Search

The recognition of food images is of great significance for nutrition monitoring, food retrieval and food recommendation. However, the accuracy of recognition had not been high enough due to the complex background of food images and the characteristics of small inter-class differences and large intra-class differences. To solve these problems, this paper proposed a food image recognition method based on transfer learning and ensemble learning. Firstly, generic image features were extracted by using the convolutional neural network models (VGG19, ResNet50, MobileNet V2, AlexNet) pre-trained on the ImageNet dataset. Secondly, the 4 pre-trained models were transferred to the food image dataset for model fine-tuning. Finally, different basic learner combination strategies were adopted to establish the ensemble model and classify feature information. In this paper, several kinds of experiments were performed to compare the results of food image recognition between single models and ensemble models on food-11 dataset. The experimental results demonstrated that the accuracy of the ensemble model was the highest, reaching 96.88%, which was superior to any base learner. Therefore, the convolutional neural network model based on transfer learning and ensemble learning has strong learning ability and generalization ability, and it is feasible and practical to apply the method to food image recognition. [ABSTRACT FROM AUTHOR]

Brain extraction is an important prerequisite for the automated diagnosis of intracranial lesions and determines, to a certain extent, the accuracy of subsequent lesion identification, localization, and segmentation. To address the problem that the current traditional image segmentation methods are fast in extraction but poor in robustness, while the Full Convolutional Neural Network (FCN) is robust and accurate but relatively slow in extraction, this paper proposes an adaptive mask-based brain extraction method, namely AMBBEM, to achieve brain extraction better. The method first uses threshold segmentation, median filtering, and closed operations for segmentation, generates a mask for the first time, then combines the ResNet50 model, region growing algorithm, and image properties analysis to further segment the mask, and finally complete brain extraction by multiplying the original image and the mask. The algorithm was tested on 22 test sets containing different lesions, and the results showed MPA = 0.9963, MIoU = 0.9924, and MBF = 0.9914, which were equivalent to the extraction effect of the Deeplabv3+ model. However, the method can complete brain extraction of approximately 6.16 head CT images in 1 second, much faster than Deeplabv3+, U-net, and SegNet models. In summary, this method can achieve accurate brain extraction from head CT images more quickly, creating good conditions for subsequent brain volume measurement and feature extraction of intracranial lesions. [ABSTRACT FROM AUTHOR]

To accurately locate faulty components in analog circuits, an analog circuit fault diagnosis method based on Tunable Q-factor Wavelet Transform(TQWT) and Convolutional Neural Network (CNN) is proposed in this paper. Firstly, the Grey Wolf algorithm (GWO) is used to improve the TQWT. The improved TQWT can adaptively determine the parameters Q-factor and decomposition level. Secondly, The signal is decomposed, and single-branch reconstruction is conducted with TQWT to facilitate adequate feature extraction. Thirdly, to capture the time-frequency features in the signal, a CNN-LSTM network is built by combining CNN and LSTM for feature extraction. Finally, CNN, which introduces Fully Convolutional Network (FCN) layers and a Batch Normalization layer, is used to fault diagnosis. The method was comprehensively evaluated with a second-order bandpass filter circuit. The experimental results illustrate that the proposed fault diagnosis method can achieve excellent fault diagnosis accuracy, and the average accuracy is 98.96%. [ABSTRACT FROM AUTHOR]

In the field of surface electromyography (sEMG) gesture recognition, how to improve recognition accuracy has been a research hotspot. The rapid development of deep learning provides a new solution to this problem. At present, the main applications of deep learning for sEMG gesture feature extraction are based on convolutional neural network (CNN) structures to capture spatial morphological information of the multichannel sEMG or based on long short-term memory network (LSTM) to extract time-dependent information of the single-channel sEMG. However, there are few methods to comprehensively consider the distribution area of the sEMG signal acquisition electrode sensor and the arrangement of the sEMG signal morphological features and electrode spatial features. In this paper, a novel multi-stream feature fusion network (MSFF-Net) model is proposed for sEMG gesture recognition. The model adopts a divide-and-conquer strategy to learn the relationship between different muscle regions and specific gestures. Firstly, a multi-stream convolutional neural network (Multi-stream CNN) and a convolutional block attention module integrated with a resblock (ResCBAM) are used to extract multi-dimensional spatial features from signal morphology, electrode space, and feature map space. Then the learned multi-view depth features are fused by a view aggregation network consisting of an early fusion network and a late fusion network. The results of all subjects and gesture movement validation experiments in the sEMG signal acquired from 12 sensors provided by NinaPro's DB2 and DB4 sub-databases show that the proposed model in this paper has better performance in terms of gesture recognition accuracy compared with the existing models. [ABSTRACT FROM AUTHOR]

Extracting rich feature representations is a key challenge in person re-identification (Re-ID) tasks. However, traditional Convolutional Neural Networks (CNN) based methods could ignore a part of information when processing local regions of person images, which leads to incomplete feature extraction. To this end, this paper proposes a person Re-ID method based on vision Transformer with hierarchical structure and window shifting. When extracting person image features, the hierarchical Transformer model is constructed by introducing the hierarchical construction method commonly used in CNN. Then, considering the importance of local information of person images for complete feature extraction, the self-attention calculation is performed by shifting within the window region. Finally, experiments on three standard datasets demonstrate the effectiveness and superiority of the proposed method. [ABSTRACT FROM AUTHOR]

Siamese-based trackers have achieved excellent performance and attracted extensive attention, which regard the tracking task as a similarity learning between the target template and search regions. However, most Siamese-based trackers do not effectively exploit correlations of the spatial and channel-wise information to represent targets. Meanwhile, the cross-correlation is a linear matching method and neglects the structured and part-level information. In this paper, we propose a novel tracking algorithm for feature extraction of target templates and search region images. Based on convolutional neural networks and shuffle attention, the tracking algorithm computes the similarity between the template and a search region through a graph attention matching. The proposed tracking algorithm exploits the correlations between the spatial and channel-wise information to highlight the target region. Moreover, the graph matching can greatly alleviate the influences of appearance variations such as partial occlusions. Extensive experiments demonstrate that the proposed tracking algorithm achieves excellent tracking results on multiple challenging benchmarks. Compared with other state-of-the-art methods, the proposed tracking algorithm achieves excellent tracking performance. [ABSTRACT FROM AUTHOR]

The design of deep convolutional neural networks has resulted in significant advances and successes in the field of object detection. However, despite these achievements, the high computational and memory costs of such object detection networks on the edge or in mobile scenarios are one of the most significant barriers to their broad adoption. To solve this problem, this paper introduces an improved lightweight real-time convolutional neural network based on YOLOv5, called Alpha-SGANet: A multi-attention-scale feature pyramid network combined with a lightweight network based on Alpha-IoU loss. Firstly, we add one more prediction head to detect different-scale objects, design a lightweight and efficient feature extraction network using ShuffleNetV2 in the backbone, and reduce information loss using the SPP module with a smaller convolutional nucleus. Then, cleverly, employ GAFPN to improve feature transition processing in the neck region, including the usage of the Ghost module to construct efficient feature maps to help prediction. The CBAM module was further integrated to find areas of interest in the scene; finally, combined with Alpha-IOU loss for model supervision training, the biggest performance improvement was achieved. The experiment results show that, compared with YOLOv5s, our proposed method can achieve higher accuracy with fewer parameters and has real-time speed through verification on the PASCAL VOC dataset and MS COCO dataset. [ABSTRACT FROM AUTHOR]

Various physical information can be leaked while the encryption algorithm is running in the device. Side-channel analysis exploits these leakages to recover keys. Due to the sensitivity of deep learning to the data features, the efficiency and accuracy of side channel analysis are effectively improved with the application of deep learning algorithms. However, a considerable part of existing reserches are based on traditional neural networks. The effectiveness of key recovery is improved by increasing the size of the network. However, the computational complexity of the algorithm increases accordingly. Problems such as overfitting, low training efficiency, and low feature extraction ability also occur. In this paper, we construct an improved lightweight convolutional neural network based on the feature fusion network. The new network and the traditional neural networks are respectively applied to the side-channel analysis for comparative experiments. The results show that the new network has faster convergence, better robustness and higher accuracy. No overfitting has occurred. A heatmap visualization method was introduced for analysis. The new network has higher heat value and more concentration in the key interval. Side-channel analysis based on feature fusion network has better performance, compared with the ones based on traditional neural networks. [ABSTRACT FROM AUTHOR]

Visual tracking is a fundamental research task in vision computer. It has broad application prospects, such as military defense and civil security. Visual tracking encounters many challenges in practical application, such as occlusion, fast motion and background clutter. Siamese based trackers achieve superior tracking performance in balanced accuracy and tracking speed. The deep feature extraction with Convolutional Neural Network (CNN) is an essential component in Siamese tracking framework. Although existing trackers take full advantage of deep feature information, the spatial structure and semantic information are not adequately exploited, which are helpful for enhancing target representations. The lack of these spatial and semantic information may lead to tracking drift. In this paper, we design a CNN feature extraction subnetwork based on a Depthwise Over-parameterized Convolutional layer (DO-Conv). A joint convolution method is introduced, namely the conventional and depthwise convolution. The depthwise convolution kernel explores independent channel information, which effectively extracts shallow spatial information and deep semantic information, and discards background information. Based on DO-Conv, we propose a novel tracking algorithm in Siamese framework (named DOSiam). Extensive experiments conducted on five benchmarks including OTB2015, VOT2016, VOT2018, GOT-10k and VOT2019-RGBT(TIR) show that the proposed DOSiam achieves leading tracking performance with real-time tracking speed at 60 FPS against state-of-the-art trackers. [ABSTRACT FROM AUTHOR]

Most image content modelling methods are designed for English description which is different form Chinese in syntax structure. The few existing Chinese image description models do not fully integrate the global features and the local features of an image, limiting the capability of the models to represent the details of the image. In this paper, an encoder-decoder architecture based on the fusion of global and local features is used to describe the Chinese image content. In the encoding stage, the global and local features of the image are extracted by the Convolutional Neural Network (CNN) and the target detection network, and fed to the feature fusion module. In the decoding stage, an image feature attention mechanism is used to calculate the weights of word vectors, and a new gating mechanism is added to the traditional Long Short-Term Memory (LSTM) network to emphasize the fused image features, and the corresponding word vectors. In the description generation stage, the beam search algorithm is used to optimize the word vector generation process. The integration of global and local features of the image is strengthened to allow the model to fully understand the details of the image through the above three stages. The experimental results show that the model improves the quality of Chinese description of image content. Compared with the baseline model, the score of CIDEr evaluation index improves by 20.07%, and other evaluation indices also improve significantly. [ABSTRACT FROM AUTHOR]

The measurement of work time for individual tasks by using video has made a significant contribution to a framework for productivity improvement such as value stream mapping (VSM). In the past, the work time has been often measured manually, but this process is quite costly and labor-intensive. For these reasons, automation of work analysis at the worksite is needed. There are two main methods for computing spatio-temporal information: by 3D-CNN, and by temporal computation using LSTM after feature extraction in the spatial domain by 2D-CNN. These methods has high computational cost but high model representational power, and the latter has low computational cost but relatively low model representational power. In the manufacturing industry, the use of local computers to make inferences is often required for practicality and confidentiality reasons, necessitating a low computational cost, and so the latter, a lightweight model, needs to have improved performance. Therefore, in this paper, we propose a method that pre-trains the image encoder module of a work detection model using an image segmentation model. This is based on the CNN-LSTM structure, which separates spatial and temporal computation and enables us to include heuristics such as workers' body parts and work tools in the CNN module. Experimental results demonstrate that our pre-training method reduces over-fitting and provides a greater improvement in detection performance than pre-training on ImageNet. [ABSTRACT FROM AUTHOR]

With the development of recent years, the field of deep learning has made great progress. Compared with the traditional machine learning algorithm, deep learning can better find the rules in the data and achieve better fitting effect. In this paper, we propose a hybrid stock forecasting model based on Feature Selection, Convolutional Neural Network and Bidirectional Gated Recurrent Unit (FS-CNN-BGRU). Feature Selection (FS) can select the data with better performance for the results as the input data after data normalization. Convolutional Neural Network (CNN) is responsible for feature extraction. It can extract the local features of the data, pay attention to more local information, and reduce the amount of calculation. The Bidirectional Gated Recurrent Unit (BGRU) can process the data with time series, so that it can have better performance for the data with time series attributes. In the experiment, we used single CNN, LSTM and GRU models and mixed models CNN-LSTM, CNN-GRU and FS-CNN-BGRU (the model used in this manuscript). The results show that the performance of the hybrid model (FS-CNN-BGRU) is better than other single models, which has a certain reference value. [ABSTRACT FROM AUTHOR]

The accurate segmentation of retinal vessels images can not only be used to evaluate and monitor various ophthalmic diseases, but also timely reflect systemic diseases such as diabetes and blood diseases. Therefore, the study on segmentation of retinal vessels images is of great significance for the diagnosis of visually threatening diseases. In recent years, especially the convolutional neural networks (CNN) based on UNet and its variant have been widely used in various medical image tasks. However, although CNN has achieved excellent performance, it cannot learn global and long-distance semantic information interaction well due to the local computing characteristics of convolution operation, which limits the development of medical image segmentation tasks. Transformer, currently popular in computer vision, has global computing features, but due to the lack of low-level details, local feature information extraction is insufficient. In this paper, we propose Patches Convolution Attention based Transformer UNet (PCAT-UNet), which is a U-shaped network based on Transformer with a Convolution branch. We use skip connection to fuse the deep and shallow features of both sides. By taking advantage of the complementary advantages of both sides, we can effectively capture the global dependence relationship and the details of the underlying feature space, thus improving the current problems of insufficient extraction of retinal micro vessels feature information and low sensitivity caused by easily predicting of pixels as background. In addition, our method enables end-to-end training and rapid inference. Finally, three publicly available retinal vessels datasets (DRIVE, STARE and CHASE_DB1) were used to evaluate PCAT-UNet. The experimental results show that the proposed PCAT-UNET method achieves good retinal vessel segmentation performance on these three datasets, and is superior to other architectures in terms of AUC, Accuracy and Sensitivity performance indicators. AUC reached 0.9872, 0.9953 and 0.9925, Accuracy reached 0.9622, 0.9796 and 0.9812, Sensitivity reached 0.8576, 0.8703 and 0.8493, respectively. In addition, PCAT-UNET also achieved good results in two other F1-Score and Specificity indicators. [ABSTRACT FROM AUTHOR]

When approaching thyroid gland tumor classification, the differentiation between samples with and without "papillary thyroid carcinoma-like" nuclei is a daunting task with high inter-observer variability among pathologists. Thus, there is increasing interest in the use of machine learning approaches to provide pathologists real-time decision support. In this paper, we optimize and quantitatively compare two automated machine learning methods for thyroid gland tumor classification on two datasets to assist pathologists in decision-making regarding these methods and their parameters. The first method is a feature-based classification originating from common image processing and consists of cell nucleus segmentation, feature extraction, and subsequent thyroid gland tumor classification utilizing different classifiers. The second method is a deep learning-based classification which directly classifies the input images with a convolutional neural network without the need for cell nucleus segmentation. On the Tharun and Thompson dataset, the feature-based classification achieves an accuracy of 89.7% (Cohen's Kappa 0.79), compared to the deep learning-based classification of 89.1% (Cohen's Kappa 0.78). On the Nikiforov dataset, the feature-based classification achieves an accuracy of 83.5% (Cohen's Kappa 0.46) compared to the deep learning-based classification 77.4% (Cohen's Kappa 0.35). Thus, both automated thyroid tumor classification methods can reach the classification level of an expert pathologist. To our knowledge, this is the first study comparing feature-based and deep learning-based classification regarding their ability to classify samples with and without papillary thyroid carcinoma-like nuclei on two large-scale datasets. [ABSTRACT FROM AUTHOR]

Low-end LiDAR sensor provides an alternative for depth measurement and object recognition for lightweight devices. However due to low computing capacity, complicated algorithms are incompatible to be performed on the device, with sparse information further limits the feature available for extraction. Therefore, a classification method which could receive sparse input, while providing ample leverage for the classification process to accurately differentiate objects within limited computing capability is required. To achieve reliable feature extraction from a sparse LiDAR point cloud, this paper proposes a novel Clustered Extraction and Centroid Based Clustered Extraction Method (CE-CBCE) method for feature extraction followed by a convolutional neural network (CNN) object classifier. The integration of the CE-CBCE and CNN methods enable us to utilize lightweight actuated LiDAR input and provides low computing means of classification while maintaining accurate detection. Based on genuine LiDAR data, the final result shows reliable accuracy of 97% through the method proposed. [ABSTRACT FROM AUTHOR]

Occupationally-induced back pain is a leading cause of reduced productivity in industry. Detecting when a worker is lifting incorrectly and at increased risk of back injury presents significant possible benefits. These include increased quality of life for the worker due to lower rates of back injury and fewer workers' compensation claims and missed time for the employer. However, recognizing lifting risk provides a challenge due to typically small datasets and subtle underlying features in accelerometer and gyroscope data. A novel method to classify a lifting dataset using a 2D convolutional neural network (CNN) and no manual feature extraction is proposed in this paper; the dataset consisted of 10 subjects lifting at various relative distances from the body with 720 total trials. The proposed deep CNN displayed greater accuracy (90.6%) compared to an alternative CNN and multilayer perceptron (MLP). A deep CNN could be adapted to classify many other activities that traditionally pose greater challenges in industrial environments due to their size and complexity. [ABSTRACT FROM AUTHOR]