Detection of key component defect for electric transmission tower by deep texture analogy

There are many key components in the transmission line, which often forms different defects and occupies a small size in aerial images. In this paper, we propose a new technique to identify the defect of key components by deep texture analogy, which depends on the deep neural network model SSD combined with the feature fusion module(FFM). Using the trained model of SSD+FFM to extract the feature map to compare. Due to the difference in the content structure and texture between images, the number of feature mismatches is increased in the compare process. The block matching base on the nearest field search method, which enhanced by bidirectional constraint can effectively reduce mismatching, and acquire the spatial different flow map between the analogy images. Extracting the significantly spatial different zone of texture defects, and finally the defect region is outlined. Test and analysis results show that, the method proposed in this paper can effectively identify multiclass defects of key components, greatly improving the intelligence level of the overall inspection.