遥感尾矿库影像检测的深度学习算法研究  

Research on deep learning algorithm for remote sensing tailings storage image detection

作　　者：党政军 李琨[1] DANG Zhengjun;LI Kun(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650504,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,云南昆明650504

出　　处：《陕西理工大学学报(自然科学版)》2025年第1期50-60,共11页Journal of Shaanxi University of Technology:Natural Science Edition

摘　　要：针对当前尾矿库自动检测的局限性,提出一种基于深度学习模型的尾矿库自动检测方法。首先,将基于Transformer和ViT的RepViT作为YOLOv8的主干特征提取网络;其次,将包含注意力机制的C2f_SENetV2模块代替原始的C2f模块;然后,采用WIoUv3代替原始的CIoU作为损失函数;最后,将模型在构建好的尾矿库数据集进行实验验证。结果表明,改进的YOLOv8模型相比原模型在尾矿库的检测上准确率、召回率以及平均准确率等性能指标有显著提升;方案有助于实现尾矿库的自动化识别检测。To address the limitations of the current automatic tailing pond detection method,a deep learning model-based automatic tailing pond detection method is proposed.Firstly,RepViT,which is based on Transformer and Vision Transformer(ViT),is used as the backbone feature extraction network of YOLOv8.Secondly,the C2f_SENetV2 module,which contains an attention mechanism,is used instead of the original C2f module.Thirdly,WIoUv3 is used instead of the original CIoU as the loss function.Finally,the model is validated experimentally on the constructed tailing pond dataset.The experimental results demonstrate that the improved YOLOv8 model markedly enhances the performance metrics,including accuracy,recall,and average precision,in the detection of tailing ponds when compared to the original model.The proposed scheme in this paper facilitates the automated identification and detection of tailing ponds.

关 键 词：尾矿库 YOLOv8 目标检测 注意力机制 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]