Research On Single Image Demining Algorithm Based On Background Reconstruction And Detail Supplement

In recent years,with the vigorous development of software and hardware technology,computer vision has been widely used in military reconnaissance,public security and video surveillance fields.However,due to the influence of complex environmental factors under outdoor conditions,many mature technologies still suffer from a significant drop in accuracy,such as missing a large number of detection targets in vehicle detection tasks.Rainy scene,as one of the complex environmental factors,makes the images containing a large amount of raindrop information,resulting in the image blur and target loss,leading to serious deviations in subsequent image recognition or detection.Therefore,in order to solve the negative effect of raindrops in rainy scene images,rain removal technology has been proposed and continuously developed in recent years.This technology aims to restore and improve the visual quality of images or videos,so that subsequent computer vision algorithms can still achieve the desired effect in rainy scene.At present,the methods for removing rain from images are mostly based on deep learning technology.They mainly remove rain layer information from the physical imaging model of the rainy images or directly reconstruct the background to obtain a rain-free image to achieve rain removal.Compared with the physical imaging model,the method of directly reconstructing the background does not require the construction of the physical model,thereby avoiding the negative influence caused by the inaccurate physical model.Therefore,based on the method of direct reconstruction of the background layer,this paper proposes two new image rain removal schemes,which aim to solve the problems of residual rain streaks and loss of background detail information that are common in the current mainstream rain removal methods.The main innovations are summarized as follows:(1)A single image deraining method based on progressive residual detail supplement mechanism is proposed.The whole rain removal process adopts a progressive iterative method to gradually obtain high-quality rain removal results from coarse to fine.In the network architecture,this paper proposes a diamond residual block as the main iterative module to extract rich feature information.At the same time,in order to reduce the loss of details in the process of rain removal,a detail supplement mechanism composed of two information supplement methods is proposed to restore detail information of the background layer through the information transfer between iterations.A large number of experimental results prove that the proposed method can achieve more thoroughly rain removal effect and retain background details well.In addition,the algorithm can also be transferred to the field of image denoising to achieve high-quality denoising effects.(2)Taking into account the intricacies of rain layer information and the loss of details in the process of rain removal,this paper proposes a single image deraining method based on information multi-distillation method and multi-order difference detail extraction.The solution based on information distillation achieves gradual rain layer removal and background information reconstruction by constructing detail supplement information distillation block.At the same time,a multi-order differential detail block is constructed in the method,and the detail information can be restored by extracting the different detail information under different receptive fields.A large number of experimental results show that compared with most mainstream rain removal methods,this method has better performance in subjective visual quality and objective evaluation indicators.