Research On Image Segmentation Of Industrial Soot Based On Improved Lpf Level Set Model

With the increasing number of factories,the continuous discharge of industrial soot has caused serious pressure on the atmosphere,which has attracted worldwide attention.Our country also pay high attention to the environmental problems,constantly put forward new policies to control the environment.In the treatment of atmospheric environment,there are strict policy requirements for industrial soot emissions,therefore,the emission of industrial soot needs real-time detection.With the development of computer vision technology,it is historical inevitability to apply relevant technology to the field of industrial soot detection.In the process of realizing the detection of industrial smoke,it is a very key point to separate the smoke area from the background,but the characteristics of the industrial smoke image itself make it difficult to achieve segmentation,often appear wrong classification,leakage phenomenon.In this dissertation,the traditional level set segmentation method is proposed to study the separation of chemical dust.The main research work of this dissertation is as follows:(1)The traditional bilateral filtering is improved.There are many interferors in the background of industrial smoke image,there are a small range of interferors,and the contrast between smoke and background is low.These characteristics cause great difficulties in the segmentation of smoke.In order to reduce the influence of these factors on the segmentation effect,an improved bilateral filter is proposed to preprocess the industrial soot image.Based on the analysis of the traditional bilateral filtering,it is proposed to add a local average term to the range weight of the bilateral filtering to adjust the range weight,reduce the loss of edge information in the filtering process,and realize the fuzzy processing of the small range of interference.The experimental results show that the improved bilateral filter has a better effect on the subsequent segmentation,can blur the interference in the industrial soot image,and has an enhanced effect on the target.(2)A study of a new algorithm for industrial smoke image segmentation based on an improved Local Pre-fitting(LPF)level set model.Although existing segmentation models based on local information can achieve segmentation of images with inhomogeneous intensity,they do not deal with the noise problem and are susceptible to interference from noise during segmentation,and most of them need to repeatedly calculate the local fit values in iterations,so the models are not efficient.Firstly,the industrial smoke image is pre-processed using improved bilateral filtering to deal with the disturbing factors in the background;then a new data fitting term based on local pre-fitting is constructed by using a local pre-fitting method to calculate the industrial smoke image;finally,a new level set energy functional is constructed by constructing a regular term based on a logarithmic function with polynomial and then introducing a length energy term.The experimental results verify that the comprehensive performance of the proposed model is effective in the segmentation of industrial smoke and dust images.(3)A new algorithm for industrial soot image segmentation based on improved LPF and Signed Pressure Force(SPF)models is investigated.Aiming at the problems that industrial soot images are easily disturbed by the background during segmentation,the large number of interfering objects in the industrial soot background and the miscellaneous types of interfering objects,a new active contour model based on improved LPF and SPF is obtained by combining the local pre-fitting idea with the SPF model.Based on the pre-processing of the input image using improved bilateral filtering,the model proposes a new symbolic distance function using the local pre-fitting idea,and constructs a new local-based region-based level set model.The model was chosen to use selective binary Gaussian filtering in the regularisation method,so that the model retains the ability to have local or global segmentation.The results of the self-evaluation experiments against the other five models illustrate the effectiveness of the new algorithm for industrial fume segmentation based on the LPF modified SPF model for industrial fume segmentation and the comprehensive performance is stronger than the five comparison models.