摘要
目前的绝缘子及缺陷目标检测算法中普遍存在着诸如误检、漏检和检测精度低等一系列问题,提出一种改进的YOLOv9绝缘子及缺陷目标检测算法来更好地检测绝缘子及其残缺缺陷。首先,在YOLOv9的核心模块RepNCSPELAN中嵌入多样性分支块DBB,DBB可以增强单个卷积的表示能力,丰富特征空间,提高模型的特征提取能力,提升模型性能,同时基本不增加推理时间成本。其次,使用Haar小波的下采样HWD替换传统下采样,可以降低特征图的空间分辨率,同时保留尽可能多的信息,并且与传统的下采样方法相比,可以有效降低信息不确定性。最后使用MPDIoU作为模型的损失函数,MPDIoU通过直接计算预测框和真实框之间的关键点距离,能更准确地反映预测框和真实框之间的差异,从而提升模型的平均精度。在绝缘子及缺陷数据集上,改进后的算法YOLOv9-DHM的平均检测精度(Mean Average Precision,mAP)提高至96.8%,相比于原始算法提高了2.2%,精确率和召回率分别提高至95.4%和94.5%。改进后的算法相比原始算法,平均检测精度有明显提升,证明了算法改进后的可行性。
There are a series of issues in the present insulator and defect target detection algorithms,such as false positives,missed detections,and low detection accuracy.An improved YOLOv9 insulator and defect detection algorithm to achieve better detection of insulators and their defects,such as incomplete insulators is proposed.First,diversity branch blocks(DBB)are embedded into the core module RepNCSPELAN of YOLOv9.DBB can enhance the representation ability of individual convolutions,enrich the feature space,improve the model's feature extraction capability,and enhance model performance,while not increasing any inference time cost.Second,Haar wavelet downsampling(HWD)is used to replace traditional downsampling.It can reduce the spatial resolution of the feature map while retaining as much information as possible,and compared with traditional downsampling methods,it can effectively reduce information uncertainty.Finally,MPDIoU is used as the model’s loss function.MPDIoU can more accurately reflect the difference between the predicted and actual bounding boxes by directly calculating the distance between key points,thereby improving the model’s average precision.On the insulator and defect dataset,the improved algorithm YOLOv9-DHM’s mean average precision(mAP)is increased to 96.8%,an increase of 2.2%compared to the original algorithm,with precision and recall rates increased to 95.4%and 94.5%,respectively.The improved algorithm shows a significant increase in average detection accuracy compared to the original algorithm,proving the feasibility of the algorithm improvement.
作者
罗希
贺强
张宁轩
石超君
Luo Xi;He Qiang;Zhang Ningxuan;Shi Chaojun(Gui’an Power Supply Bureau,Guizhou Power Grid Co.,Ltd.,Guiyang 550029,China;Guiyang Bureau of UHV Power Transmission Company,China Southern Power Grid Co.,Ltd.,Guiyang 550002,China;Department of Electronic and Communication Engineering,North China Electric Power University,Baoding,Hebei 071003,China)
出处
《机电工程技术》
2024年第10期197-202,共6页
Mechanical & Electrical Engineering Technology
基金
中国南方电网有限责任公司科技项目(010200KK52222001)。
