A Target Detection Algorithm of Aerial Images in Power Grid Inspection Based on Transfer Learning

Abstract

When the traditional algorithm is applied to detect a target in aerial images during power grid inspection, it is difficult to carry out optical correction due to the great influence of the noise of aerial images. In the target coincidence rate range of 29.56%–49.56% of aerial images, there is the problem of high recall rate. Therefore, this paper proposes a target detection algorithm for aerial images in power grid inspection based on transfer learning. First of all, the aerial images are preprocessed, which involves optical correction, image restoration, geometric correction and so on. Then the SIFT algorithm based on the Gaussian scale space theory is used to extract the target feature points of the aerial images. The main implementation process involves: establishing the boundaries of the scale space, the feature point location, the feature point orientation, and the feature point descriptor generation. Based on the transfer learning algorithm, the detection of targets in aerial images in power grid inspection is realized, and the target detection algorithm is completed. In order to prove that the algorithm has a low recall rate in the range of 29.56%–49.56% of the target coincidence rate of the aerial images in power grid inspection, compared with the original algorithm, the experimental results show that the recall rate of the algorithm is always lower than that of the other two algorithms, and the performance is better.

Keywords: transfer learning, aerial images in power grid inspection, target detection algorithm, optical correction

Cite As

F. Wang, L. Shen, W. Li, "A Target Detection Algorithm of Aerial Images in Power Grid Inspection Based on Transfer
Learning", vol. 30 no. 5, 341-351, 2022.