Defect Detection of Small-size Fittings Based on Reformative Cascade RCNN Network

This research suggests a reformative Cascade RCNN defect detection network to identify and detect defects in small-sized fittings on transmission lines and lower the risk of power system operation. Firstly, we improve the FPN network, integrate flat edge information layer by layer into the deep network, and improve the properties of small accessories. Secondly, the feature sampling loss of the down-sampling operation is reduced by using multi-branch dilated convolution during feature fusion. According to experiment results, the combined optimized Cascade RCNN network is better than the unimproved Cascade RCNN network and other common object detection networks at detecting flawed small-sized fittings in practical applications.