Study On Granularity Model Of Sintering Mixture Based On Deep Learning

At present,in the production of sintered mixture,particle size detection mainly relies on manual work.Continuous and accurate data cannot be obtained,and it is difficult to determine the quantitative relationship between particle size distribution and production parameters.Therefore,a particle size model of sintered mixture is proposed to provide timely and accurate particle size data.A CCD industrial camera and an industrial light source were selected,and a set of image acquisition equipment was assembled with a machine vision experimental bracket.Analysis of the images found that the mixture was mostly oval or irregular.The image parameters are analyzed by the image histogram,the grayscale processing is completed by the weighted average method,and the noise reduction and detail enhancement are performed by the bilateral filtering method to improve the detection accuracy.Compared with Sobel,Laplace and Canny,an improved Canny algorithm segmentation model is proposed.Added binarization and morphology optimization.Granularity data are plotted by circumscribed circles.The segmentation models were tested using two batches of images of sintered mixtures with different compositions and compared with the model processing results based on the manual segmentation results.The average segmentation accuracy is 81.77% and 67.49%,and the average segmentation accuracy error is 17.27% and 24.57%.Compositional changes can add character to the image and affect the result.Affected by over-segmentation and under-segmentation,the error of the average particle size will fluctuate,affecting the results.Comparative analysis,select Unet network to further improve the accuracy.Before training the Unet network,the annotated images are segmented,which solves the problem of insufficient training datasets.Compare the before and after segmentation results.The segmentation accuracy is improved by 9.39% and 20%,and the segmentation accuracy error is reduced by 6.72% and 11.6%.The particle size distribution and average particle size results are closer to manual statistical results.It is found that the Unet network segmentation model has better segmentation effect on the sintered mixture and is more suitable for the particle size model of the sintered mixture.Figure 57;Table 5;Reference 73...