Low-light image enhancement method based on deep convolution attention and multi-scale feature fusion

The invention relates to the technical field of image processing, in particular to a low-light image enhancement method based on deep convolution attention and multi-scale feature fusion. A new low-light attention module and a multi-scale feature compensation module are designed, the LLAB is composed of a low-light multi-head self-attention module, a double-branch equalization module and two normalization layers, and the low-light multi-head self-attention module is designed to be used for extracting semantic information of different channels; and feature weights among different channels are balanced by calculating an attention map among the channels, so that the visibility and the contrast ratio of the image are improved. The double-branch equalization module further improves the contrast of the image; a multi-scale feature compensation module is provided to make up for loss of detail information of the image in a low light attention module and a down-sampling stage, and deep space information of images of d