Research On Single Image Rain Removal Algorithm Based On Lightweight Convolutional Neural Network

At present,computer vision is widely used in outdoor video surveillance,unmanned driving,license plate recognition and other scenarios.However,the existing algorithms are based on the premise of good weather conditions,without considering the impact of different weather conditions on the performance of the algorithm.In the case of rain,snow,fog,etc.,the image quality is lower,and the performance of the algorithm will be reduced.At the same time,the photos taken on rainy days have problems with rain streaks,raindrops,rain and fog,which leads to the loss of details of the subject in the photo and affects the viewing of the photo.Therefore,in the field of computer vision,image de-raining can be used as a preprocessing for algorithm application,improve the performance of the algorithm in bad weather,and enhance the robustness of the algorithm.In addition,the image rain removal algorithm can remove the raindrops that affect the subject of the photo,restore the information of the subject,improve the quality of the photo,and facilitate viewing.Since single image de-raining does not require the information of continuous multiple frames of images,the use of scenes is more extensive,which can not only replace the existing video image de-raining algorithm,but also satisfy many image de-raining scenarios that can only obtain a single image scene.demand.The image de-raining algorithm is mainly deployed on mobile devices such as outdoor devices or mobile phones with limited computing resources However,most existing deep learning-based de-raining algorithms do not consider the computational complexity of the model,which makes it difficult to apply the model in actual scenarios.The model proposed in this paper fully considers the rain removal effect and running speed of the model.Using residual learning and local spatial attention modules to improve the rain removal effect of the model.Compared to directly obtaining clean images from rainy images,it is simpler and easier to treat image rain removal as removing rain streaks in rainy images.At the same time,the attention mechanism is introduced to further improve the model performance.A larger spatial attention range can enhance the features related to raindrops in the model,suppress irrelevant features,and improve the performance of the model in detecting and removing raindrops.In order to improve the speed of the model,deep separable convolution is used instead of standard convolution to extract features,which significantly reduces the calculation of the model.At the same time,the use of depth-wise convolution with dilated in the local attention module requires only a small calculation A large amount of spatial attention can be obtained by the amount,which not only improves the model effect,but also reduces the calculation amount of the model.Finally,in order to verify the effect of the model,a comparative test was conducted on multiple public data sets to prove that the model has a good rain-removing effect and runs significantly faster.