New Water Extraction Method From Landsat Remote Sensing Image Based on Band Characteristics

It is difficult to determine the optimal threshold value when extracting lake water from the existing index model, and the extraction accuracy is affected by the omission and wrong extraction. In order to solve the problem, this paper proposes a new water extraction algorithm based on the band characteristics of Landsat remote sensing image. Firstly, six common bands in Landsat 7 ETM+ and Landsat 8 OLI were selected and the spectral sensitivity of two types of water bodies are analyzed; Secondly, five types of typical ground objects are determined by their differences of reflectance spectrum on the six common bands. Then, the water extraction model is determined by the combination of the spectrum value of the common sensitive bands of water and five typical ground objects. Finally, the fused spectral information is established by computing the weight value of the spectrum values in the model. The binary image is generated by binary segmentation of water body and threshold 0 to get the water extraction. The experiments show that our method has higher extraction accuracy, and can extract water with the stable threshold 0, avoiding the problem how to determine the optimal threshold.