基于U-Net++的烟草种植信息提取

Tobacco Planting Information Extraction Based on U-Net++

以洛宁县为研究区,高分二号为数据源,利用U-Net++模型对研究区烟草进行提取。结果表明:基于U-Net++模型烟草提取的准确率、F1分数和Kappa系数相比于原始的U-Net模型提取精度分别从86.33%、73.86%和0.65提高到91.85%、80.13%和0.82。综上可知,本文提出的U-Net++网络模型在烟草提取中是可行的,为烟草信息的快速提取提供了技术支持。

This paper takes Luoning county as the study area and GF-2 satellite images as the data source,and uses U-Net++model to extract tobacco in the study area.The results showed that the accuracy,F1 score and kappa coefficient of tobacco extraction based on U-Net++model were improved from 86.33%,73.86%and 0.65 to 91.85%,80.13%and 0.82 respectively as compared with original U-Net model.To sum up,the U-Net++network model proposed in this paper is feasible in tobacco extraction,and provides technical support for the rapid extraction of tobacco information.