Research On Agricultural And Forestry Image Rain Removal Based On Joint Model-data Driving Algorithm

The image acquired in the rainstorm scene will have a lot of rain and fog,and the subject and background information will be seriously degraded and cannot be distinguished.In computer vision tasks,this will seriously affect the processing of subsequent tasks,such as object detection,object tracking,etc.At present,the rain removal method based on deep learning has achieved good results in the recovery of rain maps with only rain streaks,but there is still a lot of room for improvement in the case of rain and fog coexisting in heavy rain scenes.Aiming at the coexistence of rain and fog in heavy rain scenes of agriculture and forestry,this thesis establishes an indoor rain dataset with real depth maps based on the physical model of rain and fog,named Indoor-Rain.At the same time,a two-stage network TSF-Net combining model-driven and data-driven was constructed.The model-driven network constitutes the initial step.First,we use a guided filter to filter the high and low frequencies of the image and optimize it with Selective Kernel Feature Fusion(SKFF).Then,the rain strip information is obtained from the high-frequency information,the atmospheric light is obtained from the low-frequency information,and the transmission map is obtained from the high-frequency and low-frequency mixed information.The second stage is a fully convolutional neural network with a U-Net structure.This paper proposes a Multi-Scale Project Fusion(MSPF)module based on model-driven and data-driven.The module is used to perceive the spatial information of the image across scales,and the residual rain and blurred parts are removed in the first step,so as to obtain better clean images.Our method is evaluated experimentally numerically and visually against the state-ofthe-art rain removal methods on the Outdoor-Rain,the Indoor-Rain,and Agriculture and Forestry datasets.Experiments show that the proposed method improves 1.01dB and 1.6dB on Indoor-Rain and Outdoor-Rain respectively,which is superior to other rain removal methods in terms of restoration effect.At the same time,the restored agricultural and forestry images also have good visual clarity and details.