摘要
针对露天矿非结构化道路信息无法有效提取或提取精度不高的问题,提出一种基于改进DeepLabv3+网络融合SE注意力机制的露天矿道路识别方法,使用不同采样率的空洞卷积并行采样获取目标图像的高级特征。引入SE注意力模块对采样获取的高级特征和骨干网络提取的低级特征进行特征权衡,以区分不同特征的重要性,提高融合后特征信息的准确性。试验证明,该网络在矿山道路识别中优于其他算法,各项道路识别评价指标均得到提高,可有效识别非结构化的露天矿山道路。
Aiming at the problems of low extraction accuracy and inability to effectively extract unstructured road information when extracting unstructured road such as open-pit mine,an open-pit road recognition method based on improved DeepLabv3+network fusion SE attention mechanism was proposed,and cavity convolution parallel sampling with different sampling rates was used to obtain advanced features of target images.SE attention module was introduced to carry out feature balance between high-level features obtained from sampling and low-level features extracted from backbone network,so as to distinguish the importance of different features and improve the accuracy of feature information after fusion.The experimental results show that the network is superior to other algorithms in mine road recognition,and all the evaluation indexes of road recognition are improved,which can effectively identify the unstructured open-pit mine road.
作者
金磊
杨晓伟
张浩
杜勇志
李新鹏
戴春田
JIN Lei;YANG Xiaowei;ZHANG Hao;DU Yongzhi;LI Xinpeng;DAI Chuntian(CHN Energy Baorixile Energy Co.,Ltd.,Hulunbuir 021000,China)
出处
《煤炭工程》
北大核心
2024年第7期200-204,共5页
Coal Engineering
