Endoscopic Image Smoke Removal Algorithm Based On Attention Feature And Bilateral Grid Learning

In minimally invasive surgery,it is of great practical significance to remove smoke in endoscopic images due to the smoke generated by surgical operations such as cutting tissues and laser ablation from electrothermal energy sources,which reduces the quality of endoscopic images and obscures the doctor’s surgical field of view,resulting in increased surgical risks.Most of the existing smoke removal algorithms are based on image dehazing algorithms,which can remove smoke but at the same time there will be information loss,and cannot maintain the details of endoscopic images well,so this paper proposes two endoscopic image smoke removal algorithms for the detailed recovery of endoscopic image smoke removal,which can eliminate endoscopic image smoke while maintaining the natural appearance of organ tissue surface.The main contents and innovations of this article are as follows:(1)Aiming at the problem that the existing endoscopic image dehazing algorithm will cause a certain degree of image color distortion and excessive local noise,this paper proposes a MRMGB-De Smoke algorithm by introducing Multi Res Unet for feature extraction,combining bilateral grid and multi-guided learning to construct a high-quality recovery guidance tensor.Multi Res Unet can identify smoke at different scales,retain more detailed information,and effectively extract image features,which in turn enables the proposed MRMGB-De Smoke algorithm to restore the original style color of endoscopic images,solve the problem of local noise of images,and achieve good results in the removal of endoscopic image smoke.(2)Aiming at the problem that the existing endoscopic image dehazing algorithm has a certain degree of smoke residue and the local detail recovery is not fine enough,this paper introduces the attention mechanism module and perception loss,constructs a bilateral grid,and combines multi-guided learning to propose the CPMGB-De Smoke algorithm.By introducing the attention mechanism module,the channel attention of important frequencies is emphasized,and the details of the output image can be improved;In addition,on the basis of the original loss,combined with the perceived loss,by comparing the feature similarity between the recovered image and the real image,the structure and details of the image can be preserved,so as to obtain clear effects and vivid colors.The results show that the MRMGB-De Smoke algorithm proposed in this paper can effectively solve the problems of image color distortion and local noise in the existing algorithms.The CPMGB-De Smoke algorithm not only solves the problems of smoke residue and insufficient detail recovery,but also can better restore the original color of the image,making the recovered image more similar in structure to the target image.