The paper considers a neural network approach to cleaning images from the noise component in the form of rain, the use of which will improve the quality of detection of moving objects in adverse weather conditions. A generative adversarial network was chosen as the neural network architecture. The main idea of image processing in order to remove the noise component in the form of rain is that a rectangular area of 256 by 256 pixels is selected from the original image with the rain component (the fragment is selected randomly). Then this fragment is fed to the generator, which cleans it from the rain component. Then the generator passes the processed fragment to the discriminator, which, in turn, tries to understand whether the fragment it received is processed or reference. Thus, the better the discriminator works, the better the generator performs the assessment and vice versa. The paper presents two approaches to cleaning images from the rain component: an approach to eliminating the rain component in the form of stripes; an approach to eliminating the rain component in the form of a curtain (fog) effect.