Anping Song
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Anping Song has a total of 43 co-authors. Here are the top co-authors by number of shared publications:
Jianjiao Chen
University of Miami
3 shared publicationsXuehai Ding
School of Computer Engineering and Science, Shanghai University, Shanghai, China
3 shared publications- Xuehai Ding
School of Computer Engineering and Science, Shanghai University, Shanghai, China
3 shared publications - Chenbei Zhang
School of Computer Engineering and Science, Shanghai University, Shanghai 200444, China
2 shared publications - Wendong Qi
School of Computer Engineering and Science, Shanghai University, Shanghai 200444, China
2 shared publications - Xinyi Zhai
School of Computer Engineering and Science, Shanghai University, Shanghai 200444, China
2 shared publications
Latest Publications
- Open Access
FFICL-Net: A Fusing Symmetric Feature-Importance Ranking Contrastive-Learning Network for Multivariate Time-Series Classification
Supervised contrastive learning has emerged as a novel method to help solve the problem of multivariate time-series classification. By utilizing labeled data, it maximally learns the feature-representation differences between various categories. However, existing supervised contrastive-learning approaches lack model interpretability, making it difficult to analyze the importance ranking among features. Experimentally, different preprocessing of the data often results in variations in feature-importance ranking. Therefore, we propose FFICL-Net, which combines LSTM, to analyze the importance of sequence variables, and ITransformer, to treat each variable as a token, learning the self-attention relationships between variables and their relationship to the final feature representation. This approach contrasts the feature importance derived from the two modules, making two feature-importance ranking results more similar and forming a kind of symmetry, allowing the resultant feature representation to fuse the characteristics of both models, leading to more stable and accurate feature-importance ranking results and aiding in improving classification accuracy. We conducted comparative experiments on all 30 public UEA datasets and achieved the best results on most of these datasets compared to the current top models. The average accuracy reached 72.8%, which is an improvement of 0.7% over the best-performing model.
Open AccessFFICL-Net: A Fusing Symmetric Feature-Importance Ranking Contrastive-Learning Network for Multivariate Time-Series Classification
Supervised contrastive learning has emerged as a novel method to help solve the problem of multivariate time-series classification. By utilizing labeled data, it maximally learns the feature-representation differences between various categories. However, existing supervised contrastive-learning approaches lack model interpretability, making it difficult to analyze the importance ranking among features. Experimentally, different preprocessing of the data often results in variations in feature-importance ranking. Therefore, we propose FFICL-Net, which combines LSTM, to analyze the importance of sequence variables, and ITransformer, to treat each variable as a token, learning the self-attention relationships between variables and their relationship to the final feature representation. This approach contrasts the feature importance derived from the two modules, making two feature-importance ranking results more similar and forming a kind of symmetry, allowing the resultant feature representation to fuse the characteristics of both models, leading to more stable and accurate feature-importance ranking results and aiding in improving classification accuracy. We conducted comparative experiments on all 30 public UEA datasets and achieved the best results on most of these datasets compared to the current top models. The average accuracy reached 72.8%, which is an improvement of 0.7% over the best-performing model.
- Open Access
- 10 February 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14 (4) , 1461
FlexBFT: A Flexible and Effective Optimistic Asynchronous BFT Protocol
Currently, integrating partially synchronous Byzantine-fault-tolerant protocols into asynchronous protocols as fast lanes represents a trade-off between robustness and efficiency, a concept known as optimistic asynchronous protocols. Existing optimistic asynchronous protocols follow a fixed path order: they execute the fast lane first, switch to the slow lane after a timeout failure, and restart the fast lane after the slow lane execution is completed. However, when confronted with prolonged network fluctuations, this fixed path sequence results in frequent failures and fast lane switches, leading to overhead that diminishes the efficiency of optimistic asynchronous protocols compared with their asynchronous counterparts. In response to this challenge, this article introduces FlexBFT, a novel and flexible optimistic asynchronous consensus framework designed to significantly enhance overall consensus performance. The key innovation behind FlexBFT lies in the persistence of slow lanes. In the presence of persistent network latency, FlexBFT can continually operate round after round within the slow lane—the current optimal path—until the network conditions improve. Furthermore, FlexBFT offers the flexibility to combine consensus modules adaptively, further enhancing its performance. Particularly in challenging network conditions, FlexBFT’s experimental outcomes highlight its superiority across a range of network scenarios compared with state-of-the-art algorithms. It achieves a performance with 31.6% lower latency than BDT, effectively merging the low latency characteristic of deterministic protocols with the robustness inherent in asynchronous protocols.
- 10 February 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14 (4) , 1461
Open AccessFlexBFT: A Flexible and Effective Optimistic Asynchronous BFT Protocol
Currently, integrating partially synchronous Byzantine-fault-tolerant protocols into asynchronous protocols as fast lanes represents a trade-off between robustness and efficiency, a concept known as optimistic asynchronous protocols. Existing optimistic asynchronous protocols follow a fixed path order: they execute the fast lane first, switch to the slow lane after a timeout failure, and restart the fast lane after the slow lane execution is completed. However, when confronted with prolonged network fluctuations, this fixed path sequence results in frequent failures and fast lane switches, leading to overhead that diminishes the efficiency of optimistic asynchronous protocols compared with their asynchronous counterparts. In response to this challenge, this article introduces FlexBFT, a novel and flexible optimistic asynchronous consensus framework designed to significantly enhance overall consensus performance. The key innovation behind FlexBFT lies in the persistence of slow lanes. In the presence of persistent network latency, FlexBFT can continually operate round after round within the slow lane—the current optimal path—until the network conditions improve. Furthermore, FlexBFT offers the flexibility to combine consensus modules adaptively, further enhancing its performance. Particularly in challenging network conditions, FlexBFT’s experimental outcomes highlight its superiority across a range of network scenarios compared with state-of-the-art algorithms. It achieves a performance with 31.6% lower latency than BDT, effectively merging the low latency characteristic of deterministic protocols with the robustness inherent in asynchronous protocols.
- Open Access
- 1 February 2024
- Published by Frontiers Media SA in Frontiers in Bioengineering and Biotechnology
- Vol. 12, 1330713
TSE-GAN: strain elastography using generative adversarial network for thyroid disease diagnosis
Over the past 35 years, studies conducted worldwide have revealed a threefold increase in the incidence of thyroid cancer. Strain elastography is a new imaging technique to identify benign and malignant thyroid nodules due to its sensitivity to tissue stiffness. However, there are certain limitations of this technique, particularly in terms of standardization of the compression process, evaluation of results and several assumptions used in commercial strain elastography modes for the purpose of simplifying imaging analysis. In this work, we propose a novel conditional generative adversarial network (TSE-GAN) for automatically generating thyroid strain elastograms, which adopts a global-to-local architecture to improve the ability of extracting multi-scale features and develops an adaptive deformable U-net structure in the sub-generator to apply effective deformation. Furthermore, we introduce a Lab-based loss function to induce the networks to generate realistic thyroid elastograms that conform to the probability distribution of the target domain. Qualitative and quantitative assessments are conducted on a clinical dataset provided by Shanghai Sixth People’s Hospital. Experimental results demonstrate that thyroid elastograms generated by the proposed TSE-GAN outperform state-of-the-art image translation methods in meeting the needs of clinical diagnostic applications and providing practical value.
- 1 February 2024
- Published by Frontiers Media SA in Frontiers in Bioengineering and Biotechnology
- Vol. 12, 1330713
Open AccessTSE-GAN: strain elastography using generative adversarial network for thyroid disease diagnosis
Over the past 35 years, studies conducted worldwide have revealed a threefold increase in the incidence of thyroid cancer. Strain elastography is a new imaging technique to identify benign and malignant thyroid nodules due to its sensitivity to tissue stiffness. However, there are certain limitations of this technique, particularly in terms of standardization of the compression process, evaluation of results and several assumptions used in commercial strain elastography modes for the purpose of simplifying imaging analysis. In this work, we propose a novel conditional generative adversarial network (TSE-GAN) for automatically generating thyroid strain elastograms, which adopts a global-to-local architecture to improve the ability of extracting multi-scale features and develops an adaptive deformable U-net structure in the sub-generator to apply effective deformation. Furthermore, we introduce a Lab-based loss function to induce the networks to generate realistic thyroid elastograms that conform to the probability distribution of the target domain. Qualitative and quantitative assessments are conducted on a clinical dataset provided by Shanghai Sixth People’s Hospital. Experimental results demonstrate that thyroid elastograms generated by the proposed TSE-GAN outperform state-of-the-art image translation methods in meeting the needs of clinical diagnostic applications and providing practical value.
- Open Access
- 19 January 2023
- Published by MDPI AG in Applied Sciences
- Vol. 13 (3) , 1367
RGCLN: Relational Graph Convolutional Ladder-Shaped Networks for Signed Network Clustering
Node embeddings are increasingly used in various analysis tasks of networks due to their excellent dimensional compression and feature representation capabilities. However, most researchers’ priorities have always been link prediction, which leads to signed network clustering being under-explored. Therefore, we propose an asymmetric ladder-shaped architecture called RGCLN based on multi-relational graph convolution that can fuse deep node features to generate node representations with great representational power. RGCLN adopts a deep framework to capture and convey information instead of using the common method in signed networks—balance theory. In addition, RGCLN adds a size constraint to the loss function to prevent image-like overfitting during the unsupervised learning process. Based on the node features learned by this end-to-end trained model, RGCLN performs community detection in a large number of real-world networks and generative networks, and the results indicate that our model has an advantage over state-of-the-art network embedding algorithms.
- 19 January 2023
- Published by MDPI AG in Applied Sciences
- Vol. 13 (3) , 1367
Open AccessRGCLN: Relational Graph Convolutional Ladder-Shaped Networks for Signed Network Clustering
Node embeddings are increasingly used in various analysis tasks of networks due to their excellent dimensional compression and feature representation capabilities. However, most researchers’ priorities have always been link prediction, which leads to signed network clustering being under-explored. Therefore, we propose an asymmetric ladder-shaped architecture called RGCLN based on multi-relational graph convolution that can fuse deep node features to generate node representations with great representational power. RGCLN adopts a deep framework to capture and convey information instead of using the common method in signed networks—balance theory. In addition, RGCLN adds a size constraint to the loss function to prevent image-like overfitting during the unsupervised learning process. Based on the node features learned by this end-to-end trained model, RGCLN performs community detection in a large number of real-world networks and generative networks, and the results indicate that our model has an advantage over state-of-the-art network embedding algorithms.
- Open Access
- 1 January 2023
- Published by American Institute of Mathematical Sciences (AIMS) in Mathematical Biosciences and Engineering
- Vol. 20 (6) , 10153-10173
GL-FusionNet: Fusing global and local features to classify deep and superficial partial thickness burn
Burns constitute one of the most common injuries in the world, and they can be very painful for the patient. Especially in the judgment of superficial partial thickness burns and deep partial thickness burns, many inexperienced clinicians are easily confused. Therefore, in order to make burn depth classification automated as well as accurate, we have introduced the deep learning method. This methodology uses a U-Net to segment burn wounds. On this basis, a new thickness burn classification model that fuses global and local features (GL-FusionNet) is proposed. For the thickness burn classification model, we use a ResNet50 to extract local features, use a ResNet101 to extract global features, and finally implement the add method to perform feature fusion and obtain the deep partial or superficial partial thickness burn classification results. Burns images are collected clinically, and they are segmented and labeled by professional physicians. Among the segmentation methods, the U-Net used achieved a Dice score of 85.352 and IoU score of 83.916, which are the best results among all of the comparative experiments. In the classification model, different existing classification networks are mainly used, as well as a fusion strategy and feature extraction method that are adjusted to conduct experiments; the proposed fusion network model also achieved the best results. Our method yielded the following: accuracy of 93.523, recall of 93.67, precision of 93.51, and F1-score of 93.513. In addition, the proposed method can quickly complete the auxiliary diagnosis of the wound in the clinic, which can greatly improve the efficiency of the initial diagnosis of burns and the nursing care of clinical medical staff.
- 1 January 2023
- Published by American Institute of Mathematical Sciences (AIMS) in Mathematical Biosciences and Engineering
- Vol. 20 (6) , 10153-10173
Open AccessGL-FusionNet: Fusing global and local features to classify deep and superficial partial thickness burn
Burns constitute one of the most common injuries in the world, and they can be very painful for the patient. Especially in the judgment of superficial partial thickness burns and deep partial thickness burns, many inexperienced clinicians are easily confused. Therefore, in order to make burn depth classification automated as well as accurate, we have introduced the deep learning method. This methodology uses a U-Net to segment burn wounds. On this basis, a new thickness burn classification model that fuses global and local features (GL-FusionNet) is proposed. For the thickness burn classification model, we use a ResNet50 to extract local features, use a ResNet101 to extract global features, and finally implement the add method to perform feature fusion and obtain the deep partial or superficial partial thickness burn classification results. Burns images are collected clinically, and they are segmented and labeled by professional physicians. Among the segmentation methods, the U-Net used achieved a Dice score of 85.352 and IoU score of 83.916, which are the best results among all of the comparative experiments. In the classification model, different existing classification networks are mainly used, as well as a fusion strategy and feature extraction method that are adjusted to conduct experiments; the proposed fusion network model also achieved the best results. Our method yielded the following: accuracy of 93.523, recall of 93.67, precision of 93.51, and F1-score of 93.513. In addition, the proposed method can quickly complete the auxiliary diagnosis of the wound in the clinic, which can greatly improve the efficiency of the initial diagnosis of burns and the nursing care of clinical medical staff.
- 18 November 2021
- Published by Springer Nature
- p. 118-129
An Efficient Two-Stage Fusion Network for Computer-Aided Diagnosis of Diabetic Foot
- 18 November 2021
- Published by Springer Nature
- p. 118-129
An Efficient Two-Stage Fusion Network for Computer-Aided Diagnosis of Diabetic Foot
- Open Access
- 27 February 2021
- Published by MDPI AG in Electronics
- Vol. 10 (5) , 558
SACN: A Novel Rotating Face Detector Based on Architecture Search
Rotation-Invariant Face Detection (RIPD) has been widely used in practical applications; however, the problem of the adjusting of the rotation-in-plane (RIP) angle of the human face still remains. Recently, several methods based on neural networks have been proposed to solve the RIP angle problem. However, these methods have various limitations, including low detecting speed, model size, and detecting accuracy. To solve the aforementioned problems, we propose a new network, called the Searching Architecture Calibration Network (SACN), which utilizes architecture search, fully convolutional network (FCN) and bounding box center cluster (CC). SACN was tested on the challenging Multi-Oriented Face Detection Data Set and Benchmark (MOFDDB) and achieved a higher detecting accuracy and almost the same speed as existing detectors. Moreover, the average angle error is optimized from the current 12.6° to 10.5°.
- 27 February 2021
- Published by MDPI AG in Electronics
- Vol. 10 (5) , 558
Open AccessSACN: A Novel Rotating Face Detector Based on Architecture Search
Rotation-Invariant Face Detection (RIPD) has been widely used in practical applications; however, the problem of the adjusting of the rotation-in-plane (RIP) angle of the human face still remains. Recently, several methods based on neural networks have been proposed to solve the RIP angle problem. However, these methods have various limitations, including low detecting speed, model size, and detecting accuracy. To solve the aforementioned problems, we propose a new network, called the Searching Architecture Calibration Network (SACN), which utilizes architecture search, fully convolutional network (FCN) and bounding box center cluster (CC). SACN was tested on the challenging Multi-Oriented Face Detection Data Set and Benchmark (MOFDDB) and achieved a higher detecting accuracy and almost the same speed as existing detectors. Moreover, the average angle error is optimized from the current 12.6° to 10.5°.
- Open Access
- 6 March 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 45244-45253
Similar Face Recognition Using the IE-CNN Model
In the field of face recognition, similar face recognition is difficult due to the high degree of similarity of the face structure. The following two factors are needed to make progress in this field: (i) the availability of large scale similar face training datasets, and (ii) a fine-grained face recognition method. With the above factors fulfilled, we make two contributions. First, we show how a large scale similar face dataset (SFD) can be assembled by a combination of automation and human in the loop, and divide the dataset into five grades according to different degrees of similarity. Second, a new fine-grained face feature extraction method is proposed to solve this problem using the attention mechanism which combines the Internal Features and External Features. The Labeled Faces in the Wild (LFW) database, CASIA-WebFace and similar face dataset (SFD) were selected for experiments. It turns out that the true positive rate is improved by 1.94 - 5.66% and the recognition accuracy rate improved by 2.08 - 5.8% for the LFW and CASIA-WebFace database, respectively. Meanwhile for SFD, the recognition accuracy rate improved by 18.80 - 35.84%.
- 6 March 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 45244-45253
Open AccessSimilar Face Recognition Using the IE-CNN Model
In the field of face recognition, similar face recognition is difficult due to the high degree of similarity of the face structure. The following two factors are needed to make progress in this field: (i) the availability of large scale similar face training datasets, and (ii) a fine-grained face recognition method. With the above factors fulfilled, we make two contributions. First, we show how a large scale similar face dataset (SFD) can be assembled by a combination of automation and human in the loop, and divide the dataset into five grades according to different degrees of similarity. Second, a new fine-grained face feature extraction method is proposed to solve this problem using the attention mechanism which combines the Internal Features and External Features. The Labeled Faces in the Wild (LFW) database, CASIA-WebFace and similar face dataset (SFD) were selected for experiments. It turns out that the true positive rate is improved by 1.94 - 5.66% and the recognition accuracy rate improved by 2.08 - 5.8% for the LFW and CASIA-WebFace database, respectively. Meanwhile for SFD, the recognition accuracy rate improved by 18.80 - 35.84%.
- Open Access
- 16 November 2018
- Published by MDPI AG in Future Internet
- Vol. 10 (11) , 111
Neurologist Standard Classification of Facial Nerve Paralysis with Deep Neural Networks
Facial nerve paralysis (FNP) is the most common form of facial nerve damage, which leads to significant physical pain and abnormal function in patients. Traditional FNP detection methods are based on visual diagnosis, which relies solely on the physician’s assessment. The use of objective measurements can reduce the frequency of errors which are caused by subjective methods. Hence, a fast, accurate, and objective computer method for FNP classification is proposed that uses a single Convolutional neural network (CNN), trained end-to-end directly from images, with only pixels and disease labels as inputs. We trained the CNN using a dataset of 1049 clinical images and divided the dataset into 7 categories based on classification standards with the help of neurologists. We tested its performance against the neurologists’ ground truth, and our results matched the neurologists’ level with 97.5% accuracy.
- 16 November 2018
- Published by MDPI AG in Future Internet
- Vol. 10 (11) , 111
Open AccessNeurologist Standard Classification of Facial Nerve Paralysis with Deep Neural Networks
Facial nerve paralysis (FNP) is the most common form of facial nerve damage, which leads to significant physical pain and abnormal function in patients. Traditional FNP detection methods are based on visual diagnosis, which relies solely on the physician’s assessment. The use of objective measurements can reduce the frequency of errors which are caused by subjective methods. Hence, a fast, accurate, and objective computer method for FNP classification is proposed that uses a single Convolutional neural network (CNN), trained end-to-end directly from images, with only pixels and disease labels as inputs. We trained the CNN using a dataset of 1049 clinical images and divided the dataset into 7 categories based on classification standards with the help of neurologists. We tested its performance against the neurologists’ ground truth, and our results matched the neurologists’ level with 97.5% accuracy.
- 31 October 2017
- Published by Springer Nature in Australasian Physics & Engineering Sciences in Medicine
- Vol. 40 (4) , 851-860
Assessment for facial nerve paralysis based on facial asymmetry
Facial nerve paralysis (FNP) is a loss of facial movement due to facial nerve damage, which will lead to significant physical pain and abnormal function in patients. Traditional FNP grading methods are solely based on clinician’s judgment and are time-consuming and subjective. Hence, an accurate, quantitative and objective method of evaluating FNP is proposed for constructing a standard system, which will be an invaluable tool for clinicians who treat the patient with FNP. In this paper, we introduce a novel method for quantitative assessment of FNP which combines an effective facial landmark estimation (FLE) algorithm and facial asymmetrical feature (FAF) by processing facial movement image. The facial landmarks can be detected automatically and accurately using FLE. The FAF is based on the angle of key facial landmark connection and mirror degree of multiple regions on human face. Our method provides significant contribution as it describes the displacement of facial organ and the changes of facial organ exposure during performing facial movements. Experiments show that our method is effective, accurate and convenient in practice, which is beneficial to FNP diagnosis and personalized rehabilitation therapy for each patient.
- 31 October 2017
- Published by Springer Nature in Australasian Physics & Engineering Sciences in Medicine
- Vol. 40 (4) , 851-860
Assessment for facial nerve paralysis based on facial asymmetry
Facial nerve paralysis (FNP) is a loss of facial movement due to facial nerve damage, which will lead to significant physical pain and abnormal function in patients. Traditional FNP grading methods are solely based on clinician’s judgment and are time-consuming and subjective. Hence, an accurate, quantitative and objective method of evaluating FNP is proposed for constructing a standard system, which will be an invaluable tool for clinicians who treat the patient with FNP. In this paper, we introduce a novel method for quantitative assessment of FNP which combines an effective facial landmark estimation (FLE) algorithm and facial asymmetrical feature (FAF) by processing facial movement image. The facial landmarks can be detected automatically and accurately using FLE. The FAF is based on the angle of key facial landmark connection and mirror degree of multiple regions on human face. Our method provides significant contribution as it describes the displacement of facial organ and the changes of facial organ exposure during performing facial movements. Experiments show that our method is effective, accurate and convenient in practice, which is beneficial to FNP diagnosis and personalized rehabilitation therapy for each patient.