土地覆盖分类是全球变化研究的核心,精准分类是开展土地覆盖分类变化研究的基础.基于此,本文以Deeplab-v3+模型为基础,在编码阶段改进ASPP模型,并采用Xception-ResNet结合的网络结构,解码阶段引入跃层特征融合优化模块,得到改进后的Deeplab-v3+模型,对研究区进行土地覆盖分类提取.结果表明:改进后的模型分类精度为82.93％,训练速度为7 h 22 min,相比原始模型分别提高了4％,模型训练速度提升了25％.综上可知,改进后的Deeplab-v3+模型可以实现快速且高精度的土地覆盖分类,可为土地覆盖分类研究提供技术支持.
Land Cover Classification Based on Improved Deeplab-V3+
Land cover classification is the core of global change research, and accurate classification is the basis of land cover classifi-cation change research. Based on this, based on the Deeplab-v3+ model, this paper improves the ASPP model in the coding stage, a-dopts the network structure combined with Xception-ResNet, introduces the thermocline feature fusion optimization module in the de-coding stage, obtains the improved Deeplab-v3+ model, and extracts the land cover classification of the study area. The results show that the classification accuracy of the improved model is 82.93％ and the training speed is 7 h 22 min, which is 4％ higher than the o-riginal model and 25％ faster than the original model. To sum up, the improved Deeplab-v3+ model proposed in this paper can realize fast and high-precision land cover classification and provide technical support for land cover classification research.
deep learningsemantic segmentationGF-2land classification
万方数据
维普
