Infrared small target detection method based on multi-scale feature fusion

Abstract Infrared small target detection is widely used in aerial and tracking system because of its long distance strong anti-jamming ability. Aiming at the shortcomings current infrared algorithm, such as low precision rate high false alarm when dealing with complex background. In this paper, we propose an end-to-end model (called MFSSD) based on multi-scale feature fusion. Considering trait targets, a fusion module by using refinement map method, improve correlation different channels through SP Module. The experimental results three sequences image show that average accuracy MFSSD algorithm 87.8%. Compared traditional both recall have been significantly improved.