Research on credibility evaluation method of infrared small target detection network based on YOLOv5

Abstract

At present, the credibility evaluation system of infrared small target detection network is not perfect. In this paper, we propose a set of credibility evaluation methods for infrared small target detection networks, including generalization evaluation and robustness evaluation. The generalization evaluation uses the original dataset to test the trained YOLOv5 network, and then obtains test indexes such as accuracy, precision, recall and so on. Robustness evaluation process the dataset with noise, which includes Gaussian noise, salt and pepper noise and random occlusion, test the trained network with the dataset after noise processing, obtain the robustness test index, and evaluate the anti-interference ability of the model.