Water Body Recognition From Remote Sensing Based On Multi-Task Semantic Image Segmentation

Water is not only a basic condition for human survival,but also essential resource for social production.With the change of global climate and the expansion of human production activities,the shortage of water has become increasingly prominent.It is of great significance to timely and accurately identify the coverage of surface water bodies and detect changes in water resources for many research directions such as the investigation of water resources,water pollution,flood monitoring,etc.The current water body recognition technologies proposed by scholars are mostly focused on obtaining corresponding indexes by analyzing the spectral properties of water bodies or using classification techniques of machine learning and deep learning to implement automatic water body identification.These methods are not only difficult to achieve high degree of pixel-level accuracy for water body recognition,but also can’t make full use of the multi-spectral information of satellite remote sensing data.This paper proposes an efficient method to recognize the water body based on image semantic segmentation with multi-task learning.On the one hand,it solves the problem of how to efficiently use multi-channel information from remote sensing based on existing image processing technologies by multi-task learning.On the other hand,it identifies recognition of water body as a semantic segmentation task to avoid errors from classifying the water boundary caused by object-level classification techniques and reach a new height of accuracy.The multi-level experimental results show that the proposed method for water recognition is significantly better than other compared methods.