Using WorldView-2 high-resolution remote sensing images,this study proposed a SS-UNet network model for raft aquaculture area information extraction.The SS-UNet model introduces the SPM module and the SA module on the"U"structure of the U-Net model,taking into account the shape characteristics of the raft aquaculture area.The test results showed that the SS-UNet model achieved the highest level of accuracy,in which the MIoU and Kappa coefficient reached 91.72％and 0.912 3,respectively.And the model could extract the raft aquaculture areas more accurately,with less phenomena such as omission and misclassification.Compared with the U-Net model,the MIoU and Kappa coefficient of the SS-UNet model were improved by 10.41％and 0.126,respectively,with a very small increase in model complexity.The results showed that the SS-UNet model realized the effective improvement of the result accuracy and extraction performance of raft aquaculture areas in offshore waters in high-resolution remote sensing images.
维普
万方数据