摘要
针对巡检机器人无法有效识别电力直流系统蓄电池缺陷的问题进行了深入研究。采用改进的YOLOv4网络结构,对骨干层网络进行轻量化设计,在保证特征提取能力不变的情况下降低网络模型复杂度。通过对蓄电池缺陷图像进行预处理,得到直方图均衡化处理后的灰度图像,利用Faster R-CNN网络提取缺陷图像特征,对激活函数进行优化,增加注意力机制,提高模型的缺陷检测能力。测试表明,该改进模型对蓄电池常见缺陷的检测率达90%以上,为电力直流系统蓄电池缺陷检测提供了新路径。
In depth research has been conducted on the problem of substation inspection robots being unable to effectively identify battery defects in DC power systems.An improved YOLOv4 network structure is adopted to lightweight the backbone layer network,reducing the complexity of the network model while ensuring the same feature extraction ability.The battery defect image is preprocessed to obtain a histogram balanced grayscale image.The Faster R-CNN network is used to extract the defect image features.The activation function is optimized and attention mechanism is added to improve the defect detection ability of the model.Tests have shown that the improved model has a detection rate of over 90%for common defects in batteries,providing a new method for detecting battery defects in DC power systems.
作者
贾红兵
黄泰山
王月灿
李庚生
张航
JIA Hongbing;HUANG Taishan;WANG Yuecan;LI Gengsheng;ZHANG Hang(Wudongde Hydropower Plant,Luquan 651512,China)
出处
《电子设计工程》
2025年第24期118-122,共5页
Electronic Design Engineering
基金
三峡金沙江云川水电开发有限公司科技项目(5223020031)。
