Double-branch low-illumination image enhancement method based on Retinex theory

The invention relates to the field of image enhancement, in particular to a double-branch low-illumination image enhancement method based on the Retinex theory, which comprises the following steps of: S1, acquiring an illumination component and a reflection component by using a decomposition network based on the Retinex theory, and denoising the reflection component and enhancing the illumination through an enhancement network; s2, carrying out initialization feature extraction on the input low-illumination image through parallel connection of hole convolution with different expansion rates, and carrying out deeper feature extraction on the initialized image through an enhancement unit in an iterative learning module and a U-Net network; and S3, obtaining a corrected illumination image through a fusion unit, and fusing the obtained illumination image and the reflection component after three times of iterative learning to obtain a final enhanced image, the method provided by the invention effectively improves