Defect Detection Algorithm of Anti-vibration Hammer Based on Improved Cascade R-CNN

Aiming at the problem that it is difficult to accurately locate and identify the defects of anti-vibration hammer components in high-voltage transmission lines, this paper propose a detection method for anti-vibration hammer defects based on the improved Cascade R-CNN algorithm. In dataset: Firstly, this research construct a dataset of anti-vibration hammer defects based on common anti-vibration hammer defect categories; secondly, this research perform preprocessing methods such as cropping, flipping, gamma transformation and CLAHE on training samples to improve the generalization ability of the network and avoid over-fitting. In algorithm: This research use ResNeXt-101 as the backbone network of the Cascade R-CNN algorithm; add FPN module for extracting multi-scale features to extract more effective information; use Focal Loss function to improve the classification loss of RPN module to solve the dataset category imbalance problem. Experimental results show that the improved Cascade R-CNN algorithm has a detection accuracy of 91.2% on the anti-vibration hammer defect test set, which is 3.5% higher than the original Cascade R-CNN algorithm and is better than other mainstream object detection algorithms.