Ship-YOLOv8:一种轻量级高分辨遥感图像船舶细粒度检测算法

Ship-YOLOv8:A Lightweight Ship Fine-grained Detection Algorithm of High-resolution Remote Sensing Images

针对高分辨率影像的船舶细粒度目标检测分类任务中类内差异大、类间相似性高、物体和场景的尺度变化范围大、特征提取困难、样本小等特点,提出了一种基于YOLOv8为基础的改进算法。首先,在骨干网络中引入SimAM注意力机制,使得模型在复杂背景中更加聚焦船舶对象;其次,在颈部引入SPD-Conv模块,改善复杂背景下船舶尺度变化大和小目标检测的问题;最后针对细粒度船舶目标检测的特点,替换Mish激活函数和Focal-Loss损失函数,加快模型收敛,提高模型精度。经对比实验可知,改进的算法在保证检测速度和模型参数量的同时,在FAIR1M_Ship数据集取得了94.49%的检测精度,与目前流行的目标检测算法相比,在检测精度上有一定的提升。

Aiming at the characteristics of large intra-class differences,high similarity between classes,large scale changes of objects and scenes,difficulty in feature extraction,and small samples in the ship fine-grained target detection and classification task of high-resolution images,an improved algorithm based on YOLOv8 is proposed.Firstly,the SimAM Attention Mechanism is introduced into the backbone network to make the algorithm model more focused on the ship object when running in the complex background.Secondly,the SPD-Conv module is introduced in the neck to improve the problems of large ship scale changes and small target detection in complex backgrounds.Finally,for the characteristics of fine-grained ship target detection,it replaces the Mish activation function and Focal-Loss loss function to speed up model convergence and improve model accuracy.Comparative experiments show that the improved algorithm achieves a detection accuracy of 94.49%in the FAIR1M_Ship dataset while ensuring the detection speed and number of model parameters.Compared with the currently popular target detection algorithms,the detection accuracy of the improved algorithm has been improved to a certain extent.