Attention Model and Soft-NMS-Based Transmission Line Small Target Detection Method

In the defect detection of transmission lines, the bird's nest, plastic, rags and other suspended solids are mostly small targets. They have few pixels in the images and are easy to be disturbed by the background, which make the detection accuracy needs to be improved. In this paper, a new two-stage object detection algorithm is designed to improve the detection effect of bird nests and suspended solids in transmission lines. In order to improve the detection performance of small targets, the attention mechanism is integrated into the feature extraction network to learn more rich context information. In addition, in the detection network, a post-processing method based on softer non maximum suppression algorithm is designed to reduce the loss of small targets. Compared with the commonly used two-stage object detection algorithms, the proposed method improves the average accuracy of the two categories by about 4.7% and 5.9%, respectively, and has greater value in practical applications.