基于深度学习算法的航拍巡检图像的绝缘子识别

针对绝缘子检测目标在航拍图像中尺寸变化剧烈的问题,提出一种改进Faster R-CNN的绝缘子检测算法.首先将FPN特征金字塔结构网络与Faster R-CNN算法进行结合,将不同尺度下的特征进行融合;然后,改进最大池化层,提升检测框的坐标精度;针对遮挡现象,采用Soft-NMS算法规避不同目标检测框因重叠而被误删的情况.经过对绝缘子航拍数据集的检测验证,对比原Faster R-CNN网络,本改进网络结构提高了平均准确率(MAP),且可以更有效地识别图像中更小比例的绝缘子目标.

Insulator recognition of aerial patrol image based on deep learning algorithm

As an important component of transmission lines, insulators are of great significance to the stable operation and safety of transmission lines. Aiming at the problem that the size of insulator detection target changes dramatically in aerial image, this paper proposes an improved insulator detection algorithm of Faster R-CNN. Firstly, FPN characteristic pyramid structure network was combined with Faster R-CNN algorithm to fuse the characteristics at different scales. Then the maximum pooling layer is improved to improve the coordinate accuracy of the detection box. Soft -NMS algorithm is used to avoid the overlap of different detection frames. After the detection and verification of the insulator aerial data set, compared with the original Faster R-CNN network, the improved network structure proposed in this paper improves the average accuracy (MAP), and can more effectively identify smaller proportion of insulator targets in the image.