Abstract: Abstract: To address the problems of missed detection and false detection in maritime ship detection tasks under complex weather environments such as rain, fog, and low light, a ship detection method based on an enhanced YOLOv8 algorithm for maritime complex weather environments was proposed. First, a Cross-Granularity Local Global Attention Fusion Block (CGLGAFB) was proposed. A refined local and global feature fusion mechanism was constructed and multi-path feature fusion strategies were combined to integrate multi-source feature information from different levels. The model's feature fusion capability was enhanced while noise interference and information redundancy were suppressed. Then, the original C2f (Faster Implementation of CSP Bottleneck with 2 convolutions) module was improved to an adaptive mixed C2f module (C2f-Adaptive Mixer Block, C2f-AMB). Through depth convolution branches with adaptive receptive field adjustment capabilities, target features of different scales and complex spatial structures can be captured more flexibly and efficiently, thereby enhancing feature extraction capabilities. Finally, a Multi-scale Spatial Perception Pyramid (MSPP) module was proposed to replace SPPF. Dilated convolutions with different dilation rates were utilized to construct multi-scale receptive fields, obtain comprehensive contextual information, and reduce the omission of key information. Experiments on the enhanced dataset SeaShips_aug show that the improved detection algorithm achieves mAP@50 and recall rates of 84.7% and 79.3% respectively, which are 2.6 and 3.9 percentage points higher than the baseline YOLOv8 model. The results prove that the enhanced model is more suitable for maritime ship detection tasks in complex weather environments.

Key words: Keywords: complex weather conditions, YOLOv8, deep learning, ship detection, target recognition

摘要： 摘 要: 针对海面船舶检测任务在雨雾、低光照等复杂天气条件下存在的漏检与误检问题，提出一种基于改进YOLOv8的复杂天气环境海面船舶检测算法。首先，提出跨层次局部与全局注意力融合模块(Cross-Granularity Local Global Attention Fusion Block, CGLGAFB)，通过构建精细化的局部和全局特征融合机制，结合多路特征融合策略整合来自不同层次的多源特征信息，提升模型特征融合能力，抑制噪声干扰与信息冗余；其次，改进原C2f(Faster Implementation of CSP Bottleneck with 2 convolutions)模块为自适应混合C2f模块(C2f-Adaptive Mixer Block, C2f-AMB)，通过有自适应感受野调节能力的深度卷积分支，使模型能够更灵活、更高效地捕获不同尺度与复杂空间结构的目标特征，增强特征提取能力；最后,提出多尺度空间感知金字塔（Multi-scale Spatial Perception Pyramid, MSPP）模块替换SPPF，利用不同空洞率的空洞卷积构建多尺度感受野，获取全面的的上下文信息，减少关键信息遗漏。在增强后的数据集SeaShips_aug上实验证明，改进检测算法的mAP@50和召回率分别达到84.7%和79.3%，比基线模型YOLOv8高2.6个百分点和3.9个百分点，证明改进模型更适合复杂天气环境下的海面船舶检测任务。

关键词: 关键词: 复杂天气环境, YOLOv8, 深度学习, 船舶检测, 目标识别