现代电子技术

针对合成孔径雷达图像舰船目标检测算法在精度、计算效率和模型复杂度之间难以兼顾等问题，提出一种基于改进YOLOv8的SAR图像舰船目标检测算法。首先，在Head部分增加一个P2检测头，提高对小尺度目标的检测能力；其次，在C2f模块中引入增强的多尺度通道感知结构，以增强特征表达能力并优化多尺度目标检测效果；同时，在检测头前增加卷积注意力模块，提升模型对关键特征的关注度；此外，采用Ghost轻量化卷积以减少计算量，提高模型推理速度。在HRSID上的实验结果显示：相较于原始YOLOv8，改进后的算法在SAR图像舰船目标检测平均精度均值（mAP）上提升了2.8%、召回率（R）提升了4.2%，检测速度（FPS）提高了27.1 f/s、计算量GFLOPs降低了25.17%。与RCSA⁃YOLO相比，虽然计算量略微增加，但文中算法的mAP值高出4.7%，准确率也高于RCSA⁃YOLO；与其他算法相比，文中算法在保证较高检测精度的情况下大幅降低了模型参数量和计算量，提高了检测效率。实验结果表明，改进后的YOLOv8算法较好地兼顾了检测精度、检测效率和模型复杂度，对复杂背景下的SAR小尺度舰船检测具有较高的实用价值，可为海上监视与港口安防等实时应用提供支持。

Since the synthetic aperture radar (SAR) image ship target detection algorithm is difficult to balance between accuracy, computational efficiency and model complexity, an SAR image ship target detection algorithm based on improved YOLOv8 is proposed. Firstly, a P2 detection head is added to the part of the Head to improve the detection ability of small⁃scale targets. Secondly, the enhanced multi⁃scale channel perception (EMSCP) structure is introduced into the C2f module to enhance the feature expression ability and optimize the multi ⁃ scale object detection effect. The convolutional block attention module (CBAM) is added in front of the detection head to improve the model's attention to key features. In addition, Ghost lightweight convolution is used to reduce the computation burden and improve the inference speed of the model. The experimental results on HRSID (high⁃resolution SAR images dataset for ship detection) show that in comparison with the original YOLOv8, the mean average precision (mAP) of ship target detection of the improved algorithm is improved by 2.8%, its recall rate is increased by 4.2%, its detection speed FPS is increased by 27.1 f/s, and its computation burden GFLOPs is reduced by 25.17%. In comparison with RCSA⁃YOLO, although the computation burden of the proposed algorithm is slightly increased, its mAP is 4.7% higher, and its accuracy is also higher than that of RCSA⁃YOLO. In comparison with the other algorithms, the proposed algorithm reduces the number of model parameters and computation burden greatly and improves the detection efficiency while ensuring high detection accuracy. Experimental results show that the improved YOLOv8 algorithm achieves a balance between detection accuracy, detection efficiency and model complexity, so it has high practical value for SAR small⁃scale ship detection in complex background, and can provide support for real⁃time applications such as maritime surveillance and port security.