Search

The enterprises of the make-to-order production mode are required to make accurate judgments and reasonable decisions when dealing with the disturbance factors in the manufacturing process to ensure the timely completion of orders. Order remaining completion time (ORCT) prediction can quantify the production process and is an important basis to ensure the delivery date of orders. However, an accurate prediction model of the ORCT is challenging because of the spatial and temporal (ST) characteristics of the manufacturing process. An ST features-based prediction method is proposed to solve this problem. Firstly, the ST data sets are established on the basis of analysing ST characteristics of the manufacturing process. Secondly, the spatial feature extraction network and temporal feature extraction network are built to make a more comprehensive analysis. Finally, the prediction model of the ORCT is proposed by fusing the spatial and temporal features and validated in a practical workshop. Experimental results show that considering the spatial and temporal characteristics of the manufacturing process can significantly improve the prediction model's performance, and the proposed prediction model is superior to graph convolutional neural network, gated recurrent unit network, deep auto-encoder, deep neural network, and back propagation neural network. [ABSTRACT FROM AUTHOR]

GoogLeNet is a pre-trained Convolutional Neural Network (CNN) that allows transfer learning and has achieved high recognition rates in image classification tasks. A Recurrence Plot (RP) is an imaging method that depicts the recurrence of the state space system using coloured points and lines in 2D images. This work contributes to facilitating time series feature extraction by proposing a method that applies the GoogLeNet to time series images obtained with RP. The developed method is tested using simulated time series and selected time series from the M3 competition dataset. The results shows that the transfer learning approach allowed the extraction of business time series features by means of a GoogLeNet fine-tuned using 100 simulated time series. The combination of GoogLeNet and RPs outperforms the alternative and easier combination of GoogLeNet and plots of the time series and support the convenience of the RP transformation step. This application of deep learning techniques to business time series imaging offers opportunity for further developments. [ABSTRACT FROM AUTHOR]

Depression is a pressing yet underdiagnosed issue in health management. Because depressed patients share their symptoms, life events, and treatments on digital platforms, information systems (IS) scholars resort to user-generated digital traces for depression detection. While they facilitate innovative information technology (IT) approaches to alleviate the social and economic burden of depression, most studies lack effective means to incorporate domain knowledge in depression detection systems or suffer from feature extraction difficulties. Following the design science research in IS, we propose a Deep-Knowledge-Aware Depression Detection system to detect social media users at risk of depression and explain the detection factors. We deploy extensive empirical analyses to evaluate our designed IT artifact, which shows domain knowledge greatly improves performance. Our work has significant implications for IS research in knowledge-aware machine learning, digital traces utilization, and generalizable design principles. Practically, the early detection and factor explanation from our IT artifact can assist depression management and enable large-scale assessment of the population's mental health. [ABSTRACT FROM AUTHOR]

In the semiconductor industry, virtual metrology (VM) is a cost-effective and efficient technique for monitoring the processes from one wafer to another. This technique is implemented by generating a predictive model that uses real-time data from equipment sensors in conjunction with measured wafer quality characteristics. Before establishing a prediction model for the VM system, appropriate selection of relevant input variables should be performed to maintain the efficiency of subsequent analyses considering the large dimensionality of the sensor data inputs. However, wafer production processes usually employ multiple sensors, which leads to cost escalations. Herein, we propose a variant of the sparse principal component analysis (PCA) called true sparse PCA (TSPCA). The proposed method uses a small number of input variables in the first few principal components. The main contribution of the proposed TSPCA is reducing the number of essential sensors. Our experimental results demonstrate that compared to the existing sparse PCA methods, the proposed approach can reduce the number of sensors required while explaining an approximately equivalent amount of variance. [ABSTRACT FROM AUTHOR]

Summary: Anomaly event recognition and identification has a crucial part in several areas, particularly in night vision environments. Conventional techniques of event recognition are hugely based upon data extracted from certain images for classification purposes. This needs users to select suitable features to establish the feature depictions for actual images per definite situations. Manual feature selection is laborious as well as heuristic tasks and the features obtained in this manner generally have worse robustness. Here, a Faster Region‐based Convolutional fused Social Generative Adversarial Network (FRC‐SGAN) is designed for anomaly event recognition in a night vision environment. At the cloud, key frame extraction, pre‐processing, feature extraction, human detection (HD) and anomalous event recognition are carried out. Initially, input video from the database is subjected to perform pre‐processing. The visibility enhancement is utilized for pre‐processing. Thereafter, features like ResNet features, texture features and statistical features are extracted. Then, HD is accomplished by DeepJoint segmentation with chord distance. Finally, anomalous detection is done by FRC‐SGAN that is the incorporation of Fast Regional Convolutional Neural Network (FR‐CNN) and Social Generative Adversarial Network (SGAN). In addition, FRC‐SGAN acquired 90.8% of accuracy, 89.7% of precision, and 89.2% of recall. [ABSTRACT FROM AUTHOR]

To maintain a qualified product, it is necessary to control the final machined quality approximately. To this end, massive research has been devoted to modeling and controlling the machining-induced surface roughness. However, a generalized surface roughness prediction model is hard to develop due to the complex modeling process and insufficient data. And a feasible surface roughness stabilization method is often missing in the existing studies. To this end, this paper proposed a novel self-adaptive machining parameters adjustment method for stabilizing the machining-induced surface roughness. In the proposed method, a physical surface roughness prediction model is developed at first. Then, a CNN-LSTM is employed to realize spatial–temporal feature extraction. Next, the MMD-MSE-based method is employed to realize the transfer learning process. Finally, a self-adaptive process parameter tuning system using the gradient descent method is developed, based on the surface prediction method. Experiments are conducted on a milling machine, and results indicate that the proposed method can realize a high accuracy and generalization prediction of surface roughness. In terms of the machined surface roughness, the proposed method effectively maintains the surface roughness under 1.6 μm. [ABSTRACT FROM AUTHOR]

Silent speech recognition (SSR) based on surface electromyography (sEMG) is a voice interaction technology proposed for scenarios requiring silent operations. This article abstracts the SSR task based on sEMG into a short‐term image sequence classification task. Time‐frequency domain feature extraction and data reconstruction on the muscle activity segment data is performed. Additionally, the temporal and spatial dimensions to capture the intrinsic correlation representation of muscle activity is analysed. The SVIT‐SSR model is proposed based on the vision transformer (VIT) framework. Finally, experiments to identify 33 types of typical silent speech commands in the SSR dataset are designed. The results demonstrate that the proposed model achieves an accuracy of 96.67 ± 1.15%, outperforming similar algorithms. [ABSTRACT FROM AUTHOR]

As the court put forward higher requirements for quantitative evaluation and scientific standards of forensic evidence, how to objectively and scientifically express identification opinions has become a challenge for traditional forensic identification methods. Score‐based likelihood ratios are mathematical methods for quantitative evaluation of forensic evidence. However, due to the subtle differences in inter‐class barefootprints, there is no automatic barefootprints matching algorithm with high accuracy under large‐scale dataset validation, and there are few studies related to deep learning barefootprint features for evidence evaluation in court. Therefore, score‐based likelihood ratios for barefootprint evidence using deep learning features are proposed by this paper. Firstly, the largest barefootprint dataset (BFD) is constructed, which contains 54,118 barefootprint images from 3000 individuals. Then, an automatic barefootprint feature extraction and matching algorithm is proposed, which achieves a retrieval accuracy of 98.4% on BFD and an AUC of 0.989 for barefootprint validation. Next, Cosine distance, Euclidean distance and Manhattan distance are employed to measure the comparison scores between intra‐class and inter‐class barefootprints using deep learning features in four dimensions of 64, 128, 512 and 1024, respectively. The performance of proposed model is evaluated by comparing the Cllr$$ {C}_{llr} $$ values and the Tippett plot. Finally, simulated crime scene barefootprint samples are constructed to verify the practical application of the proposed method, which provide further support for the quantitative evaluation of barefootprint evidence in court. [ABSTRACT FROM AUTHOR]

Leaf nitrogen content (LNC) is an essential indicator of crop nitrogen status. To rapidly and correctly estimate the LNC using hyperspectral remote sensing, the canopy hyperspectral reflectance of coffee trees treated with five levels of nitrogen fertilization in a greenhouse was obtained in this study. Five methods were used for hyperspectral data preprocessing, namely, Savitzky–Golay (SG) smoothing, a combination of SG and standard normal variate transformation (SG-SNV), a combination of SG and first-order derivative (SG-FD), a combination of SG and second-order derivative (SG-SD), and a combination of SG and multiplicative scatter correction (SG-MSC). Feature wavelengths were extracted using variables combination population analysis (VCPA), competitive adaptive reweighted sampling (CARS), and the combination (CARS-SPA) of CARS and successive projections algorithm (SPA). Vegetation indexes (VIs) were constructed and subjected to correlation analysis and variance inflation factor (VIF) analysis. Linear and nonlinear models including partial least squares regression (PLSR), back propagation neural network (BPNN), extreme learning machine (ELM), random forest regression (RFR), and support vector regression (SVR), were adopted to construct LNC retrieval models for coffee trees. The results indicated that SG-MSC could increase the signal-to-noise ratio of hyperspectral data well. The wavelengths selected by CARS-SPA were more relevant to LNC, and combined with ELM resulted in the best performance of LNC prediction (R2P = 0.901, RMSEP = 0.825 g·kg−1, RPD = 3.229). Ten VIs were obtained through correlation analysis and VIF, and the VIs-based ELM prediction model also performed moderately well (R2P = 0.814, RMSEP = 1.131 g·kg−1, RPD = 2.354). By comparing the coffee LNC prediction models established by different methods, two coffee LNC inversion models with better prediction accuracy were obtained, which provide a scientific basis for accurate diagnosis of coffee trees LNC, and are of great significance for optimizing the field management. [ABSTRACT FROM AUTHOR]

Blood Pressure (BP) is considered an essential factor that provides information regarding cardiovascular function. Regular monitoring of the BP is required for proper healthcare maintenance that avoids the high risk of life due to high and low BP. Several methods were devised for the estimation of BP, but the estimation accuracy is still a challenging task. Hence this research introduces an efficient BP estimation technique using the Fact Finding Instructor (FFI) based clustering method by considering the speech signal of the patients. An efficient BP extraction technique is introduced using the FFI Optimization algorithm an integration of the mannerism of the fact finder that identifies the suspect who commits the criminal offense and, with the instructor with good knowledge, these make the trainee more efficient. The detection and suspect's arrest contain two phases, the fact-finding phase and the chasing phase. Initially, the speech signal is collected from the database and pre-processed for removing noise and artifacts. Then feature extraction is used for the minimization of the computation overhead that generates a feature vector. The clustering of BP is employed with the k-means clustering algorithm and the proposed FFI optimization algorithm. The FFI Optimization algorithm provides a fast convergence rate due to the fact-finding phase and provides accurate detection of the suspect's location along with that the clustering of classes of patients' BP by considering the feature of the speech signal. The clusters formed using the FFI optimization algorithm are combined with the K-means clustering, by multiplying the clusters the BP estimation is implemented on three criteria Low BP, Normal, and, High BP. Finally, the output generated by both the clustering operations is multiplied together for the estimation of the BP. The performance of the proposed method is evaluated using the metrics like Davies Bouldin score, Homogeneity score, Completeness score, Jacquard Similarity score, Silhouette score, and Dunn's Index which acquired the improvement rate of 0.98, 0.96, 0.96, 0.98, 0.95, and 0.98 for training percentage 90, respectively to the existing Teaching Learning Based Optimization(TLBO) clustering technique. [ABSTRACT FROM AUTHOR]

Injection molding is a widely utilized manufacturing process across various industries. The cooling time in injection molding is an important factor that affects the productivity and energy consumption of the process. The production efficiency is directly proportional to the cooling efficiency, yet optimizing this cooling process presents significant challenges. The fixed cooling parameters are not suitable for all types of material, thus increasing the molding deviation. To address these challenges, this research work proposed a Regularized with Mish Linear Unit-based Deep Learning Neural Network-based material prediction with optimization of derived variables using the Banana Function-centric Green Anaconda Optimization approach. Initially, the features are extracted from the raw materials, and then the material type is predicted by using Regularized with Mish Linear Unit-based Deep Learning Neural Network. Conversely, various machine properties are clustered utilizing the binomial distribution with damping factor-based farthest first clustering methodology. Subsequently, parameters are derived from both the predicted material and the clustered machine property. The pressure of the machine is controlled by a quasi-conservative with an optimized proportional integral derivative controller. Then, the optimal parameters are selected from the Banana Function-centric Green Anaconda Optimization algorithm. In the optimal parameter selection, the multi-objective is considered by minimization of temperature, cooling time, pressure drop, and power consumption. In experimental analysis, the performance of the proposed approach is analyzed with the existing approaches. The proposed approach attains 98.9% accuracy, which is higher than existing approaches. [ABSTRACT FROM AUTHOR]

Background: Sleep spindles have emerged as valuable biomarkers for assessing cognitive abilities and related disorders, underscoring the importance of their detection in clinical research. However, template matching-based algorithms using fixed templates may not be able to fully adapt to spindles of different durations. Moreover, inspired by the multiscale feature extraction of images, the use of multiscale feature extraction methods can be used to better adapt to spindles of different frequencies and durations. Methods: Therefore, this study proposes a novel automatic spindle detection algorithm based on elastic time windows and spatial pyramid pooling (SPP) for extracting multiscale features. The algorithm utilizes elastic time windows to segment electroencephalogram (EEG) signals, enabling the extraction of features across multiple scales. This approach accommodates significant variations in spindle duration and polarization positioning during different EEG epochs. Additionally, spatial pyramid pooling is integrated into a depthwise separable convolutional (DSC) network to perform multiscale pooling on the segmented spindle signal features at different scales. Results: Compared with existing template matching algorithms, this algorithm's spindle wave polarization positioning is more consistent with the real situation. Experimental results conducted on the public dataset DREAMS show that the average accuracy of this algorithm reaches 95.75%, with an average negative predictive value (NPV) of 96.55%, indicating its advanced performance. Conclusions: The effectiveness of each module was verified through thorough ablation experiments. More importantly, the algorithm shows strong robustness when faced with changes in different experimental subjects. This feature makes the algorithm more accurate at identifying sleep spindles and is expected to help experts automatically detect spindles in sleep EEG signals, reduce the workload and time of manual detection, and improve efficiency. [ABSTRACT FROM AUTHOR]

Face recognition is used in many applications such as access control, automobile security, criminal identification, immigration, healthcare, cyber security, and so on. Each person has his/her own unique face, so the face can help distinguish people from each other. Feature extraction process plays a fundamental role in accuracy of face recognition, and many algorithms have been presented to extract more informative features from the face image. In this paper, an efficient texture descriptor is proposed based on local information of the face image. In the proposed method, at first, face image is split into several sub-images in such a way that each sub-image includes one of the facial parts such as eyes, nose, and lips. Second, texture features are extracted from each sub-image using a new local pattern descriptor, and then features of sub-images are concatenated to construct feature vector. Finally, the face image is compared to images in a dataset based on a similarity measure. In addition, particle swarm optimization algorithm is used to assign weight to the features of different parts of the face image. To evaluate the proposed algorithm, four face datasets, Yale, ORL, GT and KDEF, are used. Implementation results show that the proposed method outperforms recent methods in terms of accuracy, receiver operating characteristic (ROC) curve, and area under ROC curve. [ABSTRACT FROM AUTHOR]

Accurately predicting the capacity and remaining useful life (RUL) of lithium-ion batteries is crucial for their reliable and safe functioning. In this research, we propose a dual Siamese transformer-encoder-based network consisting of two subnetworks to improve the RUL prediction capability and for wider applications. The first subnetwork, autoTrans, adopts a transformer-encoder architecture to form an autoencoder structure for the feature extraction. The second subnetwork, regTrans, is a transformer-encoder-based regressor model which takes the featuring encodings from autoTrans as inputs and makes the RUL prediction. To enhance the model's robustness, both subnetworks employ the Siamese architecture for handling raw and noisy inputs. A joint training strategy is applied on autoTrans and regTrans to optimize the proposed approach. The experimental verification is conducted on the NASA battery, and our model achieves the best average results across different evaluation criteria. Furthermore, we also apply our model to the turbofan engine dataset and demonstrate promising performance as well. [ABSTRACT FROM AUTHOR]

Monocular depth estimation is essential in computer vision and robotics applications, including localization, mapping, and 3D object detection. In recent years, supervised learning algorithms that model large amounts of data have been successful in depth estimation. However, obtaining dense ground truth depth labels remains a challenge in supervised training. Therefore, unsupervised methods trained using monocular image sequences have gained wider attention. However, the depth estimation results of most existing models often produce blurred edges. Therefore, we propose various effective improvement strategies to construct a depth estimation network TL-Depth. (1) We propose a tower connection structure that utilizes convolutional processing to facilitate feature fusion, achieve precise semantic classification of pixels, and yield more accurate depth results. (2) We employ a Laplacian-filtering residual to focus on boundary information and enhance detailed results. (3) During the feature extraction stage, multiple pooling excitations are used by embedding them in the convolutional layer. This reduces redundant information while enhancing the network's feature extraction capability. The experimental results on the KITTI dataset and the Make3D dataset demonstrate that this method achieves good results compared to current methods. [ABSTRACT FROM AUTHOR]

Due to the powerful ability in capturing the global information, transformer has become an alternative architecture of CNNs for hyperspectral image classification. However, general transformer mainly considers the global spectral information while ignores the multiscale spatial information of the hyperspectral image. In this paper, we propose a multiscale spectral–spatial convolutional transformer (MultiFormer) for hyperspectral image classification. First, the developed method utilizes multiscale spatial patches as tokens to formulate the spatial transformer and generates multiscale spatial representation of each band in each pixel. Second, the spatial representation of all the bands in a given pixel are utilized as tokens to formulate the spectral transformer and generate the multiscale spectral–spatial representation of each pixel. Besides, a modified spectral–spatial CAF module is constructed in the MultiFormer to fuse cross‐layer spectral and spatial information. Therefore, the proposed MultiFormer can capture the multiscale spectral–spatial information and provide better performance than most of other architectures for hyperspectral image classification. Experiments are conducted over commonly used real‐world datasets and the comparison results show the superiority of the proposed method. [ABSTRACT FROM AUTHOR]

To effectively address common issues such as cattle being obscured by fences and images prone to colour shifts and high brightness in a ranch setting, this paper proposes an algorithm for counting cows based on multi‐scale perception and image correlation. The algorithm first adjusts the model output scale to enhance cattle detection under current conditions. It incor‐porates efficient Partial Convolution (PConv) to replace 3 × 3 convolutions in the Neck segment of the YOLOv7 network, boost‐ing computational speed and reducing complexity. To streamline feature fusion, Dynamic Head (DyHead) unifies multiple at‐tentional operations in the Neck segment, enhancing efficiency. Additionally, it introduces a novel bounding box similarity metric Minimum Point DioU (MPDIoU) based on minimum point distance, encompassing factors from existing loss functions, while simplifying computations. Experimental results demonstrate the algorithm significantly improves detection, achieving 98.8% accuracy, 99.0% recall, and a 92.1% mAP value. Compared with mainstream SOTA models, Precision increases by 0.4%, Recall by 2.0%, and mAP value by 2.2%. Model size decreases by 23.9%, parameter count by 23.0%, and computational load by 6.1%. the algorithm shows improvements across all indices, meeting the challenge of real‐time cattle counting in ranches under complex conditions. [ABSTRACT FROM AUTHOR]

Epilepsy is referred to as a neurological disorder, which is detected via examination and manual comprehension of Electroencephalogram (EEG) signals. In deep learning schemes, various enhancements have emerged to efficiently address complex issues by end-to-end learning. The major objective of this research is to propose a new seizure detection approach from EEG signals using a deep learning-based classification technique. The pre-processing is the initial stage, where denoising is performed using a Short-Time Fourier Transform (STFT). Subsequently, the statistical features, time-domain features and spectral features are extracted from the pre-processed signal. Finally, an efficient optimization approach, named Adadelta-Chameleon Swarm Algorithm (Adadelta-CSA), is proposed and employed to train Deep Neural Network (DNN) to carry out the precise seizure prediction. Here, the integration of the Adadelta concept in the Chameleon Swarm Algorithm (CSA) has resulted in Adadelta-CSA. At last, the performance of the Adadelta-CSA scheme-based DNN is compared with the existing techniques by considering accuracy, sensitivity and specificity, and it is found to produce better values of 0.951, 0.966, and 0.935, respectively. [ABSTRACT FROM AUTHOR]

Objectives Methods Results Conclusions This study aimed to develop a quantitative intratumoral heterogeneity (ITH) model for assessing the risk of early recurrence (ER) in pre‐treatment multimodal imaging for hepatocellular carcinoma (HCC) patients undergoing ablation treatments.This multi‐centre study enrolled 633 HCC patients who underwent ultrasound‐guided local ablation between January 2015 and September 2022. Among them, 422, 85, 57 and 69 patients underwent radiofrequency ablation (RFA), microwave ablation (MWA), laser ablation (LA) and irreversible electroporation (IRE) ablation, respectively. Vision‐Transformer‐based quantitative ITH (ViT‐Q‐ITH) features were extracted from the US and MRI sequences. Multivariable logistic regression analysis was used to identify variables associated with ER. A combined model integrated clinic‐radiologic and ViT‐Q‐ITH scores. The prediction performance was evaluated concerning calibration, clinical usefulness and discrimination.The final training cohort and internal validation cohort included 318 patients and 83 patients, respectively, who underwent RFA and MWA. The three external testing cohorts comprised of 106 patients treated with RFA, 57 patients treated with LA and 69 patients who underwent IRE ablation. The combined model showed excellent predictive performance for ER in the training (AUC: .99, 95% CI: .99–1.00), internal validation (AUC: .86, 95% CI: .78–.94), external testing (AUC: .83, 95% CI: .73–.92), LA (AUC: .84, 95% CI: .73–.95) and IRE (AUC: .82, 95% CI: .72–.93) cohorts, respectively. Decision curve analysis further affirmed the clinical utility of the combined model.The multimodal‐based model, incorporating clinic‐radiologic factors and ITH features, demonstrated superior performance in predicting ER among early‐stage HCC patients undergoing different ablation modalities. [ABSTRACT FROM AUTHOR]

• We propose a model to detect the spatial locus of a listener's auditory attention. • EEG is modeled as a graph with node features based on time-frequency feature fusion. • Graph convolution can take advantage of local brain neuroanatomical connectivity. • Global attention mechanism is adopted to learn global brain effective connectivity. • Right-lateralized frontotemporal activation and connectivity exist in spatial hearing. Auditory spatial attention detection (ASAD) aims to decipher the spatial locus of a listener's selective auditory attention from electroencephalogram (EEG) signals. However, current models may exhibit deficiencies in EEG feature extraction, leading to overfitting on small datasets or a decline in EEG discriminability. Furthermore, they often neglect topological relationships between EEG channels and, consequently, brain connectivities. Although graph-based EEG modeling has been employed in ASAD, effectively incorporating both local and global connectivities remains a great challenge. To address these limitations, we propose a new ASAD model. First, time-frequency feature fusion provides a more precise and discriminative EEG representation. Second, EEG segments are treated as graphs, and the graph convolution and global attention mechanism are leveraged to capture local and global brain connections, respectively. A series of experiments are conducted in a leave-trials-out cross-validation manner. On the MAD-EEG and KUL datasets, the accuracies of the proposed model are more than 9% and 3% higher than those of the corresponding state-of-the-art models, respectively, while the accuracy of the proposed model on the SNHL dataset is roughly comparable to that of the state-of-the-art model. EEG time-frequency feature fusion proves to be indispensable in the proposed model. EEG electrodes over the frontal cortex are most important for ASAD tasks, followed by those over the temporal lobe. Additionally, the proposed model performs well even on small datasets. This study contributes to a deeper understanding of the neural encoding related to human hearing and attention, with potential applications in neuro-steered hearing devices. [ABSTRACT FROM AUTHOR]

Parkinson's disease (PD) is a prevalent neurodegenerative disorder affecting millions of people globally, with substantial health risks and economic burdens. This study aims to introduce an innovative hybrid approach combining deep learning and machine learning algorithms to improve the diagnosis of PD using handwriting dynamics indicative of Parkinson's symptoms. The proposed approach integrates hybrid feature extraction using nine fine-tuned transfer learning models, i.e., InceptionV3, DenseNet201, EfficientNetB0, ResNet50, MobileNetV2, VGG16, Xception, NASNetMobile, and InceptionResNetV2. Initially, features from these models are used individually or in binary and ternary combinations. Given the limited sample size in PD datasets, some extracted features through fine-tuning may lack significance, and fully connected layers can lead to overfitting. To address this issue, Neighborhood Component Analysis (NCA) is employed to refine these features, retaining only the most informative ones. Finally, the selected features are classified using Support Vector Machines (SVM) maximizing the margin between classes and reducing the risk of overfitting. The proposed hybrid model achieves a state-of-the-art accuracy of 99.39% on the Parkinson Hand Drawings dataset. The combination of features extracted from DenseNet201, Xception, and NASNetMobile models, processed using NCA and SVM methods, has been identified as the most efficient model, balancing high accuracy with computational efficiency. Qualitative assessments further confirm the accuracy and reliability of the approach. [ABSTRACT FROM AUTHOR]

A proper system for identifying leaf disease, which is crucial for developing agricultural areas, has been addressed in this research using a neural network approach. This study helps find plant illnesses and their stages whenever they occur. Fungal, bacterial, and viral infections are very harmful to plants. Five major tomato diseases have been classified in this research: bacterial spot, early blight, late blight, leaf mold, tomato mosaic virus, and healthy tomato plant leaves. The study underscores the importance of algorithmic adaptability to attain precision in leaf disease identification and emphasizes the potential for customized strategies in achieving accuracy. The classification is done by extracting color, shape, and texture features from a healthy tomato plant leaf image. The feature extraction method is carried out to proceed with the segmentation phase. Features extracted from segmented pictures are used as inputs to a classification algorithm. These five categories were used to finalize the illness categorization stage. The variety of five kinds of tomato leaf images yielded almost 99% classification accuracy. Furthermore, various research gaps have been identified to achieve a more open approach to detecting tomato diseases. [ABSTRACT FROM AUTHOR]

In recent years, convolutional neural networks have significantly advanced the field of computer vision by automatically extracting features from image data. CNNs enable the modeling of complex and abstract image features using learnable filters, eliminating the need for manual feature extraction. However, combining feature maps obtained from CNNs with different approaches can lead to more complex and interpretable inferences, thereby enhancing model performance and generalizability. In this study, we propose a new method called RGB angle rotation to effectively obtain feature maps from RGB images. Our method rotates color channels at different angles and uses the angle information between channels to generate new feature maps. We then investigate the effects of integrating models trained with these feature maps into an ensemble architecture. Experimental results on the CIFAR-10 dataset show that using the proposed method in the ensemble model results in performance increases of 9.10 and 8.42% for the B and R channels, respectively, compared to the original model, while the effect of the G channel is very limited. For the CIFAR-100 dataset, the proposed method resulted in a 17.09% improvement in ensemble model performance for the R channel, a 5.06% increase for the B channel, and no significant improvement for the G channel compared to the original model. Additionally, we compared our method with traditional feature extraction methods like scale-invariant feature transform and local binary pattern and observed higher performance. In conclusion, it has been observed that the proposed RGB angle rotation method significantly impacts model performance. [ABSTRACT FROM AUTHOR]

The development of robust and agile locomotion skills for legged robots using reinforcement learning is challenging, particularly in demanding environments. In this study, we propose a blind locomotion control learning framework that enables fast and stable walking on challenging terrains. First, we construct an asymmetric terrain feature extraction network that uses a multilayer perceptron to effectively infer terrain features from the history of proprioceptive states, consisting only of inertial measurement unit and joint encoder data. Additionally, our asymmetric actor-critic framework implicitly infers terrain features, thereby enhancing the accuracy of terrain representation. Second, we introduce a foot trajectory generator based on prior gait behaviors, which improves the gait periodicity and provides accurate state information for terrain feature inference. Compared to state-of-the-art methods, our approach significantly increases the learning efficiency by 26.0% and enhances terrain adaptation by 5.0%. It also achieved a more periodic gait, with the state-command tracking error reduced by 38.5% compared with advanced methods. The success rate for traversing complex terrains was similar to that of the baseline methods, with a 31.3% increase in the step height on stair-like terrains. The experimental results demonstrate that the proposed method enables fast and stable walking on challenging terrains. [ABSTRACT FROM AUTHOR]

The lubrication system supplies lubrication and cleans the rotating parts and contacting machinery during the operation of an aero-engine. It is crucial to maintain an adequate amount of lubricant by predicting and analyzing the consumption rate to ensure endurance and maintenance programs are effective. This paper examines the combination of temporal and non-temporal data that impact the characteristic parameters of lubricant consumption rate in aero-engines. Our study focuses on the merging of LSTM (Long Short-Term Memory) + LightGBM (Light Gradient Boosting Machine) + CatBoost, and uses KPCA dimensionality reduction optimization, along with Stacking for the fusion of a multi-feature regression prediction algorithm. On the one hand, this study utilizes integrated learning to fuse feature extractions from LSTM for temporal information and non-temporal information by GDBT (Gradient Boosting Decision Tree). This approach considers the trend and distribution of feature samples to develop a more robust feature extraction method. On the other hand, the integrated learning framework incorporates multi-decision making and feature importance extraction to strengthen the mapping relationship with the predicted output of lubrication oil consumption rate, enabling regression prediction. The algorithm for regression prediction has been executed and the results indicate a final regression prediction MAPE (Mean Absolute Percentage Error) of less than 3%. MSE and RMSE reached 1.28% and 1.33%, the results are in an ideal state. The algorithms used in this paper will be applied in the future to aero-engine lubricant systems and eventually to engines in general. [ABSTRACT FROM AUTHOR]

In the multi-focus image fusion process, generating a focus decision map is typically a requisite step before the actual fusion of images. Many multi-focus image fusion methods often require optimization through post-processing techniques such as hole filling and removal of small regions in the decision map. To solve these problems and obtain better fusion images, we propose a multi-focus image fusion algorithm based on an end-to-end decision optimization model called EDOM-MFIF. Firstly, we design a cross-scale feature extraction encoder based on the joint attention fusion of upper and lower information, which can fully extract the shallow texture information and high-level semantic information of images. Secondly, we design a feature fusion module using convolution and spatial attention for comprehensive feature fusion and maximal retention of global image information. Finally, we achieve end-to-end training and optimization of the focusing decision map by integrating the decision map's optimization into the whole image fusion architecture. This prevents evident pixel faults in the immediate area from spreading to other areas during post-processing, which could lead to inaccurate classification findings, in addition to maintaining pixel consistency with the original image to the greatest extent possible. To facilitate training of the multi-focus image fusion model, we construct a large-scale multi-focus image dataset with a supervised decision map and test the algorithm on three different types of public datasets. The experimental results show that the proposed algorithm outperforms other advanced multi-focus image fusion algorithms in both objective evaluation and visual evaluation. [ABSTRACT FROM AUTHOR]