Pub Date : 2022-11-05DOI: 10.1109/CISP-BMEI56279.2022.9980116
Jialiang Luo, Wanbo Yu
There is a demand for more image encryption techniques due to the safety of image transmission and archiving. so a means of encrypting images based on random number matrix iterations is suggested. In the encryption process, the algorithm uses two random number matrices with values ranging from 1 to 256, which are randomly generated. Use two matrices to iterate. During the iteration, take out the elements of the first matrix, then take out the elements in the same position of the second matrix, combining the element values of the two to form a new position index, and get the value of the new position index in the first matrix finally. Iterate over the value multiple times, using it to replace the element value of the current first matrix, and then continue to iterate the other position elements of the matrix. After the iteration of all elements is completed, the first random number matrix will be converted into a new matrix, and it will be used as an encrypted sequence. A plaintext image is XOR with the sequence to generate a ciphertext image. The testing findings demonstrate the algorithm's superior security and encryption performance. It features a big key spacing, strong key sensitivity, and good diffusion and obfuscation capabilities, and are resistant to conventional assaults including differential and brute force assaults.
{"title":"An image encryption method based on random number matrix iterations","authors":"Jialiang Luo, Wanbo Yu","doi":"10.1109/CISP-BMEI56279.2022.9980116","DOIUrl":"https://doi.org/10.1109/CISP-BMEI56279.2022.9980116","url":null,"abstract":"There is a demand for more image encryption techniques due to the safety of image transmission and archiving. so a means of encrypting images based on random number matrix iterations is suggested. In the encryption process, the algorithm uses two random number matrices with values ranging from 1 to 256, which are randomly generated. Use two matrices to iterate. During the iteration, take out the elements of the first matrix, then take out the elements in the same position of the second matrix, combining the element values of the two to form a new position index, and get the value of the new position index in the first matrix finally. Iterate over the value multiple times, using it to replace the element value of the current first matrix, and then continue to iterate the other position elements of the matrix. After the iteration of all elements is completed, the first random number matrix will be converted into a new matrix, and it will be used as an encrypted sequence. A plaintext image is XOR with the sequence to generate a ciphertext image. The testing findings demonstrate the algorithm's superior security and encryption performance. It features a big key spacing, strong key sensitivity, and good diffusion and obfuscation capabilities, and are resistant to conventional assaults including differential and brute force assaults.","PeriodicalId":198522,"journal":{"name":"2022 15th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI)","volume":"74 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128708159","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-11-05DOI: 10.1109/CISP-BMEI56279.2022.9979902
A. Yu, Longyun Chen, C. Qiao
At present, deep learning has been widely used in the research of brain structure, brain connectivity, brain diseases and other related fields. In particular, research on attention deficit hyperactivity disorder (ADHD) has been applied to assist in diagnosis, follow-up treatment, etc. However, there is a lack of explainable studies on abnormal functional connectivity in ADHD. In addition, the small amount of information available on ADHD lead to poor recognition accuracy and performance of deep learning. Therefore, we propose an explainable Graph convolutional networks (GCN) with attentional mechanisms to improve diagnostic accuracy and find abnormal neural markers of ADHD. We experiment with the method on fMRI clinical dataset of Connectomics in Neuroimaging Transfer Learning Challenge (CNI-TLC). The experimental results validate the reliability of the model, and find the abnormal regions and connections in ADHD patients. These abnormal regions and connections are mainly concentrated in cognitive and emotion-related regions such as frontal, parietal and temporal lobes.
{"title":"Graph Convolutional Network with Attention Mechanism for Discovering the Brain's Abnormal Activity of Attention Deficit Hyperactivity Disorder","authors":"A. Yu, Longyun Chen, C. Qiao","doi":"10.1109/CISP-BMEI56279.2022.9979902","DOIUrl":"https://doi.org/10.1109/CISP-BMEI56279.2022.9979902","url":null,"abstract":"At present, deep learning has been widely used in the research of brain structure, brain connectivity, brain diseases and other related fields. In particular, research on attention deficit hyperactivity disorder (ADHD) has been applied to assist in diagnosis, follow-up treatment, etc. However, there is a lack of explainable studies on abnormal functional connectivity in ADHD. In addition, the small amount of information available on ADHD lead to poor recognition accuracy and performance of deep learning. Therefore, we propose an explainable Graph convolutional networks (GCN) with attentional mechanisms to improve diagnostic accuracy and find abnormal neural markers of ADHD. We experiment with the method on fMRI clinical dataset of Connectomics in Neuroimaging Transfer Learning Challenge (CNI-TLC). The experimental results validate the reliability of the model, and find the abnormal regions and connections in ADHD patients. These abnormal regions and connections are mainly concentrated in cognitive and emotion-related regions such as frontal, parietal and temporal lobes.","PeriodicalId":198522,"journal":{"name":"2022 15th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI)","volume":"5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131139249","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-11-05DOI: 10.1109/CISP-BMEI56279.2022.9980125
Qiming Ma, Yu Zhu
Current 3D object detection frameworks based on LiDAR mainly used sparse convolution as the backbone network after voxelization to extract features, and applied grids to further refine the proposal boxes. While these operations may limit the accuracy improvement of 3D object detection, because they destroyed the geometric characteristics of point clouds to a large extent, including density and object shape. Therefore, in this paper, we proposed a method to refine the proposals by estimating density-aware information in the second stage. A certain number of key points were sampled in each proposal, and then applied the self-attention module to study the relations between these key points. Then the designed spatial-channel-wise decoder fused channel-wise and spatial-wise features to obtain the global representation of the object. Finally, the global representation was fed into the detect head to obtain a more accurate box. The performance of our proposed 3D detection model was evaluated on the KITTI dataset, and the average accuracy of car class on the test set and validation split was 80.62% and 85.39% respectively, and the average accuracy of three classes in KITTI on the validation split was 72.41%.
{"title":"Keypoints Representation of Density-aware and the Spatial-Channel-wise Decoder for 3D Object Detection","authors":"Qiming Ma, Yu Zhu","doi":"10.1109/CISP-BMEI56279.2022.9980125","DOIUrl":"https://doi.org/10.1109/CISP-BMEI56279.2022.9980125","url":null,"abstract":"Current 3D object detection frameworks based on LiDAR mainly used sparse convolution as the backbone network after voxelization to extract features, and applied grids to further refine the proposal boxes. While these operations may limit the accuracy improvement of 3D object detection, because they destroyed the geometric characteristics of point clouds to a large extent, including density and object shape. Therefore, in this paper, we proposed a method to refine the proposals by estimating density-aware information in the second stage. A certain number of key points were sampled in each proposal, and then applied the self-attention module to study the relations between these key points. Then the designed spatial-channel-wise decoder fused channel-wise and spatial-wise features to obtain the global representation of the object. Finally, the global representation was fed into the detect head to obtain a more accurate box. The performance of our proposed 3D detection model was evaluated on the KITTI dataset, and the average accuracy of car class on the test set and validation split was 80.62% and 85.39% respectively, and the average accuracy of three classes in KITTI on the validation split was 72.41%.","PeriodicalId":198522,"journal":{"name":"2022 15th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI)","volume":"378 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117169993","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-11-05DOI: 10.1109/CISP-BMEI56279.2022.9979883
Jiaping Qin, Jing-yu Gong, Zhengyang Feng, Xin Tan, Lizhuang Ma
Geometry plays a vital role in 3D point cloud semantic segmentation since each category of object exhibits a specific geometric pattern. However, popular point cloud semantic segmentation methods ignore this property during feature aggregation. In this paper, we propose a novel Covariance-based Geometry Encoder (CGE) to learn latent geometry representation in point clouds and break this limitation. Specifically, we find that the classic covariance matrix can represent geometry implicitly in a point neighborhood, and we can learn geometry representation through simple multi-layer perceptrons to enhance the point features in a deep network. The proposed CGE module is generally applicable to any point-based network, while only requiring a little extra computing. Through extensive experiments, our method shows competitive performance on both indoor and outdoor benchmark datasets. Code will be publicly available.
{"title":"Understanding Geometry for Point Cloud Segmentation via Covariance","authors":"Jiaping Qin, Jing-yu Gong, Zhengyang Feng, Xin Tan, Lizhuang Ma","doi":"10.1109/CISP-BMEI56279.2022.9979883","DOIUrl":"https://doi.org/10.1109/CISP-BMEI56279.2022.9979883","url":null,"abstract":"Geometry plays a vital role in 3D point cloud semantic segmentation since each category of object exhibits a specific geometric pattern. However, popular point cloud semantic segmentation methods ignore this property during feature aggregation. In this paper, we propose a novel Covariance-based Geometry Encoder (CGE) to learn latent geometry representation in point clouds and break this limitation. Specifically, we find that the classic covariance matrix can represent geometry implicitly in a point neighborhood, and we can learn geometry representation through simple multi-layer perceptrons to enhance the point features in a deep network. The proposed CGE module is generally applicable to any point-based network, while only requiring a little extra computing. Through extensive experiments, our method shows competitive performance on both indoor and outdoor benchmark datasets. Code will be publicly available.","PeriodicalId":198522,"journal":{"name":"2022 15th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123487429","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-11-05DOI: 10.1109/CISP-BMEI56279.2022.9979906
Kunkun Xiong, Wensheng Li, Shuai Dong, Yuanlie He, Zhihua Yang
Biscuit packaging must be thoroughly inspected after being sealed. However, biscuit packaging defect detection remains a challenging task due to both subtle and complex textures at the seal and the lack of defective samples. To overcome these difficulties, we propose a multi-task defect detection framework based on level set map. First, level set map (LSM), a new segmentation task label is proposed, which can represent both contour information and defect location information by using image gray value. Then, a multi-task framework based on LSM is designed, the main task of which is the binary classification task to predict the defect state of the packaging, and the auxiliary task is the semantic segmentation task of extracting biscuit packaging contours and locating defects. The two tasks share the feature extractor, and the auxiliary task provides a supervised spatial attention to guide the feature extractor to focus on the contour of the packaging. To verify the performance of the multi-task framework, two real datasets under different acquisition environments are established. The experimental results show that, compared with other classification networks and object detection framework, the multi-task framework based on LSM can significantly improve the accuracy of the biscuit packaging defect detection task.
{"title":"Defect detection of biscuit packaging based on level set map","authors":"Kunkun Xiong, Wensheng Li, Shuai Dong, Yuanlie He, Zhihua Yang","doi":"10.1109/CISP-BMEI56279.2022.9979906","DOIUrl":"https://doi.org/10.1109/CISP-BMEI56279.2022.9979906","url":null,"abstract":"Biscuit packaging must be thoroughly inspected after being sealed. However, biscuit packaging defect detection remains a challenging task due to both subtle and complex textures at the seal and the lack of defective samples. To overcome these difficulties, we propose a multi-task defect detection framework based on level set map. First, level set map (LSM), a new segmentation task label is proposed, which can represent both contour information and defect location information by using image gray value. Then, a multi-task framework based on LSM is designed, the main task of which is the binary classification task to predict the defect state of the packaging, and the auxiliary task is the semantic segmentation task of extracting biscuit packaging contours and locating defects. The two tasks share the feature extractor, and the auxiliary task provides a supervised spatial attention to guide the feature extractor to focus on the contour of the packaging. To verify the performance of the multi-task framework, two real datasets under different acquisition environments are established. The experimental results show that, compared with other classification networks and object detection framework, the multi-task framework based on LSM can significantly improve the accuracy of the biscuit packaging defect detection task.","PeriodicalId":198522,"journal":{"name":"2022 15th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI)","volume":"15 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115331620","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In the cross-subject classification task, a subject-agnostic model is trained for the classification task of other subjects, according to the prior knowledge from EEG data of some subjects. It is one of the challenges for ERP classification in the RSVP-based BCI system. So far, convolutional neural networks (CNNs) for RSVP classification only use a fixed-size kernel for each layer to extract features in the temporal domain, which limits the ability of the network to detect ERP. In this work, a multi-scale EEGNet model (MS-EEGNet) for cross-subject RSVP classification task was proposed, which adopted parallel convolution layers with multi-scale kernels to extract discrimination information in the temporal domain, and increased the robustness of the model. The proposed model was used for the BCI Controlled Robot Contest in the World Robot Contest 2022 and achieved good results. The UAR of the A and B datasets got 0.493 and 0.528, respectively. Compared with other CNN algorithms including EEGNet and PLNet, the proposed model had better classification performance.
{"title":"A multi-scale EEGNet for cross-subject RSVP-based BCI system","authors":"Xuepu Wang, Yanfei Lin, Ying Tan, Rongxiao Guo, Xiaorong Gao","doi":"10.1109/CISP-BMEI56279.2022.9980258","DOIUrl":"https://doi.org/10.1109/CISP-BMEI56279.2022.9980258","url":null,"abstract":"In the cross-subject classification task, a subject-agnostic model is trained for the classification task of other subjects, according to the prior knowledge from EEG data of some subjects. It is one of the challenges for ERP classification in the RSVP-based BCI system. So far, convolutional neural networks (CNNs) for RSVP classification only use a fixed-size kernel for each layer to extract features in the temporal domain, which limits the ability of the network to detect ERP. In this work, a multi-scale EEGNet model (MS-EEGNet) for cross-subject RSVP classification task was proposed, which adopted parallel convolution layers with multi-scale kernels to extract discrimination information in the temporal domain, and increased the robustness of the model. The proposed model was used for the BCI Controlled Robot Contest in the World Robot Contest 2022 and achieved good results. The UAR of the A and B datasets got 0.493 and 0.528, respectively. Compared with other CNN algorithms including EEGNet and PLNet, the proposed model had better classification performance.","PeriodicalId":198522,"journal":{"name":"2022 15th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI)","volume":"29 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122651020","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-11-05DOI: 10.1109/CISP-BMEI56279.2022.9980081
Ruoyu Du, Jingjie Huang, Shujin Zhu
To solve the problem that traditional epileptic seizure detection methods are cumbersome and prone to human errors, a hybrid model combining optimized feature convolutional neural network (CNN) model and traditional machine learning model is proposed, and its performance is verified on two small sample epileptic EEG datasets of Bonn and Hauz Khas. The model is based on the optimized feature CNN model for feature extraction, and the support vector machine (SVM) and random forest (RF) classifiers are selected to detect and recognize the Epileptic Electroencephalogram (EEG) seizure and normal state. The results showed that the optimized feature CNN-SVM model performs well in the binary classification tasks of epileptic EEG detection, with the highest accuracy of 99.57% and 98.00%. Compared with the traditional SVM and RF model, the classification performance is better, which can be improved by 3.92 %. The results indicated that the advantages of the deep learning algorithm in automatic feature extraction could improve the classification performance of the traditional machine learning model, and the traditional machine learning model is more suitable for small sample binary classification detection tasks than the deep learning model. It provides a scientific reference for the research of machine learning models and the clinical diagnosis of epilepsy.
{"title":"EEG-Based Epileptic Seizure Detection Model Using CNN Feature Optimization","authors":"Ruoyu Du, Jingjie Huang, Shujin Zhu","doi":"10.1109/CISP-BMEI56279.2022.9980081","DOIUrl":"https://doi.org/10.1109/CISP-BMEI56279.2022.9980081","url":null,"abstract":"To solve the problem that traditional epileptic seizure detection methods are cumbersome and prone to human errors, a hybrid model combining optimized feature convolutional neural network (CNN) model and traditional machine learning model is proposed, and its performance is verified on two small sample epileptic EEG datasets of Bonn and Hauz Khas. The model is based on the optimized feature CNN model for feature extraction, and the support vector machine (SVM) and random forest (RF) classifiers are selected to detect and recognize the Epileptic Electroencephalogram (EEG) seizure and normal state. The results showed that the optimized feature CNN-SVM model performs well in the binary classification tasks of epileptic EEG detection, with the highest accuracy of 99.57% and 98.00%. Compared with the traditional SVM and RF model, the classification performance is better, which can be improved by 3.92 %. The results indicated that the advantages of the deep learning algorithm in automatic feature extraction could improve the classification performance of the traditional machine learning model, and the traditional machine learning model is more suitable for small sample binary classification detection tasks than the deep learning model. It provides a scientific reference for the research of machine learning models and the clinical diagnosis of epilepsy.","PeriodicalId":198522,"journal":{"name":"2022 15th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI)","volume":"61 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125069639","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-11-05DOI: 10.1109/CISP-BMEI56279.2022.9979914
Qing Ding
In recent years, the application of LED is more and more extensive, and the constant-current accuracy requirement of the flyback LED driver controlled by the original side is also increasing. In this paper, the constant current error of the flyback LED driver controlled by the original side is analyzed, and on this basis, the two errors are compensated one by one. An adaptive Toff time compensation scheme is proposed, which can detect the Toff time error in different states in real time, so as to compensate the error well and achieve high constant current accuracy. In addition, a variable with Vin is introduced to compensate the sampling error of the original peak current by adjusting the external resistance. The experimental results show that through the above compensation action, the constant current accuracy can be controlled within 0.3 %, so as to meet the engineering requirements for constant current accuracy.
{"title":"Constant Current Precision of Flyback LED Driver with Primary Side Control","authors":"Qing Ding","doi":"10.1109/CISP-BMEI56279.2022.9979914","DOIUrl":"https://doi.org/10.1109/CISP-BMEI56279.2022.9979914","url":null,"abstract":"In recent years, the application of LED is more and more extensive, and the constant-current accuracy requirement of the flyback LED driver controlled by the original side is also increasing. In this paper, the constant current error of the flyback LED driver controlled by the original side is analyzed, and on this basis, the two errors are compensated one by one. An adaptive Toff time compensation scheme is proposed, which can detect the Toff time error in different states in real time, so as to compensate the error well and achieve high constant current accuracy. In addition, a variable with Vin is introduced to compensate the sampling error of the original peak current by adjusting the external resistance. The experimental results show that through the above compensation action, the constant current accuracy can be controlled within 0.3 %, so as to meet the engineering requirements for constant current accuracy.","PeriodicalId":198522,"journal":{"name":"2022 15th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI)","volume":"01 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129908025","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-11-05DOI: 10.1109/CISP-BMEI56279.2022.9980137
Cheng Fei, Jianxu Luo
Accurate pancreas segmentation is of great significance for the diagnosis and treatment of pancreatic cancer. Exploiting FCN, UNet or their variants, these CNN-based methods, to complete this task has become the de-facto standard and great success has been achieved. However, convolutional operation fails to build long-range dependency which is crucial for segmentation, hindering the further development of CNN-based methods. To address the difficulty, we propose the DTUnet network, which introduces the DenseASPP module and Transformer on the basis of UNet and stacks the two in a sequential manner. Transformer connects each pixel of the input feature maps to generate a global receptive field, thus capturing the global context information and realizing the construction of long-range dependency. Meanwhile, to alleviate the training challenges of Transformer's data-hungry, DTUnet employs DenseASPP module to generate rich and multi-scale high-level semantic feature maps as the input of Transformer, ensuring that Transformer can fully leverage global modeling capability even when applied to a small size pancreas segmentation dataset. Benefiting from the combination of the two, the proposed DTUnet generates more efficient and reliable global context information, and ultimately achieves an average Dice coefficient score of 84.77% ±4.65 on the public NIH pancreas segmentation dataset, which is 1.87% higher than UNet. The result is also higher than advanced methods in recent years, indicating that DTU net has the potential to assist doctors to segment the pancreas in clinical application.
{"title":"DTUnet: A Transformer-based UNet Combined with DenseASPP Module for Pancreas Segmentation","authors":"Cheng Fei, Jianxu Luo","doi":"10.1109/CISP-BMEI56279.2022.9980137","DOIUrl":"https://doi.org/10.1109/CISP-BMEI56279.2022.9980137","url":null,"abstract":"Accurate pancreas segmentation is of great significance for the diagnosis and treatment of pancreatic cancer. Exploiting FCN, UNet or their variants, these CNN-based methods, to complete this task has become the de-facto standard and great success has been achieved. However, convolutional operation fails to build long-range dependency which is crucial for segmentation, hindering the further development of CNN-based methods. To address the difficulty, we propose the DTUnet network, which introduces the DenseASPP module and Transformer on the basis of UNet and stacks the two in a sequential manner. Transformer connects each pixel of the input feature maps to generate a global receptive field, thus capturing the global context information and realizing the construction of long-range dependency. Meanwhile, to alleviate the training challenges of Transformer's data-hungry, DTUnet employs DenseASPP module to generate rich and multi-scale high-level semantic feature maps as the input of Transformer, ensuring that Transformer can fully leverage global modeling capability even when applied to a small size pancreas segmentation dataset. Benefiting from the combination of the two, the proposed DTUnet generates more efficient and reliable global context information, and ultimately achieves an average Dice coefficient score of 84.77% ±4.65 on the public NIH pancreas segmentation dataset, which is 1.87% higher than UNet. The result is also higher than advanced methods in recent years, indicating that DTU net has the potential to assist doctors to segment the pancreas in clinical application.","PeriodicalId":198522,"journal":{"name":"2022 15th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI)","volume":"16 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125575410","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-11-05DOI: 10.1109/CISP-BMEI56279.2022.9980303
Xinru Li, Qingyu Li, Fanke Meng, Zixin Xu, Zhan Xu, Yi Gong
Non-linear MIMO technology is proven to work out the excessive power consumption issue caused by the base station with more than 100 antenna pairs that have been adopted for the Internet of Vehicles (IoV). However, the non-linear MIMO scheme applied in the IoV scenario does not consider the real-world channel with the characteristics of vehicle motion. In addition, traditional channel estimation (CE) in non-linear MIMO technology are not robust under the variation of channel parameters in IoV.A channel estimation scheme of Half Phase Only (HPO)-MIMO based on Convolutional Neural Network (CNN) is proposed, which can get more accurate channel estimation results and achieve perfect robustness for changing channel parameters. Besides, the COST 2100 channel model is used, which is more suitable for simulating the IoV scenarios. Moreover, the channel estimation scheme based on CNN al-gorithm can be used favorably in the non-linear MIMO and IoV scenarios. Simulation results show that the CNN-based CE scheme we proposed achieves outstanding mean squared error (MSE) performance compared to the Generalized Approximate Messaging (GAMP) algorithm. Furthermore, the rationality of using the COST 2100 channel model is proven that have excellent performances.
{"title":"Deep Learning-Based Channel Estimation for HPO-MIMO Systems in IoV Scenario","authors":"Xinru Li, Qingyu Li, Fanke Meng, Zixin Xu, Zhan Xu, Yi Gong","doi":"10.1109/CISP-BMEI56279.2022.9980303","DOIUrl":"https://doi.org/10.1109/CISP-BMEI56279.2022.9980303","url":null,"abstract":"Non-linear MIMO technology is proven to work out the excessive power consumption issue caused by the base station with more than 100 antenna pairs that have been adopted for the Internet of Vehicles (IoV). However, the non-linear MIMO scheme applied in the IoV scenario does not consider the real-world channel with the characteristics of vehicle motion. In addition, traditional channel estimation (CE) in non-linear MIMO technology are not robust under the variation of channel parameters in IoV.A channel estimation scheme of Half Phase Only (HPO)-MIMO based on Convolutional Neural Network (CNN) is proposed, which can get more accurate channel estimation results and achieve perfect robustness for changing channel parameters. Besides, the COST 2100 channel model is used, which is more suitable for simulating the IoV scenarios. Moreover, the channel estimation scheme based on CNN al-gorithm can be used favorably in the non-linear MIMO and IoV scenarios. Simulation results show that the CNN-based CE scheme we proposed achieves outstanding mean squared error (MSE) performance compared to the Generalized Approximate Messaging (GAMP) algorithm. Furthermore, the rationality of using the COST 2100 channel model is proven that have excellent performances.","PeriodicalId":198522,"journal":{"name":"2022 15th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI)","volume":"167 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127579794","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：