Infrared dim and small target extraction algorithm based on improved U-Net

Single-frame infrared small target detection has always been a hard nut due to the poor imaging effect ascribed its background, which is much affected by detector accuracy and the noise of surrounding environment, and difficulties in realizing feature extraction of too little information contained in object. According to traditional method, image preprocessing work is burdensome under a complex scenario, with many control parameters controlled, resulting in a high false warning rate and low detection accuracy. An infrared small target extraction algorithm based on improved UNet was proposed in this paper on the basis of the features of infrared small targets. Based on the U-Net semantic segmentation network, the algorithm could separate single-category targets in input image from the background to finally extract target. The detection accuracy finally improves a lot by optimizing processing flow and introducing an efficient channel attention mechanism. In comparison to other detection algorithms based on semantic segmentation, the algorithm in the paper is more accurate in the test of single-frame infrared small target (SIRST) data set, making IoU value reach 76.4%. The experimental results show that the algorithm is superior in detecting single-frame infrared small target.