Research On Vegetation Extraction Of High Resolution Remote Sensing Image Based On Attention Model

Remote sensing image classification is an important way to obtain vegetation information.Our series of high remote sensing satellites and resource satellites successful launched greatly enrich the high resolution remote sensing date.Based on high-resolution remote sensing images,this study establishes a training sample set for extracting typical terrain features of vegetation,and construct vegetation extraction system based on domestic high-resolution images,attention models and semantic segmentation network,thus Automatically extract the typical vegetation in high resolution satellite remote sensing images.The main research content of this study as the follows:(1)The study establish a vegetation extraction training sample set based on high-resolution images.The sample set is the basis for the automatic extraction of vegetation elements.By calculating the characteristics of spectrum,texture,index,etc.,a sample set containing image-element annotation-features is constructed.(2)A high-resolution remote sensing image vegetation extraction model based on the attention model is designed to realize the end-to-end automatic extraction of different types of vegetation areas of high-resolution images,and design attention based on the pixel-level constraints of the encoding and decoding network.The model fully learns the spectral spatial texture information and deep semantic information of the images.(3)Through comparative experiments between Vietnam and Changsha,the overall classification accuracy of this method is 93.49% and 92.68%,and the Kappa coefficient of this method is 0.8503 and 0.8671.Through visual interpretation and quantitative calculation analysis,the attention model network proposed in this study has effectiveness and generalization ability,which is more consistent with the needs of practical applications of vegetation extraction.