Research On Ore Granularity Detection Technology Based On Machine Vision

The granularity information of the crushed ore is the main indicator of optimalcontrol for beneficiation. Taking the machine vision technology into the detection ofore granularity information, it can acquire the real-time ore granularity informationand timely feedback and adjust the crusher processing equipment parameters.Therefore it can improve beneficiation efficiency, reduce energy consumption andmake full use of mineral resources. The use of machine vision technology for oregranularity detection must firstly segment the ore particles in the image, but the oreparticles with a large quantity, adhesions, large differences in size, irregular shape andother characteristics, which brings great difficulties for dividing. Meanwhile, thecurrent segmentation algorithm is mainly using image brightness feature forsegmentation, and the current ore segmentation for complex ore images is difficult toachieve the desired effect.This thesis studied the ore granularity detection technology based on machinevision. It first studied and chose the preprocessing methods on ore images; then, Forthe problem of existing segmentation algorithm, proposed a multi-scale segmentationalgorithm with multi-feature fusion for complex ore image; finally it studiedquantitative describing methods of the ore granularity parameters. The specificresearch work and innovations are as follows:1. The preprocessing method of ore image. In this thesis it adopted the bilateralfiltering algorithm for image filtering, and can preserve edges of ore particle whilefiltering noise; it has used the method of combining the integral image and the localthreshold for image binarization to extract the ore target areas, and the algorithm withthe advantage of lighting adaptability and high efficiency; finally, it eliminated thenoise in the binary image and smoothed the edges of ore particles by morphologymethods.2. The research on adaptive segmentation algorithm for complex ore image. Inthis thesis, the local extrema point in the distance transform image was processed bythe bilateral filtering, solved over-segmentation problem of watershed segmentationalgorithm which was based on luminance characteristics. it proposed an ore imagesegmentation algorithm based on concave point detection and matching. Thissegmentation algorithm achieved the concave point detection and matching for binaryimage by combining Harris operator and circular templates, and adhesive ores were separated successfully. In order to further improve the accuracy ratio of segmentationalgorithm, it proposed a multi-scale segmentation algorithm for complex ore image bycombing with improved watershed algorithm, concave point detection and matchingalgorithm and SVM-based model of ore texture recognition. It also merged thebrightness feature, shape and texture features. Finally the result of segmentation wasevaluated. The cooperation process mechanism used characteristics and segmentationalgorithms, and combined feature extraction and image recognition. Thus, it enhancedrobustness of algorithm, and effectively decreased under-segmentation andover-segmentation region.3. Ore granularity parameter extraction and design of detection system. Due toore particles are difficult to quantify, this thesis studied the description method ofgrain granularity. And then it chose the right granularity parameter to describe oreparticles. On the basis of image segmentation, pixels in ore segmentation region werecalibrated. At the meantime, perimeter diameter, Feret’s diameter and other granularityparameters were obtained. Finally, designed hardware platform and developed oregranularity detection system.