改进的Faster R-CNN绝缘子识别模型
Improved Faster R-CNN Insulator Recognition Model
郝旭龙 1廖金 1董国芳1
作者信息
- 1. 云南民族大学电气信息工程学院 昆明 650504
- 折叠
摘要
为了适应更多更复杂的传输线路巡检要求,提高巡检实时性,提出了一种基于改进的Faster R-CNN绝缘子目标检测算法模型.对于拍摄传输的巡检绝缘子图像,提出了一种小波变换融合维纳滤波的去噪方法.针对绝缘子的形状特征,将原有的锚框比例1∶1,1∶2,2∶1改为更加适合研究数据的1∶2,2∶1,1∶3,3∶1,并用软化非极大值抑制算法代替原有的非极大值抑制算法解决因拍摄角度而存在的绝缘子遮挡的漏检情况.实验结果表明,改进的Faster R-CNN目标检测算法比原有的算法有更高的识别准确率.
Abstract
In order to meet the requirements of more complex transmission line inspection and improve the real-time perfor-mance of inspection,an improved algorithm model of Faster R-CNN insulator target detection is proposed.A wavelet transform fu-sion wiener filter method is proposed for denoising the patrol insulator images.According to the shape and characteristics of insula-tors,the original anchor frame ratios of 1∶1,1∶2 and 2∶1 are changed to 1∶2,2∶1,1∶3 and 3∶1,which are more suitable for the re-search data,and the softening non-maximum suppression algorithm is used to replace the original non-maximum suppression algo-rithm to solve the problem of insulator occlusion missing due to the shooting angle.Experimental results show that the improved Fast-er R-CNN target detection algorithm has higher recognition accuracy than the original algorithm.
关键词
Faster/R-CNN/目标检测/绝缘子/小波变换/维纳滤波Key words
Faster R-CNN/object detection/insulator/wavelet transform/Wiener filter引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据