计算机工程

• •    

基于改进Faster RCNN的输电线路航拍绝缘子检

  

  • 发布日期:2021-01-08

Aerial insulator detection of transmission line based on improved Faster RCNN

  • Published:2021-01-08

摘要: 针对已有绝缘子检测算法对输电线路中绝缘子部分遮挡、尺寸不一、检测难度大、精确度低的问题, 提出了一种基于改进 Faster RCNN 的输电线路航拍绝缘子检测方法。在 Faster RCNN 原始模型上运用多尺度训练 同时根据绝缘子自身特点调整滑动窗口产生的候选区域比例;并引入防止遮挡物体难检测的在线困难样本对手 生成策略,然后对测试样本进行检测识别。实验结果表明,改进后的网络模型相较原始 Faster RCNN 模型精确 度提升了 4.33%,能更为准确地检测出目标绝缘子。

Abstract: Aiming at the problems of existing insulator detection algorithms, such as partial occlusion, different sizes, difficult detection and low accuracy, an aerial insulator detection method based on improved Faster RCNN is proposed. The multi-scale training is applied to the original Faster RCNN model, and the proportion of candidate regions generated by sliding window is adjusted according to the characteristics of insulators; In addition, an online difficult sample opponent generation strategy is introduced to prevent the occlusion objects from being difficult to detect, and then the test samples are detected and identified. The experimental results show that the accuracy of the improved network model is improved by 4.33% compared with the original Faster RCNN model, and it can detect the target insulator more accurately.