Research On Image Segmentation Algorithm Of Multi-grain Adhesive Rice

Food security is the most important ballast for maintaining economic development and social stability.Rice is one of our important staple foods.The quality inspection of rice is of great significance to the quality of people's life and the development of the food industry.When inspecting the quality of rice,there is often a problem of adhesive rice that cannot be handled,which adversely affects the inspection work.The correct segmentation of multigrain adhesive rice is an urgent requirement for grain quality inspection.Due to the complex adhesion of rice grains,almost all segmentation algorithms for simpler situations are seen in the literature,which are far from the various complex situations in actual grain detection.In view of this,the thesis studies the multi-grain adhesive rice image,aiming to use image processing technology to propose an accurate and effective multi-grain adhesive rice image segmentation algorithm.The main innovative work of this thesis are as follows:(1)An optimized two-dimensional OTSU method(OTSU)segmentation algorithm based on improved particle swarm optimization(PSO)is proposed to realize the segmentation of multi-grain adhesive rice image target and background.Based on the standard PSO algorithm,the algorithm firstly optimizes the inertia weight coefficient and acceleration coefficient to improve the ability of particle optimization and the speed of calculation convergence,then the improved PSO is introduced into the 2D OTSU algorithm to increase the value search rate and reduce split time.This algorithm improves the traditional 2D OTSU algorithm's computational redundancy,complicated process,and long segmentation time.The segmentation experiments on different rice images have proved the effectiveness of the algorithm.(2)A segmentation algorithm of adhesive rice with holes is proposed.Aiming at the characteristics of porous multi-grain rice adhesion,first obtain the inner pit area through the hole filling and difference calculation,then perform the area closing calculation and the difference calculation to obtain the outer pit area.According to the set area threshold,filter out the pseudo-pit area,then the pits are matched and connected to the inner and outer pit areas respectively.Finally,the segmentation result is judged according to the size of the rice grains after segmentation,until the segmentation is correct.Experiments show that the algorithm is effective and feasible.(3)Aiming at various complex situations of adhesive rice images,a relatively complete general segmentation strategy is proposed.First,according to whether there are holes in the image of the adhesive rice,the adhesion of multi-grain rice are divided into two types:tandem type and hole type,and processed separately.There are only external pit areas in the tandem type,and they correspond to each other.Determine the category of the outer pit area according to the judgment line,and perform pits matching according to the minimum Euclidean distance principle.In addition to the external pit areas,there are also internal pit areas in the hole type.The pit matching of the inner pit area is to expand the circle with its center as the center,find all the outer pit area associated with the inner pit area,and then pair the pits according to the minimum Euclidean distance principle.Other remaining external pit areas are processed in accordance with the method of tandem adhesion.Finally,the preliminary segmentation result is judged according to the rice grain size criterion until the segmentation is correct.Experiments verify the effectiveness and wide adaptability of proposed strategy.