Research on Image Recognition Algorithm Based on Improved DBSCAN and Its Application in Augmented Reality Inspection System

Traditional substation inspection system has the problems of low inspection accuracy and low inspection efficiency. In this paper, an image recognition algorithm based on improved DBSCAN clustering is proposed. Firstly, the U-Net segmentation algorithm is adopted to preprocess the images. Then, the image feature points are matched by AKAZE algorithm. Finally, the adaptive parameter selection method and the concept of parameter density are utilized to improve the DBSCAN clustering algorithm, and the image features are input into the improved DBSCAN clustering algorithm for feature point clustering, and finally the image recognition is realized based on the clustering results. The experimental results show that the $F 1$ index value of the proposed recognition algorithm is 0.9672, which is 0.2174,0.1903 and 0.1556 higher than the traditional YOLOv4 algorithm, CURE parallelization algorithm and K-means algorithm, respectively. Moreover, the running time of the proposed algorithm is only 96 s, which is lower than the other three algorithms. In practical application, it is found that the proposed algorithm can improve the inspection accuracy and efficiency of the augmented reality substation inspection system, and the inspection effect is significantly improved.