Fish sonar image denoising method and system based on neural style migration

Abstract

The invention provides a fish sonar image denoising method and system based on neural style migration. The method comprises the following steps: acquiring a fish sonar image and a reference style image; inputting the fish sonar image and the reference style image into a pre-trained denoising model; wherein an encoder, a self-adaptive style attention normalization module and a decoder are arranged in the denoising model; the fish sonar image is converted into a content feature map through the encoder, and the reference style image is converted into a style feature map; based on different dimensions, inputting the content feature map and the style feature map into the adaptive style attention normalization module to generate a fusion feature map of the corresponding dimension; and inputting the fusion feature maps of different dimensions into the decoder, and generating a de-noised fish sonar image, thereby solving the problem of poor imaging quality of the existing fish sonar image, realizing the effect of removing speckle noise and environmental noise in the sonar image through the style migration capability of the neural network, retaining detail textures and edge features of the fish image, and improving the image quality of the fish sonar image. And a high-quality sonar image denoising effect is realized.

Images