In the industrial Internet environment, the introduction of network slicing supports the connection of a large number of devices with different service requirements (QoS) sharing the same physical resources. Aiming at the problem of the adaptability of massive terminal devices and networks in industrial heterogeneous scenarios, this paper proposes a network slice prediction algorithm based on multi-dimensional and deep neural network (MDNN) based on the multi-dimensional resource network requirements of different terminal devices in specific industrial scenarios. The network slice prediction algorithm predicts the network resources required by the device at the next moment according to the historical network requirements and historical slice selection of the device, and selects the appropriate network slice for the device according to the prediction result. The simulation results show that the prediction accuracy of the proposed algorithm can reach 98.70%, which greatly improves the adaptability of the device and the network.
{"title":"Industrial Internet Network Slice Prediction Algorithm Based on Multidimensional and Deep Neural Networks","authors":"Jihong Zhao, Gao-Jing Peng","doi":"10.1145/3573942.3573989","DOIUrl":"https://doi.org/10.1145/3573942.3573989","url":null,"abstract":"In the industrial Internet environment, the introduction of network slicing supports the connection of a large number of devices with different service requirements (QoS) sharing the same physical resources. Aiming at the problem of the adaptability of massive terminal devices and networks in industrial heterogeneous scenarios, this paper proposes a network slice prediction algorithm based on multi-dimensional and deep neural network (MDNN) based on the multi-dimensional resource network requirements of different terminal devices in specific industrial scenarios. The network slice prediction algorithm predicts the network resources required by the device at the next moment according to the historical network requirements and historical slice selection of the device, and selects the appropriate network slice for the device according to the prediction result. The simulation results show that the prediction accuracy of the proposed algorithm can reach 98.70%, which greatly improves the adaptability of the device and the network.","PeriodicalId":103293,"journal":{"name":"Proceedings of the 2022 5th International Conference on Artificial Intelligence and Pattern Recognition","volume":"14 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133822980","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}
With the development of deep learning, small object detection has a significant role in application fields such as smart factories and remote sensing images. In order to address the problem of difficult and low accuracy detection of small objects due to small pixel scale and little feature information. In this paper, we present a path aggregation network with residual characteristic RPANet on YOLOv3 algorithm, which can twice use the feature information of the backbone network to enhance the small object feature information, and also offer a positional convolution attention mechanism module PCAM to thoroughly learn and extract the small object feature information as well as reduce the unnecessary feature information in the background, so as to further enhance the detection capability of the model for small objects. The experimental results demonstrate that the improved YOLOv3 algorithm is more effective for small object detection.
{"title":"Small Object Detection Algorithm Based on RPANet and Positional Convolution Attention Mechanism","authors":"Zongbing Tang, Dan Yang, Junsuo Qu","doi":"10.1145/3573942.3574031","DOIUrl":"https://doi.org/10.1145/3573942.3574031","url":null,"abstract":"With the development of deep learning, small object detection has a significant role in application fields such as smart factories and remote sensing images. In order to address the problem of difficult and low accuracy detection of small objects due to small pixel scale and little feature information. In this paper, we present a path aggregation network with residual characteristic RPANet on YOLOv3 algorithm, which can twice use the feature information of the backbone network to enhance the small object feature information, and also offer a positional convolution attention mechanism module PCAM to thoroughly learn and extract the small object feature information as well as reduce the unnecessary feature information in the background, so as to further enhance the detection capability of the model for small objects. The experimental results demonstrate that the improved YOLOv3 algorithm is more effective for small object detection.","PeriodicalId":103293,"journal":{"name":"Proceedings of the 2022 5th International Conference on Artificial Intelligence and Pattern Recognition","volume":"5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124542863","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 work focuses on dealing with fine-grained recognition problems when incremental classes emerge. The task is desirable to not only distinguish subordinate visual classes based on discriminative but subtle object parts, but also recognize new coming sub-classes without suffering from catastrophic forgetting. In this paper, we first propose to localize both object- and part-level image regions for capturing powerful fine-grained patterns. Then, these fine-grained regions are fed into a bilateral network consisting of a stable branch and a flexible branch for supporting observed and incremental sub-classes recognition respectively. Moreover, a cumulative adaptation strategy is further equipped to adjust the network training during the incremental sessions. Meanwhile, to better retain the modeling capability of observed classes, we also replay samples from previous classes by a hallucination approach. Experiments are conducted on three popular fine-grained recognition datasets and results of the proposed method can reveal its superiority over state-of-the-arts.
{"title":"Fine-Grained Recognition with Incremental Classes","authors":"Yangqiaoyu Zhou","doi":"10.1145/3573942.3574078","DOIUrl":"https://doi.org/10.1145/3573942.3574078","url":null,"abstract":"This work focuses on dealing with fine-grained recognition problems when incremental classes emerge. The task is desirable to not only distinguish subordinate visual classes based on discriminative but subtle object parts, but also recognize new coming sub-classes without suffering from catastrophic forgetting. In this paper, we first propose to localize both object- and part-level image regions for capturing powerful fine-grained patterns. Then, these fine-grained regions are fed into a bilateral network consisting of a stable branch and a flexible branch for supporting observed and incremental sub-classes recognition respectively. Moreover, a cumulative adaptation strategy is further equipped to adjust the network training during the incremental sessions. Meanwhile, to better retain the modeling capability of observed classes, we also replay samples from previous classes by a hallucination approach. Experiments are conducted on three popular fine-grained recognition datasets and results of the proposed method can reveal its superiority over state-of-the-arts.","PeriodicalId":103293,"journal":{"name":"Proceedings of the 2022 5th International Conference on Artificial Intelligence and Pattern Recognition","volume":"38 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129273409","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}
To achieve the minimum resource cost of service function chain deployment under the dynamic changes of the substrate network resources and the high dimension of the network model, this paper proposes a service function chain deployment algorithm based on a double deep Q network. Firstly, according to the characteristics that the substrate network resources change dynamically with the arrival of the service function chain, the deployment of the service function chain is converted into a Markov decision process. Then, the resource cost is used as the reward function, and finally, the service function chain is solved using the double deep Q network algorithm. Dynamically arriving at optimal deployment strategy. The simulation results show that the algorithm can effectively improve the request acceptance rate and reduce the average deployment cost and delay.
{"title":"Service Function Chain Deployment Algorithm Based on Double Deep Q Network","authors":"Guohui Zhu, Chaohang Zhao","doi":"10.1145/3573942.3573990","DOIUrl":"https://doi.org/10.1145/3573942.3573990","url":null,"abstract":"To achieve the minimum resource cost of service function chain deployment under the dynamic changes of the substrate network resources and the high dimension of the network model, this paper proposes a service function chain deployment algorithm based on a double deep Q network. Firstly, according to the characteristics that the substrate network resources change dynamically with the arrival of the service function chain, the deployment of the service function chain is converted into a Markov decision process. Then, the resource cost is used as the reward function, and finally, the service function chain is solved using the double deep Q network algorithm. Dynamically arriving at optimal deployment strategy. The simulation results show that the algorithm can effectively improve the request acceptance rate and reduce the average deployment cost and delay.","PeriodicalId":103293,"journal":{"name":"Proceedings of the 2022 5th International Conference on Artificial Intelligence and Pattern Recognition","volume":"228 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116719040","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}
Driver distraction has surfaced as a significant safety issue worldwide, and the capacity to track a driver's attention via monitoring its gaze direction is one of the most critical features in the modern Driver Monitoring System (DMS). Deep learning based gaze estimation has grown in popularity due to its robustness across operating conditions. Though appropriate network structure design and parameters tuning are important, accurate ground-truth estimation for millions of gaze training images to build the model also plays a critical role in achieving high-quality gaze estimation results. This paper proposes a non-intrusive automatic 3D ground-truth data collection system for large-scale on-bench and in-car data collection, using gamified camera calibration, occlusion invariant mirror-based camera localization, and noise-robust 3D reconstruction algorithms. Experimental results are provided to demonstrate the system's accuracy and robustness even in challenging conditions.
{"title":"Non-intrusive Automatic 3D Gaze Ground-truth System","authors":"Feng Hu","doi":"10.1145/3573942.3574068","DOIUrl":"https://doi.org/10.1145/3573942.3574068","url":null,"abstract":"Driver distraction has surfaced as a significant safety issue worldwide, and the capacity to track a driver's attention via monitoring its gaze direction is one of the most critical features in the modern Driver Monitoring System (DMS). Deep learning based gaze estimation has grown in popularity due to its robustness across operating conditions. Though appropriate network structure design and parameters tuning are important, accurate ground-truth estimation for millions of gaze training images to build the model also plays a critical role in achieving high-quality gaze estimation results. This paper proposes a non-intrusive automatic 3D ground-truth data collection system for large-scale on-bench and in-car data collection, using gamified camera calibration, occlusion invariant mirror-based camera localization, and noise-robust 3D reconstruction algorithms. Experimental results are provided to demonstrate the system's accuracy and robustness even in challenging conditions.","PeriodicalId":103293,"journal":{"name":"Proceedings of the 2022 5th International Conference on Artificial Intelligence and Pattern Recognition","volume":"206 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114300620","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}
Non-Drug Intervention (NDI) is one of the important means to prevent and control the outbreak of coronavirus disease 2019 (COVID-19), and the implementation of this series of measures plays a key role in the development of the epidemic. The purpose of this paper is to study the impact of different mitigation measures on the situation of the COVID 19, and effectively respond to the prevention and control situation in the "post-epidemic era". The present work is based on the Susceptible-Exposed-Infectious-Remove-Susceptible (SEIRS) Model, and adapted the agent-based model (ABM) to construct the epidemic prevention and control model framework to simulate the COVID-19 epidemic from three aspects: social distance, personal protection, and bed resources. The experiment results show that the above NDI are effective mitigation measures for epidemic prevention and control, and can play a positive role in the recurrence of COVID-19, but a single measure cannot prevent the recurrence of infection peaks and curb the spread of the epidemic; When social distance and personal protection rules are out of control, bed resources will become an important guarantee for epidemic prevention and control. Although the spread of the epidemic cannot be curbed, it can slow down the recurrence of the peak of the epidemic; When people abide by social distance and personal protection rules, the pressure on bed resources will be eased. At the same time, under the interaction of the three measures, not only the death toll can be reduced, but the spread of the epidemic can also be effectively curbed.
{"title":"Build an Agent-Based Model for COVID-19 Effect of Mitigation Policies","authors":"Jianhua Zeng, Ping Lu, Kai-Biao Lin","doi":"10.1145/3573942.3574022","DOIUrl":"https://doi.org/10.1145/3573942.3574022","url":null,"abstract":"Non-Drug Intervention (NDI) is one of the important means to prevent and control the outbreak of coronavirus disease 2019 (COVID-19), and the implementation of this series of measures plays a key role in the development of the epidemic. The purpose of this paper is to study the impact of different mitigation measures on the situation of the COVID 19, and effectively respond to the prevention and control situation in the \"post-epidemic era\". The present work is based on the Susceptible-Exposed-Infectious-Remove-Susceptible (SEIRS) Model, and adapted the agent-based model (ABM) to construct the epidemic prevention and control model framework to simulate the COVID-19 epidemic from three aspects: social distance, personal protection, and bed resources. The experiment results show that the above NDI are effective mitigation measures for epidemic prevention and control, and can play a positive role in the recurrence of COVID-19, but a single measure cannot prevent the recurrence of infection peaks and curb the spread of the epidemic; When social distance and personal protection rules are out of control, bed resources will become an important guarantee for epidemic prevention and control. Although the spread of the epidemic cannot be curbed, it can slow down the recurrence of the peak of the epidemic; When people abide by social distance and personal protection rules, the pressure on bed resources will be eased. At the same time, under the interaction of the three measures, not only the death toll can be reduced, but the spread of the epidemic can also be effectively curbed.","PeriodicalId":103293,"journal":{"name":"Proceedings of the 2022 5th International Conference on Artificial Intelligence and Pattern Recognition","volume":"51-52 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114474369","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 recent years, deep convolution neural networks have made great progress in object detection tasks. Generally speaking, the bounding box and the type of bounding box play a very important role in object detection. However, it is not easy for convolution neural networks to directly generate disordered bounding boxes. A widely used solution is to adopt the idea of divide and conquer and introduce the concept of anchor box. At present, anchor frame mechanism has been widely used in top-level object detection framework, and has achieved good results on common datasets. The innovation of this paper is that a novel anchor frame generation method is proposed, which can generate error frames with various aspect ratios for object detection frames. Different from the previous method of generating the anchor box in a predefined way, the anchor box in this method is dynamically generated by the anchor box generator. The feature is that the anchor box generator is not fixed, but learns from anchor boxes defined by fixed rules, which means that the anchor box generator can be adapted to a variety of scenarios. In this paper, the dynamic anchor frame method is used to detect the traffic road. In addition, the weights of the anchor box generator are predicted by a small network whose inputs are predefined anchor boxes. Compared with the traditional anchor frame generation methods, the proposed anchor frame generator has the following innovations: (1) it adaptive adjusts the size and aspect ratio of the anchor frame to improve the quality of the anchor frame. (2) The adaptive IOU country value is used to balance the number of positive samples of the size target. Finally, good efficiency and results are obtained.
{"title":"Traffic Road Detection Based on Dynamic Anchor Frame","authors":"Xingya Yan, Yujiao Ding, Yue Li","doi":"10.1145/3573942.3574111","DOIUrl":"https://doi.org/10.1145/3573942.3574111","url":null,"abstract":"In recent years, deep convolution neural networks have made great progress in object detection tasks. Generally speaking, the bounding box and the type of bounding box play a very important role in object detection. However, it is not easy for convolution neural networks to directly generate disordered bounding boxes. A widely used solution is to adopt the idea of divide and conquer and introduce the concept of anchor box. At present, anchor frame mechanism has been widely used in top-level object detection framework, and has achieved good results on common datasets. The innovation of this paper is that a novel anchor frame generation method is proposed, which can generate error frames with various aspect ratios for object detection frames. Different from the previous method of generating the anchor box in a predefined way, the anchor box in this method is dynamically generated by the anchor box generator. The feature is that the anchor box generator is not fixed, but learns from anchor boxes defined by fixed rules, which means that the anchor box generator can be adapted to a variety of scenarios. In this paper, the dynamic anchor frame method is used to detect the traffic road. In addition, the weights of the anchor box generator are predicted by a small network whose inputs are predefined anchor boxes. Compared with the traditional anchor frame generation methods, the proposed anchor frame generator has the following innovations: (1) it adaptive adjusts the size and aspect ratio of the anchor frame to improve the quality of the anchor frame. (2) The adaptive IOU country value is used to balance the number of positive samples of the size target. Finally, good efficiency and results are obtained.","PeriodicalId":103293,"journal":{"name":"Proceedings of the 2022 5th International Conference on Artificial Intelligence and Pattern Recognition","volume":"18 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114786325","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}
Quantum satellite communication has the natural advantages of strong survival reliability and wide coverage, and is currently a research hotspot in the field of communication at home and abroad. The successful launch of the "Mozi" scientific experimental satellite has laid the foundation for the construction of the quantum space-earth integrated communication network. In order to further improve the communication performance of the quantum satellite-ground link, a tuning strategy based on quantum coupled associative feedback control (QCAFC) is proposed in this paper. QCAFC estimates the state information of atoms by measuring the photons leaked in the optical cavity, and adjusts the controller to change the spin state of the atoms in the optical cavity. The evolution of the system is studied, the influence of the QCAFC system on the performance parameters of the satellite-ground quantum communication is analyzed, and the simulation is verified. The simulation results show that in the amplitude damped channel, the QCAFC system can significantly improve the channel capacity and coherence. When transmitting in the plasma environment, when the particle radius is 5, when the transmission distance between the satellite and the earth increases from 50km to 200km, the bit error rate of the system without QCAFC is increased from 7.34×10-3 to 21.93×10-3, and the bit error rate of the system with QCAFC is increased from 4.81×10-3 to 14.72×10-3. Simulation results show that the use of QCAFC system has a significant improvement in the performance of quantum satellite communication.
{"title":"Based on Coupled Associative Feedback Control of Quantum Satellite Communication Performance Tuning Strategy","authors":"Yeliang Gong, Min Nie, Guang Yang","doi":"10.1145/3573942.3573986","DOIUrl":"https://doi.org/10.1145/3573942.3573986","url":null,"abstract":"Quantum satellite communication has the natural advantages of strong survival reliability and wide coverage, and is currently a research hotspot in the field of communication at home and abroad. The successful launch of the \"Mozi\" scientific experimental satellite has laid the foundation for the construction of the quantum space-earth integrated communication network. In order to further improve the communication performance of the quantum satellite-ground link, a tuning strategy based on quantum coupled associative feedback control (QCAFC) is proposed in this paper. QCAFC estimates the state information of atoms by measuring the photons leaked in the optical cavity, and adjusts the controller to change the spin state of the atoms in the optical cavity. The evolution of the system is studied, the influence of the QCAFC system on the performance parameters of the satellite-ground quantum communication is analyzed, and the simulation is verified. The simulation results show that in the amplitude damped channel, the QCAFC system can significantly improve the channel capacity and coherence. When transmitting in the plasma environment, when the particle radius is 5, when the transmission distance between the satellite and the earth increases from 50km to 200km, the bit error rate of the system without QCAFC is increased from 7.34×10-3 to 21.93×10-3, and the bit error rate of the system with QCAFC is increased from 4.81×10-3 to 14.72×10-3. Simulation results show that the use of QCAFC system has a significant improvement in the performance of quantum satellite communication.","PeriodicalId":103293,"journal":{"name":"Proceedings of the 2022 5th International Conference on Artificial Intelligence and Pattern Recognition","volume":"161 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114863085","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}
Lei Li, Yao Liu, Wenjin Zhang, Xiangyu Li, Jiantao Chang
The ultra-short-term forecast of wind conditions is mainly concentrated in the forecast range of a few minutes and has an important guiding role in wind power system dispatching, wind turbine control, and wind power load tracking. Due to the characteristics of sudden change, non-stationarity, and volatility of short-term wind direction and wind speed, these random and volatile properties bring great difficulties to the prediction of ultra-short-term wind conditions. The current research only predicts a single wind speed or wind direction and does not predict both at the same time, which also brings certain limitations to the dispatching of wind power systems. Given the above characteristics of wind speed and wind direction, the decomposition method can be used to divide it into multi-scale components, thereby reducing the complexity of the original signal, increasing the stability of the signal, and improving the accuracy of prediction. Therefore, this paper uses the VMD decomposition method to decompose the original wind direction and wind speed data constructs multi-scale prediction features, and explores the laws of each component. The bi-directional GRU model has a strong ability to capture the sequence fluctuation law, and the decomposed modal components are input into the bi-directional GRU model to predict the wind speed. Through a large number of experiments and the comparison of different methods, it is shown that the VMD-BiGRU-based model has high prediction accuracy, small error, and higher efficiency in wind direction and wind speed prediction.
{"title":"Ultra-short-term wind forecast of the wind farm based on VMD-BiGRU","authors":"Lei Li, Yao Liu, Wenjin Zhang, Xiangyu Li, Jiantao Chang","doi":"10.1145/3573942.3574009","DOIUrl":"https://doi.org/10.1145/3573942.3574009","url":null,"abstract":"The ultra-short-term forecast of wind conditions is mainly concentrated in the forecast range of a few minutes and has an important guiding role in wind power system dispatching, wind turbine control, and wind power load tracking. Due to the characteristics of sudden change, non-stationarity, and volatility of short-term wind direction and wind speed, these random and volatile properties bring great difficulties to the prediction of ultra-short-term wind conditions. The current research only predicts a single wind speed or wind direction and does not predict both at the same time, which also brings certain limitations to the dispatching of wind power systems. Given the above characteristics of wind speed and wind direction, the decomposition method can be used to divide it into multi-scale components, thereby reducing the complexity of the original signal, increasing the stability of the signal, and improving the accuracy of prediction. Therefore, this paper uses the VMD decomposition method to decompose the original wind direction and wind speed data constructs multi-scale prediction features, and explores the laws of each component. The bi-directional GRU model has a strong ability to capture the sequence fluctuation law, and the decomposed modal components are input into the bi-directional GRU model to predict the wind speed. Through a large number of experiments and the comparison of different methods, it is shown that the VMD-BiGRU-based model has high prediction accuracy, small error, and higher efficiency in wind direction and wind speed prediction.","PeriodicalId":103293,"journal":{"name":"Proceedings of the 2022 5th International Conference on Artificial Intelligence and Pattern Recognition","volume":"7 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126107307","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}
To address the problem that the existing object detection network models are large in size and complex in operation and cannot satisfy both detection speed and accuracy under the limited resources and small size platform. Based on YOLOv4 as the benchmark network, a lightweight object detection model LW-YOLO is proposed. Firstly, the backbone feature extraction network is replaced with MobileNetv1, while the number of feature fusion network parameters is significantly reduced by the depth separable convolutional module. Then the BN layer coefficients are used as scaling factors for the importance of the convolutional channels, the scaling factors are sparse using polarization regularization, the errors before and after pruning are reconstructed using least squares and channel weighting methods. The appropriate pruning thresholds are obtained by minimizing the reconstructed errors, the channels with small scaling factor values are eliminated to achieve the lightweight. The experimental results on the VOC (Visual Object Classes) dataset show that the detection accuracy of LW-YOLO is 87.00%, and the FPS(Frames Per Second ) reaches 48.89, which is better than the original YOLOv4 algorithm. It also significantly reduces the number of parameters, computation, and model size, which is more suitable for application in resource-poor embedded mobile devices.
针对现有目标检测网络模型规模大、操作复杂，在有限的资源和小型平台下无法同时满足检测速度和精度的问题。以YOLOv4为基准网络，提出了一种轻量级的目标检测模型LW-YOLO。首先，用MobileNetv1取代主干特征提取网络，同时利用深度可分卷积模块显著减少特征融合网络参数的数量;然后将BN层系数作为卷积信道重要性的标度因子，利用极化正则化对标度因子进行稀疏处理，利用最小二乘法和信道加权法重构剪枝前后的误差。通过最小化重构误差获得合适的剪枝阈值，剔除比例因子较小的通道，实现轻量化。在VOC (Visual Object Classes)数据集上的实验结果表明，LW-YOLO的检测准确率为87.00%，FPS(Frames Per Second)达到48.89，优于原来的YOLOv4算法。它还显著减少了参数数量、计算量和模型尺寸，更适合在资源贫乏的嵌入式移动设备中应用。
{"title":"Lightweight Improved Based on YOLOv4 Object Detection Algorithm","authors":"Rui Chen, Zhenzhong Li, Yuzhao Zhang, Yuehang Li","doi":"10.1145/3573942.3574027","DOIUrl":"https://doi.org/10.1145/3573942.3574027","url":null,"abstract":"To address the problem that the existing object detection network models are large in size and complex in operation and cannot satisfy both detection speed and accuracy under the limited resources and small size platform. Based on YOLOv4 as the benchmark network, a lightweight object detection model LW-YOLO is proposed. Firstly, the backbone feature extraction network is replaced with MobileNetv1, while the number of feature fusion network parameters is significantly reduced by the depth separable convolutional module. Then the BN layer coefficients are used as scaling factors for the importance of the convolutional channels, the scaling factors are sparse using polarization regularization, the errors before and after pruning are reconstructed using least squares and channel weighting methods. The appropriate pruning thresholds are obtained by minimizing the reconstructed errors, the channels with small scaling factor values are eliminated to achieve the lightweight. The experimental results on the VOC (Visual Object Classes) dataset show that the detection accuracy of LW-YOLO is 87.00%, and the FPS(Frames Per Second ) reaches 48.89, which is better than the original YOLOv4 algorithm. It also significantly reduces the number of parameters, computation, and model size, which is more suitable for application in resource-poor embedded mobile devices.","PeriodicalId":103293,"journal":{"name":"Proceedings of the 2022 5th International Conference on Artificial Intelligence and Pattern Recognition","volume":"48 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125711588","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：