基于Dual-YOLOv7模型的在轨卫星遥感图像船舶检测

在海洋监测领域，对遥感影像中的船舶进行实时、精准的目标检测是确保海上交通安全和海洋环境监管的重要任务。本文提出了一种基于双YOLOv7 （Dual You Only Look Once v7，Dual-YOLOv7）模型的船舶检测方法，旨在提高遥感影像中船舶检测的效率和准确性。引入交叉注意力融合（Criss-Cross Attention Fusion，CC Attention-Fusion）模块，并对随机融合（Shuffle-Fusion）模块进行改进，以有效融合红外图像与可见光图像的特征，从而增强模型对不同光照和天气条件下的适应能力。同时，引入深度可分离卷积模型，构造深度可分离空间金字塔池化快速（Depthwise Separable Spatial Pyramid Pooling Fast，D-SimSPPF）模块，在保证检测精度的同时，显著提升了检测速度。此外，构造了CIoU-C （Complete Intersection over Union-Complement）损失函数，使得目标框尺寸的选定更加贴近真实船舶的尺寸，进一步提高了检测的准确性。为了验证模型的实际应用能力，本文将Dual-YOLOv7 模型部署在AGX Orin 嵌入式平台上，并在多模态船舶MMShip 数据集上进行了训练和测试。实验结果表明：与现有的掩码-区域卷积神经网络（Mask Region -Convolutional Neural Networks，Mask R -CNN）、YOLOv8 分割（YOLOv8-seg）模型、YOLOv5 分割（YOLOv5-seg）模型、YOLOv7 分割（YOLOv7-seg）模型和实时检测变换器（Real-time Detection Transformer，RT-DETR）相比，Dual-YOLOv7 模型的平均精度（mean Average Precision 50%，mAP50%）指标达到了91.6%，展现出最高的检测精度。同时，该模型的总计算量仅为1176 亿次浮点运算，即117.6 GFLOPs （Giga Floating-Point Operations Per Second），算力消耗最低，表明其在实际应用中具有显著的效率优势。结果表明：Dual-YOLOv7模型在船舶检测任务中具有良好的应用前景，能够为海洋监测和管理提供有力的技术支持。

英文摘要:

In the field of maritime monitoring, real -time and accurate detection of ships in remote sensing images is a crucial task for ensuring maritime safety and regulating the marine environment. This paper proposes a ship detection method based on the Dual-YOLOv7 model, aiming to enhance the efficiency and accuracy of ship detection in remote sensing images. The study introduces the CC Attention-Fusion module and improves the Shuffle -Fusion module to effectively fuse the features of infrared and visible light images, thereby strengthening the model’s adaptability to different lighting and weather conditions. Meanwhile, the depthwise separable convolution model is introduced, and the D-SimSPPF module is constructed to significantly improve detection speed while ensuring detection accuracy. In addition, the CIoU-C loss function is constructed to make the selection of bounding box sizes closer to the actual size of ships, further enhancing detection accuracy. To verify the practical application capability of the model, the Dual-YOLOv7 model was deployed on the AGX Orin embedded platform and trained and tested on the MMShip dataset. Experimental results show that compared with existing object detection networks such as Mask R-CNN, YOLOv8-seg, YOLOv5-seg, YOLOv7-seg, and RT-DETR, the Dual-YOLOv7 model achieved the highest detection accuracy with an mAP50% of 91.6%. At the same time, the model’s total computational cost is only 117.6 GFLOPs, with the least computational consumption, indicating a significant efficiency advantage in practical applications. These results demonstrate that the Dual-YOLOv7 model has a promising application prospect in the task of ship detection in remote sensing images and can provide strong technical support for maritime monitoring and management.

查看全文查看/发表评论下载PDF阅读器