Lightweight image water mist removing method for severe environment

The invention provides a lightweight image water mist removal method for a severe environment. The method comprises the following steps: S1, obtaining a training data set; s2, extracting shallow layer features of the image water mist removing network; s3, carrying out image feature extraction and local feature fusion splicing on the initialized features through image water mist removal network deep feature extraction, and finally, adding an output feature image to an initialized feature image output by a shallow feature extraction module; and S4, image water mist removal network image reconstruction: sending a feature map output by the deep feature extraction part to an output convolution layer containing three convolution kernels, and finally performing reconstruction to obtain a defogged image. According to the invention, by using a simplified neural network structure, parameter sharing, model pruning and other technologies, lightweight design of a deep learning water mist removal model is realized, and the