Research on Dim Target Detection Model Based on Eye Movement Data

Dim targets detection under complex backgrounds is a classic problem in the field of automatic target recognition (ATR). The human visual attention mechanism can effectively avoid background clutter interference, which provides a new idea for solving this problem. Traditional visual attention models are limited in practice due to task definition, computational accuracy and other reasons. This work focuses on dim object detection and high-precision visual attention modeling by using deep neural network. Firstly, the head-eye movement measurement technology is used to conduct visual attention experiments and build a complete human visual dataset. And then, a visual attention calculation model with complete visual processing capabilities is conducted based on the generative adversarial network (GAN) framework. Finally, based on the attention model, a new algorithm for small object detection driven by visual attention is proposed. This detection method can achieve a detection probability of 0.78 and a false alarm rate of 0.19×10-4 on the test dataset, which proves its effectiveness. Therefore, the research has reference significance for promoting the development of ATR-related technologies.