EFEFMamba-YOLO:SAR图像中船舶目标检测OA

To address the challenges in ship target detection from Synthetic Aperture Radar(SAR)images,such as small target size,dense distribution,blurred features,and complex background interference,we propose an improved model based on Mamba-YOLO,namely EFEFMamba-YOLO(Enhanced Feature Extraction and Fusion Mamba-YOLO).Firstly,to enhance feature extraction capability,a Collaborative Feature Enhancement Block(CFEBlock)is designed in the backbone network of Mamba-YOLO,which can effectively capture the local and global feature dependencies.Secondly,to tackle the problem of easy loss of detailed information during feature fusion,a Channel-Enhanced Residual Spatial Pyramid Pooling Fast(CResSPPF)module is developed at the end of the backbone network.This module preserves detailed information through a residual structure and improves feature expression ability by means of a channel expansion strategy.Finally,a Four-level Adaptive Structure Feature Fusion(FASFF)detection head is designed,which effectively enhances the fusion effect of features at different levels.Experiments are conducted on the HRSID,SSDD,and LS-SSDD-v1.0 datasets.The results show that the mAP50 of EFEFMamba-YOLO reaches 94.1%on HRSID,an improvement of 1.9 percentage points compared with the baseline Mamba-YOLO model.On the SSDD and LS-SSDD-v1.0 datasets,the mAP50 values of the EFEFMamba-YOLO model reach 98.9%and 75.6%,respectively.Experimental results demonstrate that the EFEFMamba-YOLO model exhibits excellent effectiveness and reliability in ship target detection from SAR images.