Research on the detection algorithm of near-coastal ships in SAR images based on improved YOLOv5

During the detection of ship targets in SAR images, due to the complex background problems in near-coastal ports, the feature information of overlapping ship targets cannot be accurately extracted, resulting in missed detection and false detection of near-coastal ship targets. Aiming at the above problems, this paper proposes a SAR image ship detection algorithm in complex scenes. The algorithm is improved based on YOLOv5, uses the SPPF structure to enhance the extraction of feature information, and fuses the feature information extracted by the SPP structure of the original YOLOv5. This multi-level pyramid The method of parallel fusion of modules can effectively detect multi-scale ship targets, so that the feature information can be better expressed; then the GIOU in the original model is improved to CIOU, so that it can accurately return to the position of the prediction frame; finally, in order to be more reasonable The prediction box above the threshold is screened, NMS (Non-Maximum-Suppression) is improved, and the Soft-NMS method is used to penalize and decay the box score higher than the threshold, and reasonably remove the prediction box. The experimental results show that the mAP (mean Average Precision) of the improved model on SSDD and SAR-Ship-Dataset data sets is improved by 5.15% and 5.06% compared with the original model, and the improved model can effectively detect the complex background in the near coast. SAR image below the ship target.