Research Demonstration of Positive and Negative Sample Classification Algorithm for Insulator Multi-Angle Target Detection Based on Faster-RCNN

Abstract

Aiming at the problem that the traditional Faster RCNN target detection algorithm fails to reflect the real shape of the insulator from multiple angles and the horizontal IOU algorithm cannot classify positive and negative samples from multiple angles, an improved insulator target detection algorithm based on the traditional Faster-RCNN target detection algorithm is proposed. In this algorithm, the angle parameter is added to the horizontal and vertical rectangular detection box, and the parameter is added as a training parameter to the loss function for iterative regression, which can effectively solve the problem that the conventional Faster-RCNN target detection algorithm cannot correctly reflect the insulator shape. After adding the angle parameters, the horizontal IOU algorithm is no longer suitable for the classification of positive and negative samples. Therefore, the Euclidean norm positive and negative sample classification algorithm is designed to solve this problem. The experimental analysis shows that the multi-angle detection frame with the addition of Angle parameter and the replacement of the IOU algorithm can accurately reflect the shape of insulator targets in any angle direction. Compared with the AP (46.87%) of the horizontal IOU algorithm and the AP(65.00%) of the inclined IOU algorithm, the AP(75.06%) of the proposed algorithm has better comprehensive performance.