【摘要】 针对无人机海上人员搜救任务中复杂环境下目标检测精度与实时性的需求，对改进YOLOv11算法进行了研究；通过结合风车状卷积优化网络主干，设计特征增强模块(FEM)与自适应权重的双向特征金字塔网络(BiFPN),并引入动态注意力机制，实现了对海上人员微小目标及遮挡目标的特征增强与噪声抑制；采用SeaDronesSee数据集进行实验分析，测试结果表明，改进后模型的检测精度(mAP@0.5)达到78.47%,推理速度(FPS)为511.79 Hz,优于传统的YOLO系列算法；经实际应用验证，该算法能够满足海上搜救任务的高精度与实时性要求，为智能化应急救援提供了有效技术支持。更多还原

【Abstract】 In response to the detection accuracy and real-time performance of targets in complex environments for drone-based maritime search and rescue missions, research has been conducted on improved YOLOv11 algorithm. By integrating the pinwheel-shaped convolution to optimize the network backbone, designing a feature enhancement module(FEM) and a bidirectional feature pyramid network(BiFPN) with adaptive weights, and introducing a dynamic attention mechanism, the feature enhancement and noise suppression for small targets of maritime personnel and occluded targets are achieved. Experiments are conducted on the SeaDronesSee dataset, and the results indicate that the improved model achieves a detection accuracy(mAP@0.5) of 78.47% and an inference speed of 511.79 FPS, outperforming traditional YOLO series algorithms. Practical applications show that this algorithm can meet the high-precision and real-time requirements of maritime search and rescue missions and provide an effective technical support for intelligent emergency rescue.更多还原