Deep learning based extraction method for coastal wetland salt pond culture pond remote sensing image

Abstract

The deep learning-based coastal wetland salt pond aquaculture pond remote sensing image extraction method provided by the application is based on deep learning and fully considers the local-global information correlation to propose a local-global network LGDSNet coupled with a strip hollow convolution, so as to be applicable to high-resolution salt pond aquaculture pond remote sensing image extraction, capable of acquiring local and global information of the salt pond aquaculture pond, thereby improving the extraction precision of the salt pond aquaculture pond under high-resolution images, and finally applying the trained model to a salt pond aquaculture pond semantic segmentation task. The method comprises the following steps: step one, data set construction; step two, network architecture construction; the remote sensing image data set is trained by using the local-global neural network LGDSNet coupled with a multi-scale mixed strip convolution as a basic model, the network architecture of the LGDSNet model comprises an encoder, an intermediate part and a decoder; step three, salt pond aquaculture pond extraction; based on the trained LGDSNet extraction model, the extraction of the coastal wetland salt pond aquaculture pond remote sensing image is realized.

Images