Search

Aiming at the problem that the visual communication image enhances the features from a single angle, resulting in the fuzzy phenomenon of some details, a double branch network enhancement algorithm for deep features of visual communication image is proposed. The image is preprocessed by bilateral filtering, and then the global noise level of the image is obtained by blind estimation, which is used as the residual threshold in the sparse reconstruction process, and the image filtering processing is realized by combining sparse coding and dictionary learning algorithm; Calculate the image saliency value, establish the saliency matrix, divide the foreground area and background area of the image, and use the histogram equalization method to carry out histogram equalization processing on the foreground and background area respectively to realize the global enhancement of the visual communication image; The enhanced image is input into the dual branch network, and the spatial domain information and the frequency domain information after Fourier transform of the image are processed in the spatial domain branch and the frequency domain branch respectively, and the features of the two branches are guided to conduct adaptive fusion through the attention mechanism to obtain the final enhanced image. Experimental results show that the proposed method has good image denoising and enhancement effects. [ABSTRACT FROM AUTHOR]

With the continuous maturity of hyperspectral remote sensing imaging technology, it has been widely adopted by scholars to improve the performance of feature classification. However, due to the challenges in acquiring hyperspectral images and producing training samples, the limited training sample is a common problem that researchers often face. Furthermore, efficient algorithms are necessary to excavate the spatial and spectral information from these images, and then, make full use of this information with limited training samples. To solve this problem, a novel two-branch deep learning network model is proposed for extracting hyperspectral remote sensing features in this paper. In this model, one branch focuses on extracting spectral features using multi-scale convolution and a normalization-based attention module, while the other branch captures spatial features through small-scale dilation convolution and Euclidean Similarity Attention. Subsequently, pooling and layering techniques are employed to further extract abstract features after feature fusion. In the experiments conducted on two public datasets, namely, IP and UP, as well as our own labeled dataset, namely, YRE, the proposed DMAN achieves the best classification results, with overall accuracies of 96.74%, 97.4%, and 98.08%, respectively. Compared to the sub-optimal state-of-the-art methods, the overall accuracies are improved by 1.05, 0.42, and 0.51 percentage points, respectively. The advantage of this network structure is particularly evident in unbalanced sample environments. Additionally, we introduce a new strategy based on the RPNet, which utilizes a small number of principal components for feature classification after dimensionality reduction. The results demonstrate its effectiveness in uncovering compressed feature information, with an overall accuracy improvement of 0.68 percentage points. Consequently, our model helps mitigate the impact of data scarcity on model performance, thereby contributing positively to the advancement of hyperspectral remote sensing technology in practical applications. [ABSTRACT FROM AUTHOR]

In order to minimize additional computational costs in detecting forged videos, and enhance detection accuracy, this paper employs dynamic facial expression sequences as key sequences, replacing original video sequences as inputs for the detection model. A spatio-temporal dual-branch detection network is designed based on the visual Transformer architecture. Specifically, this process involves three steps. Firstly, dynamic facial expression sequences are localized as key sequences using optical flow difference algorithms. Subsequently, the spatial branch network employs the focal self-attention mechanism to focus on dynamic features of expression-relevant regions and uses Factorization Machines to facilitate feature interaction among multiple key sequences. Meanwhile, the temporal branch network concentrates on learning the temporal inconsistency of optical flow differences between adjacent frames. Finally, a binary classification linear SVM combines the Softmax values from the two branch networks to provide the ultimate detection outcome. Experimental results on the Faceforensics++ dataset demonstrate: (a) replacing whole video sequences with facial expression key sequences effectively reduces training and detection time by nearly 80% and 90%, respectively; (b) compared to state-of-the-art methods involving random sequence/frame extraction and key frame extraction based on video compression techniques, the proposed approach in this paper presents a more competitive detection accuracy.

Wound image classification is a crucial preprocessing step to many intelligent medical systems, e.g., online diagnosis and smart medical. Recently, Convolutional Neural Network (CNN) has been widely applied to the classification of wound images and obtained promising performance to some extent. Unfortunately, it is still challenging to classify multiple wound types due to the complexity and variety of wound images. Existing CNNs usually extract high- and low-frequency features at the same convolutional layer, which inevitably causes information loss and further affects the accuracy of classification. To this end, we propose a novel High and Low-frequency Guidance Network (HLG-Net) for multi-class wound classification. To be specific, HLG-Net contains two branches: High-Frequency Network (HF-Net) and Low-Frequency Network (LF-Net). We employ pre-trained models ResNet and Res2Net as the feature backbone of the HF-Net, which makes the network capture the high-frequency details and texture information of wound images. To extract much low-frequency information, we utilize a Multi-Stream Dilation Convolution Residual Block (MSDCRB) as the backbone of the LF-Net. Moreover, a fusion module is proposed to fully explore informative features at the end of these two separate feature extraction branches, and obtain the final classification result. Extensive experiments demonstrate that HLG-Net can achieve maximum accuracy of 98.00%, 92.11%, and 82.61% in two-class, three-class, and four-class wound image classifications, respectively, which outperforms the previous state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Multi-modal fusion of remote sensing images poses challenges because of the intricate imaging mechanisms and variations in radiation across different modalities. Specifically, the fusion of visible-light and vegetation-sensitive images encounters similar difficulties. Traditional methods have seldom considered the varied imaging mechanisms and radiation difference between modalities, resulting in discrepancies in the correspond features. To address the issue, we propose the VAM-Net (Vegetation-Attentive Multi-modal deep Network) combining a radiometric correction mechanism and a lightweight multi-modal adaptive feature selection method for fusing multi-modal images. First, the vegetation index (VDVI) is integrated into visible-light images to mitigate the radiometric differences between visible-light images and vegetation-sensitive images (e.g., infrared and red edge images). Then, a two-branch network incorporating attention mechanisms is designed to independently capture the texture features and select similar features cross two different modalities of images. Last, a new loss function is presented to ensure the learned features are suitable for multi-modal fusion. The VAM-Net is evaluated by visible-light and vegetation-sensitive images in three different areas, and the experimental results show that VAM-Net attains an average precision of 67.02%, and recall of 35.49%, and an average RMSE of 2.191px, demonstrating the accuracy and robustness of VAM-Net in multi-modal image fusion.

With the continuous maturity of hyperspectral remote sensing imaging technology, it has been widely adopted by scholars to improve the performance of feature classification. However, due to the challenges in acquiring hyperspectral images and producing training samples, the limited training sample is a common problem that researchers often face. Furthermore, efficient algorithms are necessary to excavate the spatial and spectral information from these images, and then, make full use of this information with limited training samples. To solve this problem, a novel two-branch deep learning network model is proposed for extracting hyperspectral remote sensing features in this paper. In this model, one branch focuses on extracting spectral features using multi-scale convolution and a normalization-based attention module, while the other branch captures spatial features through small-scale dilation convolution and Euclidean Similarity Attention. Subsequently, pooling and layering techniques are employed to further extract abstract features after feature fusion. In the experiments conducted on two public datasets, namely, IP and UP, as well as our own labeled dataset, namely, YRE, the proposed DMAN achieves the best classification results, with overall accuracies of 96.74%, 97.4%, and 98.08%, respectively. Compared to the sub-optimal state-of-the-art methods, the overall accuracies are improved by 1.05, 0.42, and 0.51 percentage points, respectively. The advantage of this network structure is particularly evident in unbalanced sample environments. Additionally, we introduce a new strategy based on the RPNet, which utilizes a small number of principal components for feature classification after dimensionality reduction. The results demonstrate its effectiveness in uncovering compressed feature information, with an overall accuracy improvement of 0.68 percentage points. Consequently, our model helps mitigate the impact of data scarcity on model performance, thereby contributing positively to the advancement of hyperspectral remote sensing technology in practical applications.

Reliable quality control of laser welding on power batteries is an important issue due to random interference in the production process. In this paper, a quality inspection framework based on a two-branch network and conventional image processing is proposed to predict welding quality while outputting corresponding parameter information. The two-branch network consists of a segmentation network and a classification network, which alleviates the problem of large training sample size requirements for deep learning by sharing feature representations among two related tasks. Moreover, coordinate attention is introduced into feature learning modules of the network to effectively capture the subtle features of defective welds. Finally, a post-processing method based on the Hough transform is used to extract the information of the segmented weld region. Extensive experiments demonstrate that the proposed model can achieve a significant classification performance on the dataset collected on an actual production line. This study provides a valuable reference for an intelligent quality inspection system in the power battery manufacturing industry. [ABSTRACT FROM AUTHOR]

This study proposes a fusion method of infrared and visible images based on feature interaction. Existing fusion methods can be classified into two categories based on a single-branch network and a two-branch network. Generative adversarial networks are widely used in single-branch-based fusion methods, which ignore the difference in feature extraction caused by different input images. Most two-branch-based fusion methods use convolutional neural networks, which do not take into account the inverse promotion of fusion results and lack the interaction between different input features. To remedy the shortcomings of these fusion methods and better utilize the feature from source images, this study proposes a two-branch feature interactions method based on a generative adversarial network for visible and infrared image fusion. In the generator part, a two-branch feature interaction approach was designed to extract features from different inputs and realize feature interaction through the network connection of different branches. In the discriminator part, a double-classification discriminator was used for visible images and infrared images. Extensive comparison experiments with state-of-the-art methods have demonstrated the advantages of this proposed generative adversarial network based on two-branch feature interaction, which can enhance the texture details of objects in fusion results and reduce the interference of noise information from source inputs. In addition, the above advantages were also confirmed in generalization experiments of object detection. [ABSTRACT FROM AUTHOR]

In complex marine environments, underwater images often suffer from color distortion, blur, and poor visibility. Existing underwater image enhancement methods predominantly rely on the U-net structure, which assigns the same weight to different resolution information. However, this approach lacks the ability to extract sufficient detailed information, resulting in problems such as blurred details and color distortion. We propose a two-branch underwater image enhancement method with an optimized original resolution information strategy to address this limitation. Our method comprises a feature enhancement subnetwork (FEnet) and an original resolution subnetwork (ORSnet). FEnet extracts multi-resolution information and utilizes an adaptive feature selection module to enhance global features in different dimensions. The enhanced features are then fed into ORSnet as complementary features, which extract local enhancement features at the original image scale to achieve semantically consistent and visually superior enhancement effects. Experimental results on the UIEB dataset demonstrate that our method achieves the best performance compared to the state-of-the-art methods. Furthermore, through comprehensive application testing, we have validated the superiority of our proposed method in feature extraction and enhancement compared to other end-to-end underwater image enhancement methods. [ABSTRACT FROM AUTHOR]

Wound image classification is a crucial preprocessing step to many intelligent medical systems, e.g., online diagnosis and smart medical. Recently, Convolutional Neural Network (CNN) has been widely applied to the classification of wound images and obtained promising performance to some extent. Unfortunately, it is still challenging to classify multiple wound types due to the complexity and variety of wound images. Existing CNNs usually extract high- and low-frequency features at the same convolutional layer, which inevitably causes information loss and further affects the accuracy of classification. To this end, we propose a novel High and Low-frequency Guidance Network (HLG-Net) for multi-class wound classification. To be specific, HLG-Net contains two branches: High-Frequency Network (HF-Net) and Low-Frequency Network (LF-Net). We employ pre-trained models ResNet and Res2Net as the feature backbone of the HF-Net, which makes the network capture the high-frequency details and texture information of wound images. To extract much low-frequency information, we utilize a Multi-Stream Dilation Convolution Residual Block (MSDCRB) as the backbone of the LF-Net. Moreover, a fusion module is proposed to fully explore informative features at the end of these two separate feature extraction branches, and obtain the final classification result. Extensive experiments demonstrate that HLG-Net can achieve maximum accuracy of 98.00%, 92.11%, and 82.61% in two-class, three-class, and four-class wound image classifications, respectively, which outperforms the previous state-of-the-art methods.

In order to minimize additional computational costs in detecting forged videos, and enhance detection accuracy, this paper employs dynamic facial expression sequences as key sequences, replacing original video sequences as inputs for the detection model. A spatio-temporal dual-branch detection network is designed based on the visual Transformer architecture. Specifically, this process involves three steps. Firstly, dynamic facial expression sequences are localized as key sequences using optical flow difference algorithms. Subsequently, the spatial branch network employs the focal self-attention mechanism to focus on dynamic features of expression-relevant regions and uses Factorization Machines to facilitate feature interaction among multiple key sequences. Meanwhile, the temporal branch network concentrates on learning the temporal inconsistency of optical flow differences between adjacent frames. Finally, a binary classification linear SVM combines the Softmax values from the two branch networks to provide the ultimate detection outcome. Experimental results on the Faceforensics++ dataset demonstrate: (a) replacing whole video sequences with facial expression key sequences effectively reduces training and detection time by nearly 80% and 90%, respectively; (b) compared to state-of-the-art methods involving random sequence/frame extraction and key frame extraction based on video compression techniques, the proposed approach in this paper presents a more competitive detection accuracy. [ABSTRACT FROM AUTHOR]

Reliable quality control of laser welding on power batteries is an important issue due to random interference in the production process. In this paper, a quality inspection framework based on a two-branch network and conventional image processing is proposed to predict welding quality while outputting corresponding parameter information. The two-branch network consists of a segmentation network and a classification network, which alleviates the problem of large training sample size requirements for deep learning by sharing feature representations among two related tasks. Moreover, coordinate attention is introduced into feature learning modules of the network to effectively capture the subtle features of defective welds. Finally, a post-processing method based on the Hough transform is used to extract the information of the segmented weld region. Extensive experiments demonstrate that the proposed model can achieve a significant classification performance on the dataset collected on an actual production line. This study provides a valuable reference for an intelligent quality inspection system in the power battery manufacturing industry.

In complex marine environments, underwater images often suffer from color distortion, blur, and poor visibility. Existing underwater image enhancement methods predominantly rely on the U-net structure, which assigns the same weight to different resolution information. However, this approach lacks the ability to extract sufficient detailed information, resulting in problems such as blurred details and color distortion. We propose a two-branch underwater image enhancement method with an optimized original resolution information strategy to address this limitation. Our method comprises a feature enhancement subnetwork (FEnet) and an original resolution subnetwork (ORSnet). FEnet extracts multi-resolution information and utilizes an adaptive feature selection module to enhance global features in different dimensions. The enhanced features are then fed into ORSnet as complementary features, which extract local enhancement features at the original image scale to achieve semantically consistent and visually superior enhancement effects. Experimental results on the UIEB dataset demonstrate that our method achieves the best performance compared to the state-of-the-art methods. Furthermore, through comprehensive application testing, we have validated the superiority of our proposed method in feature extraction and enhancement compared to other end-to-end underwater image enhancement methods.

Copyright of Journal of Computer Engineering & Applications is the property of Beijing Journal of Computer Engineering & Applications Journal Co Ltd. and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Copyright of Journal of Southeast University / Dongnan Daxue Xuebao is the property of Journal of Southeast University Editorial Office and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Gait has been deemed as an alternative biometric in video-based surveillance applications, since it can be used to recognize individuals from a far distance without their interaction and cooperation. Recently, many gait recognition methods have been proposed, aiming at reducing the influence caused by exterior factors. However, most of these methods are developed based on sufficient input gait frames, and their recognition performance will sharply decrease if the frame number drops. In the real-world scenario, it is impossible to always obtain a sufficient number of gait frames for each subject due to many reasons, e.g., occlusion and illumination. Therefore, it is necessary to improve the gait recognition performance when the available gait frames are limited. This paper starts with three different strategies, aiming at producing more input frames and eliminating the generalization error cause by insufficient input data. Meanwhile, a two-branch network is also proposed in this paper to formulate robust gait representations from the original and new generated input gait frames. According to our experiments, under the limited gait frames being used, it was verified that the proposed method can achieve a reliable performance for gait recognition.

• The proposed framework effectively enhances the performance of deep face recognition for low illumination images. • Our feature restoration network achieves computational efficiency at the cost of only a few more parameters and FLOPs compared to the original feature extraction model. In addition, the training of this network does not require a massive dataset. • Our feature restoration network and embedding refinement module can be integrated with most existing deep face recognition models to create an end-to-end face recognition pipeline. The performance of many state-of-the-art deep face recognition models deteriorates significantly for images captured under low illumination, mainly because the features of dim probe face images cannot match well with those of normal-illumination gallery images. The issue cannot be satisfactorily addressed by enhancing the illumination of face images and performing face recognition on the resulted images alone. We propose a novel deep face recognition framework that consists of a feature restoration network, a feature extraction network, and an embedding matching module. The feature restoration network adopts a two-branch structure based on the convolutional neural network to generate a feature image from the raw image and the illumination-enhanced image. The feature extraction network encodes the feature image into an embedding, which is then used by the embedding matching module for face verification and identification. The overall verification accuracy is improved from 1.1% to 6.7% when tested on the Specs on Faces (SoF) dataset. For face identification, the rank-1 identification accuracy is improved by 2.8%. [ABSTRACT FROM AUTHOR]

Gait has been deemed as an alternative biometric in video-based surveillance applications, since it can be used to recognize individuals from a far distance without their interaction and cooperation. Recently, many gait recognition methods have been proposed, aiming at reducing the influence caused by exterior factors. However, most of these methods are developed based on sufficient input gait frames, and their recognition performance will sharply decrease if the frame number drops. In the real-world scenario, it is impossible to always obtain a sufficient number of gait frames for each subject due to many reasons, e.g., occlusion and illumination. Therefore, it is necessary to improve the gait recognition performance when the available gait frames are limited. This paper starts with three different strategies, aiming at producing more input frames and eliminating the generalization error cause by insufficient input data. Meanwhile, a two-branch network is also proposed in this paper to formulate robust gait representations from the original and new generated input gait frames. According to our experiments, under the limited gait frames being used, it was verified that the proposed method can achieve a reliable performance for gait recognition., Competing Interests: The authors declare there are no competing interests., (©2021 Yao et al.)