combining deep learning with the knowledge graph, the knowledge information can be realized in the digital twin system’s accurate representation and enhance the data mining ability of the digital twin system.

Scheduling data of power plant systems plays a pivotal role in grid security. In this paper, the intrusion detection model IDP-TSW is constructed by using deep learning technology, feature extraction of raw traffic data based on density peak clustering algorithm and control variable method, and the final classification is realized by softmax. After completing the intrusion detection, for the network malicious intrusion situation further proposed the security protection strategy selection model HMS-BAG based on the Bayesian attack graph, formalized the description of the protection strategy selection problem, and proposed the optimal security protection strategy selection algorithm based on PSO. In the network intrusion detection performance experiments, the performance of the IDP-TSW model proposed in this paper outperforms the Bi-LSTM and CNN+Bi-LSTM models in terms of accuracy, recall, and F1 value. The F1 values for determining normal and abnormal data are 96.57% and 95.75%, respectively, and the precision and recall are also higher than 94%. Detecting Generic and Reconnaissance attacks is more than 90% accurate, but Dos attacks and others are relatively absent. In the network security defense performance experiment, the proposed HMS-BAG model achieved a defense success rate of 94.2% and a defense gain of 170.68.

With the improvement of the quality of human life, various industries utilize deep learning technology to meet the needs of users. In this paper, after exploring the deep temporal model and deep forest algorithm (DF) model in extracting the characteristics of user behavior features, we propose a user behavior prediction model based on CNN-LSTM and add a front embedding layer as well as a feature fusion layer to improve it and increase the prediction accuracy. The weighted average method is used to integrate CNN-LSTM with DF for algorithmic model fusion, aiming to improve the robustness and stability of the model and achieve accurate predictions of user behavior. In the empirical analysis, the correct rate of the fusion model proposed in this paper exceeds the proper rate of several other models by 11.75-15.73%, and it can reach 11.2% recall at k=20. Meanwhile, the lower the user behavior level is, the higher the prediction accuracy of the CNNLSTM+DF algorithm is, which can reach up to 95.20%. The lower the average relative error and average absolute error are, which can reach 13.45% and 3.92min respectively, which verifies the validity of the fusion model proposed in this paper, and provides a reference for the research in the related fields.

At present, most graphic design methods of 3D animation scenes only get a small part of 3D animation scenes, and all of them are in a 3D coordinate system, with observers as the core, so it is difficult to express the depth information of 3D animation scenes. This project intends to study a plane design method for 3D animation scenes based on deep learning. CNN (Convolutional Neural Network) is used to build a multi-view 3D animation scene generation network, and 3D geometry and structure of objects are reconstructed through multiple or a group of images. On this basis, the feature extraction method of 3D animation scenes is studied, and the collaborative learning model of multiple networks is established to improve the modeling accuracy of 3D animation scenes. The experimental results show that the network model is superior to the method based on multi-view and 0-1 voxel in detecting retrieval performance, and the accuracy rate can reach 91.725%. The multi-view 3D animation scene generation method in this paper has achieved better results than the current advanced methods, which proves that the multi-view feature fusion network proposed in this paper is a more reasonable method to fuse multi-view image features.

The visual system has a strong information processing ability, and visual attention tracking has various applications in various scenes. This paper mainly focuses on the sports scene of ice and snow far mobilization. It constructs a visual attention system model based on eye tracking. It first establishes an eye tracking system framework using deep learning, and improves the gaze estimation by optimizing the feature extraction network. The visual attention system model was constructed using particle filtering based on motion feature cognition. In the eye-tracking visual attention system model experiments, the Accuracy of the improved eye-tracking system in this paper can be significantly improved to 1.13°, and the error of the visual attention system can be kept within 10°. Furthermore, the four ice and snow sports scene types have an average accuracy of 85.47%, and the constructed model performs well. This study offers a guide for effectively combining eye tracking technology and visual attention.

In the contemporary educational environment, the protection and inheritance of traditional music culture faces challenges. The current diverse needs of music education urgently require innovative teaching models to integrate traditional music culture and enhance students’ aesthetic and cultural literacy. The eco-beauty perspective provides a new educational idea that emphasizes the harmonious coexistence with nature and promotes all-around development of students. Combining traditional music culture and modern education technology effectively can be achieved by combining deep learning technology and new paths in music education. This study explores the music teaching model that integrates traditional music culture with deep learning technology, under the perspective of ecological aesthetic education. The study utilizes a mixed methodology, combines quantitative and qualitative Analysis, and compares the experimental design to verify the effectiveness of the teaching model. In this paper, implementing the music teaching model under the perspective of ecological aesthetic education has improved students’ performance in music perception ability, music emotion expression, music composition and performance, and music knowledge theory. The average increase in music perception ability of the experimental class students is 5.18 points, and the average increase in music composition and performance ability is 11.57 points, showing the impact of this teaching mode on the overall improvement of students’ music ability. In this paper, the music teaching mode combining deep learning under the perspective of ecological aesthetic education can effectively integrate traditional music culture, promote the improvement of students’ music literacy, and provide a new teaching strategy for the inheritance and development of traditional music culture.

This paper analyzes the effectiveness of predictive analysis of brain imaging data based on deep learning algorithms, and improves the prediction accuracy and efficiency of brain imaging data through improved methods. The first step is to measure the local consistency of the brain imaging data using Kendall’s concordance coefficient (KCC), and to analyze the differences between the datasets using the two-sample t-test. Secondly, a batch normalized convolutional neural network (BN-CNN)-based prediction method for brain imaging data has been developed. This method extracts spatial and temporal features in two convolutional layers, followed by a fully connected layer for classification. Experimental results show that this method is helpful in predicting missing structural data in brain imaging. Secondly, a batch normalized convolutional neural network (BN-CNN) based brain imaging data prediction method is developed, which extracts spatial and temporal features in two convolutional layers. Then it connects to a fully connected layer for classification. The experimental results show that this method’s structural similarity index (SSIM) and feature similarity index (FSIM) in brain imaging data prediction of missing data reaches 0.9446 and 0.9465, respectively, which is significantly better than that of other GAN benchmarks. In applying the method to epilepsy and Parkinson’s cases, this algorithm is used to epilepsy and Parkinson’s cases, and a two-sample t-test analyzes the differences in the data sets. In the application of epilepsy and Parkinson’s cases, the algorithm in this paper has an average prediction accuracy of 93.37%, effectively reducing the rate of incorrect predictions. Deep learning algorithms are highly efficient and accurate in predicting brain imaging data, which is crucial for future clinical diagnosis and treatment.

In the context of education informatization, this paper discusses how to deeply integrate advanced educational technology with traditional teaching. By constructing a blended teaching model with the goal of deep learning and using linear regression, decision tree and other algorithms to build a data analysis framework, this paper aims to assess the practical application effect of educational technology in universities. The empirical study explores the development of deep blending between educational technology and teaching by taking the Information Technology Course of University J as an example. The results showed that after the teaching, the posttest mean of profound learning ability of students in the mixed, online, and offline groups were 1.948, 0.182 and 0.065 higher than the pretest mean, respectively. z-values ranked in the top three were the perseverance of students in the mixed group (-12.642) and the ability to think creatively (-11.682) and critically (-9.875), which were higher than those of students in the online group and the offline group, the overall profound learning ability of students in the mixed group was effectively improved, which verified the effectiveness of the blended teaching model proposed in this paper.

Currently, in the production of cigarettes in the blade, charging export moisture control means is relatively single and can not effectively guarantee the excellent quality of cigarette filament. In this paper, first of all, the working principle of the tobacco blade charging machine is introduced, and the moisture of the tobacco leaf for the charging machine is dynamically analyzed, and the influence of the return air temperature control of the charging machine on the export moisture of the blade charging process is explored. Secondly, based on the traditional PID controller, an adaptive fuzzy PID controller is established by combining adaptive fuzzy rules, and then the stacked noise-reducing self-encoder in deep learning is combined with the adaptive fuzzy PID control to design the intelligent control structure of export moisture of leaf charging process. Finally, the effectiveness of export moisture intelligence control, process capability index, and the effect before and after application were analyzed in controlled experiments, respectively. The results show that the difference between the predicted value and the real value of blade export moisture in this paper’s method is only 0.5%, and the process capability index of this paper’s method is improved by 1.48 compared with the PID controller, and it can control the temperature of the return air of the charging machine in the range of 56.86℃~57.21℃. The intelligent control method of export moisture introduced by deep learning can accurately control the export moisture of the leaf dosing process, which effectively ensures the quality of tobacco filament making.

In this paper, we use Caffe framework to implement the improved Faster R-CNN recognition technique for building images in civil engineering under Linux system and add feature pyramid network and regional feature aggregation into the ResNet-50 network and ResNet-101 network, respectively, to strengthen the training effect, and establish ResNet-101+FPN+ROI Align image recognition technique. Simulated crack experiments and concrete surface quality defect detection experiments confirm that the ResNet-101 FPN ROI Align method is accurate and detects defects at a high rate. The method established in this paper has a minimum error of only 0.4% in the simulated crack experiment, and the detection rate is much higher than that of other detection methods when detecting quality defects on the concrete surface, and the accuracy can reach up to 94% at the same time. In civil engineering, the image recognition technology established in this paper has practical significance and high application value, as demonstrated by the experiment.

In this paper, a new intelligent robot path recognition system LADDPG, is constructed by extending the DDPG algorithm with the support of the deep learning network algorithm. The path recognition system is optimized by using the memory system and the decision module guided by the attention mechanism, which solves the problems of the existing DRL intelligent robot path recognition method that cannot store long-time memory and the training time is too long. The system in this paper is validated by conducting path recognition experiments under different road conditions and conducting dynamic obstacle avoidance test experiments. The system can maintain excellent noise immunity and a short path recognition time of 51 ms under the conditions of missing road conditions, circuitous road conditions and large curvature, and in the presence of multiple dynamic obstacles, the system can maintain a collision rate close to 0 while maintaining a very high success rate of 0.98, and the required path recognition time is much lower than that of other methods, with an average reward value of 0.4151. This paper’s system is highly accurate in recognizing intelligent robot paths and has high application value and capability.

In this paper, the dependency relationship between different knowledge points is constructed by statistically counting the correct rate of answering knowledge points at adjacent times, finding the prior knowledge of each knowledge point, and the prior of the prior. The dependencies of each knowledge point are stored as key-value pairs and added into the loss function of the model as constraints, thus completing the construction of the deep learning knowledge tracking model. Aiming at the shortcomings of the deep learning knowledge tracking model, we propose to improve the DT-BKT learning knowledge tracking model and use it to design the teaching system of the university English digital course and analyze the teaching system of the university English digital course by using simulation experiments. The results show that the shortest time for acquiring the digital materials of college English is 12.35s, which is 52.87s shorter than that of the conventional English teaching resources information management system, indicating that the system in this paper is more suitable for the development and innovation of college English teaching in colleges and universities. This study can mobilize students’ willingness to participate in English learning in all aspects and has a promoting effect on the development of university English digital course teaching in colleges and universities.

This paper analyzes the current situation of physical education teaching in colleges and universities from the perspective of deep learning and innovates the teaching methodology. A multidimensional data array is used to quantify students’ interests and hobbies to create a recommendation module for intelligent sports learning materials. Using the SIFT feature extraction algorithm, identify the extreme value point of the detection target and develop the sports target extraction module. Through distance transformation and morphological features, determine the skeleton image features to construct a college sports teaching model using deep learning. The results show that the model in this paper enables the coverage rate of sports courses to reach over 97.20%, with a fluctuation of no more than 1.5%. Deep learning has contributed to some innovation in college sports teaching.

This paper first analyzes the emotional features of English and American literary works by combining semiotics of interactivity, language class emotional context and thinking operation model. Then, a B-Feature-BP text emotion feature construction model is constructed on the basis of the BRET model combined with the BP neural network, and the emotion features of English and American literary works are constructed by combining semiotics and the thought operation model. Then, based on the multi-task learning method in deep learning, a multi-task MT-GSU model is proposed to classify and recognize the emotional features of constructed text. Finally, the performance of constructing, classifying and recognizing the emotional features of English and American literary works in this paper is analyzed so as to analyze the emotional features of English and American literary works. The results show that the performance of the constructed emotional features is all greater than 0.8, and the emotional features of the characters, the environment, and the whole of the English and American literary works are above 0.87, and the classification time is between [0.338,0.721]s. The intensity of tendency of the characteristics of the emotions of the works of English and American literature is [0.68,0.78], the intensity of stability is [0.6,0.74], the intensity of profundity is [0.71,0.79], and the intensity of efficacy is [0.72,0.82]. This study has a positive impact on the appreciation and translation of English and American literary works.

In order to innovate the college music education model, this paper first analyzes the design ideas of music-diversified courses based on deep learning and carries out the feature extraction of music course content. Secondly, combined with the diversified music wisdom environment and curriculum evaluation, a diversified music curriculum system teaching model based on deep learning is designed. Finally, the teaching practice is conducted, and a control experiment is designed to analyze the effectiveness of diversified curriculum innovations. The results showed that after the teaching practice, the basic level of the experimental class increased by 0.26, the cognitive ability increased by 0.23, the learning attitude increased by 0.23, the learning style increased by 0.23, and the interests and hobbies increased by 0.25. The diversified curriculum based on this paper can effectively carry out music teaching innovation.