To address the challenges of high complexity,poor real-time performance,and low detection rates for small target vehicles in existing vehicle object detection algorithms,this paper proposes a real-time lightweight arc...To address the challenges of high complexity,poor real-time performance,and low detection rates for small target vehicles in existing vehicle object detection algorithms,this paper proposes a real-time lightweight architecture based on You Only Look Once(YOLO)v5m.Firstly,a lightweight upsampling operator called Content-Aware Reassembly of Features(CARAFE)is introduced in the feature fusion layer of the network to maximize the extraction of deep-level features for small target vehicles,reducing the missed detection rate and false detection rate.Secondly,a new prediction layer for tiny targets is added,and the feature fusion network is redesigned to enhance the detection capability for small targets.Finally,this paper applies L1 regularization to train the improved network,followed by pruning and fine-tuning operations to remove redundant channels,reducing computational and parameter complexity and enhancing the detection efficiency of the network.Training is conducted on the VisDrone2019-DET dataset.The experimental results show that the proposed algorithmreduces parameters and computation by 63.8% and 65.8%,respectively.The average detection accuracy improves by 5.15%,and the detection speed reaches 47 images per second,satisfying real-time requirements.Compared with existing approaches,including YOLOv5m and classical vehicle detection algorithms,our method achieves higher accuracy and faster speed for real-time detection of small target vehicles in edge computing.展开更多
随着深度学习技术的发展,水下图像检测近年来受到广泛的关注,为了克服在复杂水下环境下传统小鱼群的误检、漏检和识别准确率低等问题,提出一种改进YOLOv5的目标检测方法(INV-YOLOv5)。该方法包括将YOLOv5m中的Focus模块替换为卷积模块,...随着深度学习技术的发展,水下图像检测近年来受到广泛的关注,为了克服在复杂水下环境下传统小鱼群的误检、漏检和识别准确率低等问题,提出一种改进YOLOv5的目标检测方法(INV-YOLOv5)。该方法包括将YOLOv5m中的Focus模块替换为卷积模块,提高网络精度;在主干网络(Backbone)中添加多头自注意力机制,增大网络特征提取视野;最后,在网络中引入了内卷算子和加权的特征融合,降低网络的参数量,提高检测精度。在实验阶段,使用Labeled Fishes in the Wild数据集和WildFish数据集验证,该方法的平均精度(mAP)分别为81.7%和83.6%,与YOLOv5m网络相比分别提升了6%和14.5%,不仅拥有较高的识别率并且更加轻量化,而且模型大小与YOLOv5m网络相比减少了6 M(Mega)左右,验证了所提出的改进方法具有较好的效果。展开更多
Autonomous unmanned aerial vehicle(UAV)landing is a challenging task,especially on a moving platform in an unstructured environment.Under such a scenario,successful UAV landing is mainly affected by poor UAV localizat...Autonomous unmanned aerial vehicle(UAV)landing is a challenging task,especially on a moving platform in an unstructured environment.Under such a scenario,successful UAV landing is mainly affected by poor UAV localization performance.To solve this problem,we propose a coarse-to-fine visual autonomous UAV landing system based on an enhanced visual positioning approach.The landing platform is marked with a specially designed QR code marker,which is developed to improve the landing accuracy when the UAV approaches the landing site.Besides,we employ the you only look once framework to enhance the visual positioning accuracy,thereby promoting the landing platform detection when the UAV is flying far away.The framework recognizes the QR code and decodes the position of a UAV by the corner points of the QR code.Further,we use the Kalman filter to fuse the position data decoded from the QR code with those from the inertia measurement unit sensor.Then,the position data are used for UAV landing with a developed hierarchical landing strategy.To verify the effectiveness of the proposed system,we performed experiments in different environments under various light conditions.The experimental results demonstrate that the proposed system can achieve UAV landing with high accuracy,strong adaptability,and robustness.In addition,it can achieve accurate landing in different operating environments without external real-time kinematic global positioning system(RTK-GPS)signals,and the average landing error is 11.5 cm,which is similar to the landing error when using RTK-GPS signals as the ground truth.展开更多
基金funded by the General Project of Key Research and Develop-ment Plan of Shaanxi Province(No.2022NY-087).
文摘To address the challenges of high complexity,poor real-time performance,and low detection rates for small target vehicles in existing vehicle object detection algorithms,this paper proposes a real-time lightweight architecture based on You Only Look Once(YOLO)v5m.Firstly,a lightweight upsampling operator called Content-Aware Reassembly of Features(CARAFE)is introduced in the feature fusion layer of the network to maximize the extraction of deep-level features for small target vehicles,reducing the missed detection rate and false detection rate.Secondly,a new prediction layer for tiny targets is added,and the feature fusion network is redesigned to enhance the detection capability for small targets.Finally,this paper applies L1 regularization to train the improved network,followed by pruning and fine-tuning operations to remove redundant channels,reducing computational and parameter complexity and enhancing the detection efficiency of the network.Training is conducted on the VisDrone2019-DET dataset.The experimental results show that the proposed algorithmreduces parameters and computation by 63.8% and 65.8%,respectively.The average detection accuracy improves by 5.15%,and the detection speed reaches 47 images per second,satisfying real-time requirements.Compared with existing approaches,including YOLOv5m and classical vehicle detection algorithms,our method achieves higher accuracy and faster speed for real-time detection of small target vehicles in edge computing.
文摘随着深度学习技术的发展,水下图像检测近年来受到广泛的关注,为了克服在复杂水下环境下传统小鱼群的误检、漏检和识别准确率低等问题,提出一种改进YOLOv5的目标检测方法(INV-YOLOv5)。该方法包括将YOLOv5m中的Focus模块替换为卷积模块,提高网络精度;在主干网络(Backbone)中添加多头自注意力机制,增大网络特征提取视野;最后,在网络中引入了内卷算子和加权的特征融合,降低网络的参数量,提高检测精度。在实验阶段,使用Labeled Fishes in the Wild数据集和WildFish数据集验证,该方法的平均精度(mAP)分别为81.7%和83.6%,与YOLOv5m网络相比分别提升了6%和14.5%,不仅拥有较高的识别率并且更加轻量化,而且模型大小与YOLOv5m网络相比减少了6 M(Mega)左右,验证了所提出的改进方法具有较好的效果。
文摘Autonomous unmanned aerial vehicle(UAV)landing is a challenging task,especially on a moving platform in an unstructured environment.Under such a scenario,successful UAV landing is mainly affected by poor UAV localization performance.To solve this problem,we propose a coarse-to-fine visual autonomous UAV landing system based on an enhanced visual positioning approach.The landing platform is marked with a specially designed QR code marker,which is developed to improve the landing accuracy when the UAV approaches the landing site.Besides,we employ the you only look once framework to enhance the visual positioning accuracy,thereby promoting the landing platform detection when the UAV is flying far away.The framework recognizes the QR code and decodes the position of a UAV by the corner points of the QR code.Further,we use the Kalman filter to fuse the position data decoded from the QR code with those from the inertia measurement unit sensor.Then,the position data are used for UAV landing with a developed hierarchical landing strategy.To verify the effectiveness of the proposed system,we performed experiments in different environments under various light conditions.The experimental results demonstrate that the proposed system can achieve UAV landing with high accuracy,strong adaptability,and robustness.In addition,it can achieve accurate landing in different operating environments without external real-time kinematic global positioning system(RTK-GPS)signals,and the average landing error is 11.5 cm,which is similar to the landing error when using RTK-GPS signals as the ground truth.
