Chengyuan Zhuang, Xiaohui Yuan, Xuan Guo, Zhenchun Wei, Juan Xu, Yuqi Fan
Deep convolutional neural networks are powerful and popular tools as deep learning emerges in recent years for image classification in computer vision. However, it is difficult to learn convolutional filters from the examples. The innate frequency property of the data has not been well considered. To address this problem, we find high-frequency information import within deep networks and therefore propose our high-pass attention method (HPA) to help the learning process. HPA explicitly generates high-frequency information via a stage-wise high-pass filter to alleviate the burden of learning such information. Strengthened by channel attention on the concatenated features, our method demonstrates consistent improvements upon ResNet-18/ResNet-50 by 1.36%/1.60% and 1.47%/1.39% on the ImageNet-1K dataset and the Food-101 dataset, respectively, as well as the effectiveness over a variety of modules.
{"title":"Improved Convolutional Neural Networks by Integrating High-frequency Information for Image Classification","authors":"Chengyuan Zhuang, Xiaohui Yuan, Xuan Guo, Zhenchun Wei, Juan Xu, Yuqi Fan","doi":"10.1145/3590003.3590082","DOIUrl":"https://doi.org/10.1145/3590003.3590082","url":null,"abstract":"Deep convolutional neural networks are powerful and popular tools as deep learning emerges in recent years for image classification in computer vision. However, it is difficult to learn convolutional filters from the examples. The innate frequency property of the data has not been well considered. To address this problem, we find high-frequency information import within deep networks and therefore propose our high-pass attention method (HPA) to help the learning process. HPA explicitly generates high-frequency information via a stage-wise high-pass filter to alleviate the burden of learning such information. Strengthened by channel attention on the concatenated features, our method demonstrates consistent improvements upon ResNet-18/ResNet-50 by 1.36%/1.60% and 1.47%/1.39% on the ImageNet-1K dataset and the Food-101 dataset, respectively, as well as the effectiveness over a variety of modules.","PeriodicalId":340225,"journal":{"name":"Proceedings of the 2023 2nd Asia Conference on Algorithms, Computing and Machine Learning","volume":"67 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127525841","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}
Kaixuan Sun, Ketong Wu, Wenke Yuan, Guangyuan Wei, Huasen He
In the next generation network, both the satellite network layer and aeronautical network layer will play significant roles, leading the world into the era of global interconnectivity. However, the large-scale and high-mobility characteristics of aircraft networks greatly challenge the application of traditional routing algorithms. Therefore, this paper aims to solve this challenge by exploiting a Software Defined Satellite Network (Sat-SDN) to facilitate the routing in aeronautical networks. By centrally controlling aeronautical routing through satellites, the computation and communication overhead for aeronautical networks are relieved, since frequent packet flooding and broadcasting for synchronizing the rapidly-fluctuating topology of aeronautical networks can be avoided. To extend the aeronautical networking and transmission mechanism to a global scale, a multi-domain extension mechanism is proposed, while the concept of dynamic inter-domain telescope nodes is induced to greatly simplify the network topology. A Sat-SDN aided Genetic-based Aeronautical Routing (SSGAR) algorithm is further designed to solve the problem of huge routing calculation space and long convergence time in large-scale multi-node network scenarios. Moreover, experiments and simulations are conducted using real aircraft data, which demonstrate that our proposed SSGAR algorithm can effectively reduce communication costs and improve transmission quality compared to existing solutions.
{"title":"SSGAR: A Genetic-based Routing Solution for Aeronautical Networks aided by Software Defined Satellite Network","authors":"Kaixuan Sun, Ketong Wu, Wenke Yuan, Guangyuan Wei, Huasen He","doi":"10.1145/3590003.3590069","DOIUrl":"https://doi.org/10.1145/3590003.3590069","url":null,"abstract":"In the next generation network, both the satellite network layer and aeronautical network layer will play significant roles, leading the world into the era of global interconnectivity. However, the large-scale and high-mobility characteristics of aircraft networks greatly challenge the application of traditional routing algorithms. Therefore, this paper aims to solve this challenge by exploiting a Software Defined Satellite Network (Sat-SDN) to facilitate the routing in aeronautical networks. By centrally controlling aeronautical routing through satellites, the computation and communication overhead for aeronautical networks are relieved, since frequent packet flooding and broadcasting for synchronizing the rapidly-fluctuating topology of aeronautical networks can be avoided. To extend the aeronautical networking and transmission mechanism to a global scale, a multi-domain extension mechanism is proposed, while the concept of dynamic inter-domain telescope nodes is induced to greatly simplify the network topology. A Sat-SDN aided Genetic-based Aeronautical Routing (SSGAR) algorithm is further designed to solve the problem of huge routing calculation space and long convergence time in large-scale multi-node network scenarios. Moreover, experiments and simulations are conducted using real aircraft data, which demonstrate that our proposed SSGAR algorithm can effectively reduce communication costs and improve transmission quality compared to existing solutions.","PeriodicalId":340225,"journal":{"name":"Proceedings of the 2023 2nd Asia Conference on Algorithms, Computing and Machine Learning","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125920727","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}
Detecting respiratory events in sleep requires much attention and is labor consuming conventionally. With the development of technology, some kinds of software that can automatically detect the respiratory events was designed to help simplify and improve this process. However, in order to ensure its accuracy of the detection, it is necessary to provide appropriate key parameters before using it. After that the interval adjustment also needs to be done manually, which still takes a lot of time and means high demands on the technicians. In this paper, an end-to-end ConvNet was used to detect the respiratory events which does not need to provide any extra parameters. Its performance was further compared with widely used events detection software, Philips Sleepware G3 with Smonolyzer. The results show that ConvNet has higher accuracy than G3 with Smonolyzer in event detection. Such a ConvNet-based analysis system is sufficiently accurate for event detection according to the AASM classification criteria.
在睡眠中检测呼吸事件需要大量的注意力，并且是传统的劳动消耗。随着技术的发展，一些能够自动检测呼吸事件的软件被设计出来，以帮助简化和改进这一过程。但是，为了保证其检测的准确性，在使用前必须提供合适的关键参数。之后的间隔调整还需要手工完成，这仍然需要花费大量的时间，对技术人员的要求也很高。本文采用端到端卷积神经网络来检测呼吸事件，不需要提供任何额外的参数。进一步将其性能与广泛使用的事件检测软件Philips Sleepware G3 with Smonolyzer进行比较。结果表明，卷积神经网络在事件检测方面的准确率高于使用Smonolyzer的G3。根据AASM分类标准，这种基于convnet的分析系统对于事件检测具有足够的准确性。
{"title":"Detecting Respiratory Events with End-to-End ConvNet","authors":"Yanping Shuai, Zhangbo Li, Xingjun Wang, Hanrong Cheng","doi":"10.1145/3590003.3590098","DOIUrl":"https://doi.org/10.1145/3590003.3590098","url":null,"abstract":"Detecting respiratory events in sleep requires much attention and is labor consuming conventionally. With the development of technology, some kinds of software that can automatically detect the respiratory events was designed to help simplify and improve this process. However, in order to ensure its accuracy of the detection, it is necessary to provide appropriate key parameters before using it. After that the interval adjustment also needs to be done manually, which still takes a lot of time and means high demands on the technicians. In this paper, an end-to-end ConvNet was used to detect the respiratory events which does not need to provide any extra parameters. Its performance was further compared with widely used events detection software, Philips Sleepware G3 with Smonolyzer. The results show that ConvNet has higher accuracy than G3 with Smonolyzer in event detection. Such a ConvNet-based analysis system is sufficiently accurate for event detection according to the AASM classification criteria.","PeriodicalId":340225,"journal":{"name":"Proceedings of the 2023 2nd Asia Conference on Algorithms, Computing and Machine Learning","volume":"43 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117000629","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}
Graph contrastive learning has emerged as a promising method for self-supervised graph representation learning. The traditional framework conventionally imposes two graph views generated by leveraging graph data augmentations. Such an approach focuses on leading the model to learn discriminative information from graph local structures, which brings up an intrinsic issue that the model partially fails to obtain sufficient discriminative information contained by the graph global information. To this end, we propose a hypergraph-augmented view to empower the self-supervised graph representation learning model to better capture the global information from nodes and corresponding edges. In the further exploration of the graph contrastive learning, we discover a principal challenge undermining conventional contrastive methods: the false negative sample problem, i.e., specific negative samples actually belong to the same category of the anchor sample. To address this issue, we take the neighbors of nodes into consideration and propose the robust graph contrastive learning. In practice, we empirically observe that the proposed hypergraph-augmented view can further enhance the robustness of graph contrastive learning by adopting our framework. Based on these improvements, we propose a novel method called Robust Hypergraph-Augmented Graph Contrastive Learning (RH-GCL). We conduct various experiments in the settings of both transductive and inductive node classification. The results demonstrate that our method achieves the state-of-the-art (SOTA) performance on different datasets. Specifically, the accuracy of node classification on Cora dataset is 84.4%, which is 1.1% higher than that of GRACE. We also perform the ablation study to verify the effectiveness of each part of our proposed method.
{"title":"Robust Hypergraph-Augmented Graph Contrastive Learning for Graph Self-Supervised Learning","authors":"Zeming Wang, Xiaoyang Li, Rui Wang, Changwen Zheng","doi":"10.1145/3590003.3590053","DOIUrl":"https://doi.org/10.1145/3590003.3590053","url":null,"abstract":"Graph contrastive learning has emerged as a promising method for self-supervised graph representation learning. The traditional framework conventionally imposes two graph views generated by leveraging graph data augmentations. Such an approach focuses on leading the model to learn discriminative information from graph local structures, which brings up an intrinsic issue that the model partially fails to obtain sufficient discriminative information contained by the graph global information. To this end, we propose a hypergraph-augmented view to empower the self-supervised graph representation learning model to better capture the global information from nodes and corresponding edges. In the further exploration of the graph contrastive learning, we discover a principal challenge undermining conventional contrastive methods: the false negative sample problem, i.e., specific negative samples actually belong to the same category of the anchor sample. To address this issue, we take the neighbors of nodes into consideration and propose the robust graph contrastive learning. In practice, we empirically observe that the proposed hypergraph-augmented view can further enhance the robustness of graph contrastive learning by adopting our framework. Based on these improvements, we propose a novel method called Robust Hypergraph-Augmented Graph Contrastive Learning (RH-GCL). We conduct various experiments in the settings of both transductive and inductive node classification. The results demonstrate that our method achieves the state-of-the-art (SOTA) performance on different datasets. Specifically, the accuracy of node classification on Cora dataset is 84.4%, which is 1.1% higher than that of GRACE. We also perform the ablation study to verify the effectiveness of each part of our proposed method.","PeriodicalId":340225,"journal":{"name":"Proceedings of the 2023 2nd Asia Conference on Algorithms, Computing and Machine Learning","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129792753","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}
This paper proposes a modified YOLOv5 UAV target detection algorithm for the low detection accuracy caused by the dense target distribution and too small size in the UAV image. Firstly, the coordinate attention mechanism (Coordinate Attention, CA) is introduced in the backbone network CSPDarknet53 to enhance the feature extraction capability of the network; secondly, the multi-size feature pyramid network is designed to introduce a larger resolution feature map for feature fusion and prediction, and to improve the accuracy of small target detection. Experiments on the VisDrone2021 dataset, the results show that the average detection accuracy (Mean Average Precision, mAP) of the improved YOLOv5 algorithm reached 43.0%, 5.8 percentage points higher than the original algorithm, which fully proves the high efficiency of the proposed improved algorithm on the ground target detection of the UAV.
针对无人机图像中目标分布密集、尺寸过小导致检测精度不高的问题，提出了一种改进的YOLOv5无人机目标检测算法。首先，在骨干网络CSPDarknet53中引入坐标注意机制(coordinate attention, CA)，增强网络的特征提取能力;其次，设计多尺度特征金字塔网络，引入更大分辨率的特征图进行特征融合和预测，提高小目标检测的精度;在VisDrone2021数据集上的实验结果表明，改进的YOLOv5算法的平均检测精度(Mean average Precision, mAP)达到43.0%，比原算法提高5.8个百分点，充分证明了改进算法在无人机地面目标检测上的高效率。
{"title":"Improved YOLOv5 UAV Target Detection Algorithm by Fused Attention Mechanism","authors":"Yan He, Yanni Zhao, Hongfei Nie","doi":"10.1145/3590003.3590074","DOIUrl":"https://doi.org/10.1145/3590003.3590074","url":null,"abstract":"This paper proposes a modified YOLOv5 UAV target detection algorithm for the low detection accuracy caused by the dense target distribution and too small size in the UAV image. Firstly, the coordinate attention mechanism (Coordinate Attention, CA) is introduced in the backbone network CSPDarknet53 to enhance the feature extraction capability of the network; secondly, the multi-size feature pyramid network is designed to introduce a larger resolution feature map for feature fusion and prediction, and to improve the accuracy of small target detection. Experiments on the VisDrone2021 dataset, the results show that the average detection accuracy (Mean Average Precision, mAP) of the improved YOLOv5 algorithm reached 43.0%, 5.8 percentage points higher than the original algorithm, which fully proves the high efficiency of the proposed improved algorithm on the ground target detection of the UAV.","PeriodicalId":340225,"journal":{"name":"Proceedings of the 2023 2nd Asia Conference on Algorithms, Computing and Machine Learning","volume":"539 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121496746","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}
Scene-based product design is an effective way to improve user experience, and static scene library is an important basis for commercial vehicle product planners to carry out their design and planning. This paper explores the characteristics of commercial vehicle's dynamic scene and static scene, as well as the current situation and problems of commercial vehicle's static scene library. Furthermore, using the research methods of combinatorial design and task analysis, a digital solution for the construction of commercial vehicle's static scene library is proposed in this paper with reference to the construction process of vehicle's dynamic scene library. This paper innovatively proposes the combination of scene library and questionnaire system to solve the problem of scene information collection and scene creation scale by means of structured questionnaire, which provides a reference for the construction of commercial vehicle's static scene library.This research helps automobile enterprises systematically manage the survey results, expand the application scope of the survey data, extend the application time of the survey results, and reduce the enterprise research costs.
{"title":"Construction of Scene Library System for Commercial Vehicle Products Based on Multidimensional Terminal","authors":"Yunshuang Zheng, Shilan Hu, Nannan Xue","doi":"10.1145/3590003.3590025","DOIUrl":"https://doi.org/10.1145/3590003.3590025","url":null,"abstract":"Scene-based product design is an effective way to improve user experience, and static scene library is an important basis for commercial vehicle product planners to carry out their design and planning. This paper explores the characteristics of commercial vehicle's dynamic scene and static scene, as well as the current situation and problems of commercial vehicle's static scene library. Furthermore, using the research methods of combinatorial design and task analysis, a digital solution for the construction of commercial vehicle's static scene library is proposed in this paper with reference to the construction process of vehicle's dynamic scene library. This paper innovatively proposes the combination of scene library and questionnaire system to solve the problem of scene information collection and scene creation scale by means of structured questionnaire, which provides a reference for the construction of commercial vehicle's static scene library.This research helps automobile enterprises systematically manage the survey results, expand the application scope of the survey data, extend the application time of the survey results, and reduce the enterprise research costs.","PeriodicalId":340225,"journal":{"name":"Proceedings of the 2023 2nd Asia Conference on Algorithms, Computing and Machine Learning","volume":"52 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115947625","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}
Abstract— For areas with potential occurrence of blasting events, it is essential to distinguish them from natural earthquakes. An efficient processing method is needed to save manpower, especially under the current large amount of data records by seismic stations. We apply a SCOUTER algorithm to distinguish between the two types of events. The recognition precision of the trained model for natural earthquakes and blasts can reach 95% and 92.8%, respectively, and the recall can reach 93.4% and 94.6%, respectively. The testing results of data with different epicentral distances and SNR show that our method is stable, independent on regional waveform characteristics and insensitive to data of different SNR. The explanations for each classification at the final confidence also give us a profound enlightenment.
{"title":"Discrimination of seismic and non-seismic signal using SCOUTER","authors":"Kang Wang, Ji Zhang, J. Zhang","doi":"10.1145/3590003.3590045","DOIUrl":"https://doi.org/10.1145/3590003.3590045","url":null,"abstract":"Abstract— For areas with potential occurrence of blasting events, it is essential to distinguish them from natural earthquakes. An efficient processing method is needed to save manpower, especially under the current large amount of data records by seismic stations. We apply a SCOUTER algorithm to distinguish between the two types of events. The recognition precision of the trained model for natural earthquakes and blasts can reach 95% and 92.8%, respectively, and the recall can reach 93.4% and 94.6%, respectively. The testing results of data with different epicentral distances and SNR show that our method is stable, independent on regional waveform characteristics and insensitive to data of different SNR. The explanations for each classification at the final confidence also give us a profound enlightenment.","PeriodicalId":340225,"journal":{"name":"Proceedings of the 2023 2nd Asia Conference on Algorithms, Computing and Machine Learning","volume":"25 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129753285","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}
Image generation is an important area of artificial intelligence that involves creating new images from existing datasets. It involves learning the distribution of target images from randomly generated vectors. Like other deep learning models, the image generation model requires a vast refined data set to produce high-quality results. When there is little data, there is a problem that the diversity and quality of generated images are compromised. In this paper, we propose a new generative model that applies PCA to the generator of the least square error adversarial generative network that, in turn, generates high-quality images even with a small data set. Unlike the existing models that generate target data from randomly generated noise, in the proposed method the direction of the image to be generated is guided by extracting the features of the target data through PCA. The results section shows the superior performance of the proposed model against a different number of images in datasets.
{"title":"Image Generation Model Applying PCA on Latent Space","authors":"Myungseo Song, Asim Niaz, K. Choi","doi":"10.1145/3590003.3590080","DOIUrl":"https://doi.org/10.1145/3590003.3590080","url":null,"abstract":"Image generation is an important area of artificial intelligence that involves creating new images from existing datasets. It involves learning the distribution of target images from randomly generated vectors. Like other deep learning models, the image generation model requires a vast refined data set to produce high-quality results. When there is little data, there is a problem that the diversity and quality of generated images are compromised. In this paper, we propose a new generative model that applies PCA to the generator of the least square error adversarial generative network that, in turn, generates high-quality images even with a small data set. Unlike the existing models that generate target data from randomly generated noise, in the proposed method the direction of the image to be generated is guided by extracting the features of the target data through PCA. The results section shows the superior performance of the proposed model against a different number of images in datasets.","PeriodicalId":340225,"journal":{"name":"Proceedings of the 2023 2nd Asia Conference on Algorithms, Computing and Machine Learning","volume":"63 5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129575586","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}
Recently, artificial intelligence (AI) technologies have applied in the field of clinical medicine widely. And some of researches try to use AI to assist the diagnosis of spinal disease. In this study, HerniationDet: an automatic lumbar disc herniation detection method based on two stage detection framework (e.g. R-CNN, Fast R-CNN, Faster R-CNN, etc.) is presented. Firstly, after comparing the performance of various backbone networks such as VGG, ResNet, EfficientNet, etc., a feature extractor based on VGG16 is constructed to automatically and efficiently extract the necessary feature information from medical images. Secondly, we use region proposal network (RPN) to generate region proposals and provide them to the part of Fast R-CNN for classification and regression. After precisely studying the image of disc herniation, we adjust the scale and radio of the anchor, to make them more in line with the characteristics of the lumbar disc image dataset. Finally, the object detection algorithm is first used on CT images which achieved 89.50% mAP, and then applied to MR images of the lumbar disc to achieve the goal of automatically identifying lumbar disc herniation with or without calcification. Hence, artificial intelligence assisted diagnosis of calcified lumbar disc herniation on MR images can be achieved with 81.24% mAP, by further using a multi-modal learning strategy.
{"title":"Deep Vision Network Based CT Image Detection for Aiding Lumbar Herniated Disc Diagnosis","authors":"W. Xie, Fei-wei Qin, Yanli Shao","doi":"10.1145/3590003.3590092","DOIUrl":"https://doi.org/10.1145/3590003.3590092","url":null,"abstract":"Recently, artificial intelligence (AI) technologies have applied in the field of clinical medicine widely. And some of researches try to use AI to assist the diagnosis of spinal disease. In this study, HerniationDet: an automatic lumbar disc herniation detection method based on two stage detection framework (e.g. R-CNN, Fast R-CNN, Faster R-CNN, etc.) is presented. Firstly, after comparing the performance of various backbone networks such as VGG, ResNet, EfficientNet, etc., a feature extractor based on VGG16 is constructed to automatically and efficiently extract the necessary feature information from medical images. Secondly, we use region proposal network (RPN) to generate region proposals and provide them to the part of Fast R-CNN for classification and regression. After precisely studying the image of disc herniation, we adjust the scale and radio of the anchor, to make them more in line with the characteristics of the lumbar disc image dataset. Finally, the object detection algorithm is first used on CT images which achieved 89.50% mAP, and then applied to MR images of the lumbar disc to achieve the goal of automatically identifying lumbar disc herniation with or without calcification. Hence, artificial intelligence assisted diagnosis of calcified lumbar disc herniation on MR images can be achieved with 81.24% mAP, by further using a multi-modal learning strategy.","PeriodicalId":340225,"journal":{"name":"Proceedings of the 2023 2nd Asia Conference on Algorithms, Computing and Machine Learning","volume":"3 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127124341","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}
{"title":"Performance evaluation of agricultural logistics enterprises based on GA algorithm","authors":"Bin Ye","doi":"10.1145/3590003.3590068","DOIUrl":"https://doi.org/10.1145/3590003.3590068","url":null,"abstract":"","PeriodicalId":340225,"journal":{"name":"Proceedings of the 2023 2nd Asia Conference on Algorithms, Computing and Machine Learning","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127310279","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：