Training method and device, deraining method, terminal equipment, storage medium

Abstract

The invention discloses a training method and device, a rain removal method, a terminal device and a storage medium, and relates to the field of image processing. The training method comprises: reading a piece of rainy picture block on a rainy image; using the rainy picture piece as a feature image through a channel promotion module, a global information attention module, and a rain removal cyclic neural network model successively. Layer attention module and channel restoration module obtain the rainwater picture block corresponding to the feature image; subtract the rainwater picture block from the feature image corresponding to the rainwater picture block to obtain the rain-free picture block corresponding to the rain picture block; The rain-removed picture block is input into the rain-removing cyclic neural network model again as a feature image until the rainwater in the obtained rain-removed picture block is removed. The rain-removing cycle neural network model of the present invention performs rain training on input rain-bearing picture blocks, thereby improving the recognition ability of rain.

Description

technical field [0001] The present invention relates to the field of image processing, in particular to a training method and device, a rain removal method, terminal equipment, and a storage medium. Background technique [0002] With the development of computer technology, various computer vision algorithms have penetrated into all aspects of daily life. For example: the semantic segmentation algorithm for self-driving cars to detect the road conditions ahead, the face recognition algorithm for tracking fugitives deployed on Skynet cameras outdoors, the license plate number recognition algorithm deployed on outdoor traffic monitoring probes, etc. The normal operation of these algorithms plays a vital role in maintaining road safety and maintaining social stability. [0003] However, when it rains, the raindrops and rain fog brought by the rain have a serious impact on the visibility of the air, which greatly reduces the recognition accuracy of most computer vision algorithm...

AI Technical Summary

Problems solved by technology

[0005] However, based on the limitation of inertial thinking, the existing image deraining methods all input the image with rain and directly output the image without rain, which is essentially training the entire image instead of training the rain separately, although It is also possible to get the rain-removing image, but the rain-removing performance cannot be satisfied

Images