摘要
电力系统的安全对于整个能源传输过程至关重要;针对输电线路下超大工程车辆和烟火为主要的外力破坏行为,对单阶段目标检测算法YOLOv5s进行改进,首先针对输电线路多雨雾烟尘等工作环境,引入限制对比度自适应直方图均衡算法CLAHE对图片进行去雾处理,提升图片对比度;针对检测目标距离较远的问题,在YOLOv5s网络的基础上添加CA注意力机制,提升了模型对目标的定位能力;将原网络中的最邻近差值采样方式替换为轻量级通用上采样算子CARAFE,更好地捕捉特征图的同时引入较小的参数量;最后在网络的特征融合层,使用具有通道混洗思想的GSConv卷积模块代替标准卷积模块,减少模型参数量,再利用slim_neck特征融合结构,强化目标关注度,达到减少模型参数量同时提升检测精度的效果;实验结果表明:改进后的YOLOv5s网络,mAP提升了4.4%,参数量减少了3.4%,权重模型内存减小了2.7%,证明了算法的有效性。
It is crucial for the safety of the power system in the entire energy transmission process.Aiming at the main external force destruction behavior of super large engineering vehicles and fireworks in the transmission line,the single-stage target detection algorithm YOOv5s is improved.Firstly,for the working environment of the transmission line with heavy rain,fog and dust,the restricted contrast limited adaptive histogram equalization(CLAHE)algorithm is introduced to defog the image,and improve the image contrast;In response to the long distance problem of detecting targets,a coordinate attention(CA)mechanism is added to the YOLOv5s network to enhance the model s ability to locate targets;The nearest neighbor difference sampling method in the original network is replaced with the lightweight universal up-sampling operator content-aware reassembly of features(CARAFE),which better captures the feature map while introducing smaller parameter quantities;Finally,in the feature fusion layer of the network,a ghost-shuffle convolution(GSConv)module with channel shuffling idea is used to replace the standard convolution module,reducing the model parameters,and then Slim_Neck feature fusion structure is utilized to enhance the target attention,reducing the model parameters while improving the detection accuracy.The experimental results show that the mean average precision(mAP)of the improved YOLOv5s network improves by 4.4%,the number of the parameters reduces by 3.4%,and the memory of the weight model by 2.7%,proving that the algorithm is effectiveness.
作者
郑良成
曹雪虹
焦良葆
高阳
王彦生
ZHENG Liangcheng;CAO Xuehong;JIAO Liangbao;GAO Yang;WANG Yansheng(AI Industrial Technology Research Institute,Nanjing Institute of Technology,Nanjing 211167,China;Jiangsu Intelligent Perception Technology and Equipment Engineering Research Center,Nanjing 211167,China)
出处
《计算机测量与控制》
2024年第2期42-49,共8页
Computer Measurement &Control
基金
江苏省自然科学基金项目(BK20201042)
江苏省政策引导类计划项目(SZ-SQ2020007)。
