改进U-Net网络的遥感图像水质分割算法

Water Quality Segmentation Algorithm of Remote Sensing Image Based on Improved U-Net Network

针对遥感图像中一般水体、黑臭水体以及富营养化水体形状不规则以及相似难以准确分割的问题,选取研究区创建富含3类水体的数据集,利用深度学习卷积网络模型对3类水体数据集进行训练与测试。根据测试效果分析提出基于改进U-Net网络的遥感水质分割算法,改进卷积深度及在编码阶段输入层引入ASPP模块获取更加复杂的光谱信息,提高分割精度。实验表明,所提出的改进型U-Net分割算法能够显著提升水质分类的精确度和分割效果,从而实现一般、黑臭及富营养化水体的准确分类。

Aiming at the problem of irregular shape and similarity of general water body,black and smelly water body and eutrophic water body in remote sensing images,the study area was selected to create data sets rich in three kinds of water bodies,and the deep learning convolutional network model was used to train and test the three kinds of water data sets.According to the test effect analysis,a remote sensing water quality segmentation algorithm based on improved U-Net network is proposed.The convolutional depth is improved and ASPP module is introduced in the input layer of coding stage to obtain more complex spectral information and improve the segmentation accuracy.Experimental results show that the improved U-Net segmentation algorithm proposed in this paper can significantly improve the accuracy and segmentation effect of water quality classification,so as to achieve the accurate classification of general,black,smelly and eutrophic water bodies.