一种改进的YOLOv7光学遥感图像舰船目标检测算法

An Improved YOLOv7-Based Ship Target Detection Algorithm for Optical Remote Sensing Images

针对YOLOv7算法在应用于光学遥感图像舰船目标检测任务时所面临的小目标检测精度低的问题,提出一种基于多维注意力机制动态卷积的光学遥感图像舰船目标检测改进算法。首先,通过并行策略设计了融合多维注意力机制动态卷积的高效聚合网络,多维注意力机制动态卷积根据不同维度的特征重要性进行自适应调整,卷积核沿4个维度学习到注意力分布,增强了特征融合网络捕获数据中细粒度特征的能力;其次,针对舰船目标的多尺度差异特点设计多层次超大卷积核层,丰富全局特征描述,提高检测网络的感知能力。在HRSC2016和DOTA两个公共数据集上的实验结果表明,改进后算法的mAP分别达到了93.4%和90.1%,与现有主流先进算法相比取得更优的检测精度,在降低舰船小目标漏检误检率的同时提升了识别能力。

Aiming at the problem of low accuracy in small target detection faced by the YOLOv7 algorithm when applied to optical remote sensing image ship target detection tasks,an improved optical remote sensing image ship target detection algorithm based on multi-dimensional attention mechanism dynamic convolution is proposed.Firstly,an eficient aggregation network integrating multi-dimensional attention mechanism dynamic convolution is designed through parallel strategy.The multi-dimensional attention mechanism dynamic convolution adaptively adjusts the importance of features in different dimensions,and the convolution kernel learns attention distribution along four dimensions,enhancing the ability of the feature fusion network to capture fine-grained features in data;Secondly,a multi-level super large convolution kernel layer is designed based on the multi-scale differences of ship targets,enriching the global feature description and improving the perception ability of the detection network.The experimental results show that:1)The improved algorithm achieves mAP of 93.4%and 90.1%respectively on two public datasets of HRSC2016 and DOTA;and 2)Compared with existing mainstream advanced algorithms,it achieves higher detection accuracy and improves recognition ability while reducing the missed detection and false detection rate of small ship targets.