Theoretical Method And Experimental Research Of Particle Size Distribution Detection Based On Image Analysis Technology

In the construction of rockfill dam and other hydraulic engineering projects,the particle size distribution(PSD)of blasting mining materials affects the quality control of filling construction,as well as the deformation,permeability,pore pressure coefficient and shear strength of the dam,which is an important construction control parameter.In addition,the PSD also affects the loading,transportation,secondary crushing and other subsequent construction efficiency of muck-pile.The traditional sieving technology is time-consuming and laborious,which affects the production cycle and has low efficiency.Therefore,it is of great engineering significance to find a convenient and fast particle size detection method.Image analysis has become one of the alternative methods of mechanical sieving because of its advantages of convenient equipment,mature algorithm,no contact and no impact on production.Based on the study of image analysis theory and laboratory experiments,the realization of important steps of digital sieving is studied and a PSD detection system is developed in this paper.In addition,considering the shortcomings of digital sieving,the PSD detection method based on pattern recognition is studied.The specific contents and results of research are as follows:(1)The traditional watershed transform is prone to over-segmentation due to the influence of minimum regions.In this paper,two improved watershed transforms based on gray morphological operation are proposed.The first method uses the corroded image and the original gray image for morphological reconstruction,then the dilate operation and the reconstructed image are for morphological reconstruction again,after the image is inverted,the maximum value area in the rock is extracted,and then the watershed transformation is carried out;the second method is to obtain the maximum value area by combining the algebraic operation between the top hat transformation and the bottom hat transformation(2)In order to measure the particle size from the image accurately,the laboratory experiment was carried out.30 kinds of two-dimensional size or shape features are extracted from rock particle images,and NCFS is used for feature selection.Three problems in digital sieving,such as sieve grading determination,volume reconstruction and overlap discrimination,are solved.The related mathematical models and SVM classification model are proposed,and the research results are more accurate than the existing methods.(3)According to the improved watershed algorithm based on gray morphological operation and rock particle size and overlap estimation methods,a GUI system for PSD detection is developed based on the MATLAB platform,and the feasibility and practicability of the system are verified through the engineering sieving test.(4)Aiming at the problems and shortcomings of most image segmentation algorithms,such as the dependence of controllable light source,the low robustness of outdoor environment,and the need for users to adjust parameters,the PSD detection based on pattern recognition is studied.Six methods are used to extract image texture features,then PCA is used to reduce the dimension of features,and an RBF neural network model between texture features and PSD is constructed.Cross validation results show that the model has good generalization ability.The model is integrated into the GUI system and can guide the engineering practice to a certain extent.This method based on pattern recognition does not need image segmentation and manual operation.It has important theoretical research significance and should be explored in-depth practice.