基于改进DeepLabv3+和高分辨率影像的露天矿区土地覆盖分类

Land Cover Classification of Open-pit Mining Areas Based on Improved DeepLabv3+and High-resolution Images

深度学习技术和高分辨率遥感影像为矿区的土地覆盖分类提供了高效的技术手段,针对原始DeepLabv3+深度学习模型在矿区土地覆盖分类中的地物边缘分割不准确和错分、漏分等问题,本文通过在编码层进行特征融合、骨干网络添加监督分支模块和解码层融合浅层特征的方式对原始的DeepLabv3+模型进行改进。实验结果表明,改进后的DeepLabv3+模型的mIoU值平均为80.32%,相比原始的DeepLabv3+模型提升了5.71%,能够有效地进行露天矿区的土地覆盖分类。

Deep learning technology and high-resolution remote sensing images provide efficient technical means for the classification of land cover in mining areas.Aiming at the problems of inaccuracy,omission and commission of feature boundary segmentation in the classification of land cover in mining areas,this paper improves the original DeepLabv3+deep learning model by means of feature fusion in coding layer,adding supervision branch module in backbone network and merging shallow features in decoding layer.The experimental results show that the mIoU value of the improved DeepLabv3+model is 80.32% on average,which is 5.71% higher than that of the original DeepLabv3+model.It can effectively classify the land cover of open-pit mining areas.