Multi-feature fusion ship target detection based on improved YOLOv5s

In recent years, with the rapid development of artificial intelligence, deep learning has also been widely used in the field of computer vision. At the same time, synthetic aperture radar (SAR) has also become a research hotspot in the field of computer vision because of its all-day, all-weather advantages. However, in view of various complex needs in practical engineering, the current target recognition algorithm still has great limitations. In this paper, a method is proposed to enhance the contrast between the target region and the clutter region by fusing various features in the image input, so as to improve the detection performance of ship targets in SAR images. Specifically, this paper realizes multi-feature fusion by placing the SAR image after noise reduction, the original SAR image and the CFAR feature map in the three channels of the RGB image respectively. The test results show that when the improved YOLOv5s detector is used to detect the target, the indicators of the multi-feature image after fusion are significantly improved compared with the detection of each channel information alone. Compared with the original SAR image, the detection accuracy rate of multi-feature fusion image is increased by 4.3%, and the recall rate is increased by 5.5%. mAP0.5 increased by 4.5% and mAP.5:.95 increased by 6.5%, which effectively met the application requirements in practical projects.