Research On Image Segmentation Algorithm Of Touching Grains Based On Boundary Concave And Convex Points And Neural Networks

According to the classification of rock grains in the mining industry,the traditional methods of mechanical sieving and manual measurement of rock grains size had the disadvantage of poor real-time performance.Replacing the traditional complicated and time-consuming manual sampling method with digital image processing technology has the advantage of reducing the manpower,material resources and time spent on quantifying the distribution of rock grain size and estimating the size distribution automatically and accurately with smaller labor intensity.In this paper,the following digital image processing techniques are applied to the dim and blurred image of rock grains:Image enhancement of rock grains image,edge extraction of rock grain image,classification of touching rock grains and separation of touching rock grains.First,the guided filtering algorithm enhances the dim and blur image,double threshold Otsu method binary gray image,and then extract the feature parameters of rock grain image to train BP neural network for distinguishing whether the rock grain has touching parts or not,extract the touching rock grains of the main concave point and the segmentation point pair,the normalized cutting algorithm to separate the touching rock grains.The specific research content is divided into the following aspects:1,Rock grain image acquisition and preprocessing:in the ore or aggregate conveyor belt,ore-rock grains overlap each other and it is difficult to distinguish between ore and conveyor belts,but at the end of the conveyor belt,the ore and rock grains are scattered as a result of falling down and easy to distinguish the foreground and background,so the image of the rock grain gravitational flow was taken at the end of the conveyor belt.In the process of image acquisition,the image is influenced by the noise and inhomogeneous illumination,and then the different filtering algorithms are chosen according to the noise category contained in the image.2.Initial segmentation of rock grain image:After image enhancement of the original image,this paper compares and analyzes the segmentation effect of various threshold segmentation algorithms and selects double threshold Otsu to segment the rock grain image.This algorithm has the advantages of reducing the touching rock grains and preserving the basic contour of the rock.3.Distinguish whether the rock grain has touching parts or not in the image:Extract the five shape feature parameters between the binary image of the rock grains and the corresponding convex hull,label these features artificially whether the rock grain has touching parts or not,and then use these eigenvectors to supervise the training of the BP neural network.After the training,the BP neural network is used to distinguish whether the rock grain has touching parts or not.4.Separation of the touching rock grains:In the image of touching rock grains,the concave point with the deepest concave defect between the convex and the edge contour is extracted as the main concave point.The point is selected as the segmentation point pair,which is the closest to the main concave point or another concave point at the opposite side of the main concave point.Then,the segmentation point pair is incorporated into the normalized cut criterion,and the corresponding weight matrix is constructed,and the touching rock grains are separated by the normalized cut algorithm.5.The data statistics of rock grain image segmentation:According to the statistical analysis of the segmentation accuracy of this algorithm and the traditional algorithm,the experiment proves the validity of the algorithm.Finally,the size distribution of the rock grains is calculated from the segmentation result image.