A Study on Enhancing IR Small Target Detection Performance Using Sequence Data With the YOLOX Object Detection Model

This study presents a method to improve the infrared small object detection performance of the YOLOX object detection model. By using sequential frames as input through input-level fusion, we improved detection performance compared to YOLOX, a single-image-based detection technique. Furthermore, we enhanced detection performance by adding high-resolution layers to the detection stage to enable the network to identify small objects. These results demonstrate the validity of using sequence data in deep learning-based object detection, and the meaningful conclusion that adding high-resolution layers to deep learning models is beneficial for small object detection in infrared images. This provides meaningful directions for future research in the development of infrared small object surveillance technology.