Two-Stage Quality Assessment Method of Power Equipment X-Ray Image Based on Improved CenterNet

Abstract

In order to improve the recognition accuracy of X-ray inspection image quality of electric power equipment, also ensure defect detection and recognition algorithm models to obtain high-quality X-ray images, in view of the characteristics of the original X-ray image of power equipment based on twin-wire image quality meter, such as large resolution, relatively small size and compact arrangement of twin-wire pairs, a two-stage X-ray image quality assessment method for power equipment based on improved CenterNet is proposed. Firstly, in the coarse detection stage, the original CenterNet is used to detect the double-wire image quality meter region in the X-ray image. Then, in the fine detection stage, the improved CenterNet is used to detect the number of double-wire pairs in the double-wire image quality meter image. Finally, image quality assessment results are achieved according to the double-wire image quality meter evaluation standard. In order to solve the problems of low resolution and large ratio of long side to short side of double-wire pair, the backbone network structure of CenterNet is optimized and a global attention mechanism is introduced to improve the spatial resolution and representation ability of features. For detection of double-wire pairs, experimental results show that the average precision of the proposed method on the test set can reach 96.18%, and the detection accuracy of double-wire pairs with a detection error of less than 2 is as high as 97.5%, which can achieve the effective quality assessment of power equipment X-ray images.