Haze image restoration method based on deep learning and gamma correction

The invention discloses a haze image restoration method based on deep learning and gamma correction. The method comprises the following steps: taking an image restoration architecture Uform based on transform as a generator of a generative adversarial network; inputting the foggy image into a generator of the adversarial network to generate a defogged picture; the discriminator compares the defogged picture generated by the generator with a real picture without fog, and optimizes generative adversarial network model parameters by minimizing a Charbonnier loss function; and repeating the above operations until the number of iterations set by training, adding a gamma correction function in the forward propagation reasoning process of the trained generator, and performing power transformation on the image data to adjust the nonlinear response so as to realize more accurate light intensity expression. According to the method, a gamma correction image is divided into RGB channels, each channel is multiplied by a c