FSG⁃YOLO:基于YOLOv8改进的轻量化月面障碍物检测算法

FSG⁃YOLO:An Improved Lightweight Lunar Obstacle Detection Algorithm Based on YOLOv8

为了提高月面障碍物检测的效率和准确性,提出了一种基于YOLOv8改进的轻量化陨石坑和月面岩石检测算法FSG⁃YOLO。首先,采用FasterNet作为主干网络,提升模型对于陨石坑和月面岩石的特征提取能力,同时减少模型参数量和计算量,提升检测速度;然后,引入Si⁃mAM注意力机制模块,在不增加原始网络参数的同时提高模型的特征融合能力;最后,采用GIoU作为模型的损失函数,提高模型的识别准确率。实验结果表明:相较于原模型,FSG⁃YO⁃LO在自建月面障碍物数据集上的平均精度均值提升了4.0%,模型参数量减少了41.86%,计算量减少了38.27%,检测速度提高了19.99%。算法能够平衡精度和轻量化的需求,能有效适用于复杂空间环境下月面障碍物的检测。

To improve the efficiency and accuracy of lunar obstacle detection,a lightweight algo⁃rithm for detecting craters and lunar rocks,named FSG⁃YOLO,was proposed based on an improved YOLOv8.Firstly,the FasterNet was adopted as the backbone network to enhance the model’s fea⁃ture extraction capability for craters and lunar rocks,while simultaneously reducing the model’s pa⁃rameters and computational load to increase detection speed.Secondly,the SimAM attention mecha⁃nism module was introduced to improve the model’s feature fusion capability without increasing the original network parameters.In the end,GIoU was used as the model’s loss function to improve de⁃tection accuracy.Experimental results showed that compared to the original model,FSG⁃YOLO im⁃proved the mean average precision by 4.0%on a self⁃constructed lunar surface obstacle dataset,the number of model parameters was reduced by 41.86%,the computational load was decreased by 38.27%,and the detection speed was improved by 19.99%.This algorithm effectively can balance the accuracy and lightweight requirements,making it suitable for detecting lunar obstacles in com⁃plex space environments.