Colony Image Analysis And Recognition Method Based On Digital Image Processing Technology

The colony is one of the important research objects in microbial technology,which can realize the evaluation of food safety level,environmental pollution degree,therapeutic effect of medical drugs,and characteristics of agricultural fungicides.Traditional colony image research requires human visual observation and statistics,which will result in low work efficiency and high work intensity.Based on digital image processing technology,it is of practical significance to carry out the research on colony image analysis and recognition method.This paper mainly completes the following work.(1)An edge detection method of colony images based on mediocrity ant colony algorithm(MACA)was proposed to achieve edge detection of colony images.MACA combines the mediocrity rule,uses empirical functions to establish a pheromone database that can be used as a pheromone update reference table,adopts the Chebyshev distance as a weight that affects pheromone update,and combines heuristic information acquisition with maximum variance classification method and local path weights.The method that jointly affects the ant transition probability incorporates feedback rules for obtaining path weights to improve the edge detection effect.By performing edge detection simulation experiments on six colonies of three types of bacteria,and comparing with the classic edge detection operators and two classic ant colony edge detection algorithms,the detection performance,detection results and running time are proposed.The stability and accuracy of MACA algorithm is better than other methods,and the ideal results of the colony image edge detection by the ant colony algorithm are obtained.According to the image of edge detection results,a method for colony counting was realized.(2)A new method of colony image feature extraction,intrinsic dimensionality estimation and dimensionality reduction based on digital image processing technology was proposed.Firstly,nine color characteristics of colony images were extracted based on HSI image color moments,which included first,second and third order moments.Based on the gray co-occurrence matrix and its characteristic parameters,20 texture features information of the colony images were obtained.Then,three essential dimensionality estimation methods,namely the associated dimensionality estimator,the maximum likelihood estimator and the cluster quantity estimator,are used to estimate the dimensionality of the inner low-dimensional structure of the colony images.The results of different dimensionality estimation were combined with PCA,LDA,MDS,ISOMAP,SNE and NCA to form 18 different data dimensionality reduction methods.Based on the distance criterion function,an optimal dimension reduction method for the high-dimensional feature data of colony image samples corresponding to the maximum distance criterion function is obtained.(3)A colony image Support Vector Machine(SVM)classification method based on sense organs Grey Wolf Optimizer(GWO)was proposed.Four improved algorithms based on classical GWO are proposed.They are Olfactory Grey Wolf Optimizer(OGWO),Visual Grey Wolf Optimizer(VGWO),Auditory Grey Wolf Optimizer(VGWO),and Sense Grey Wolf Optimizer(SOGWO).The SVM classifiers were optimized by using five algorithms respectively.The optimized SVM classifiers perform classification experiments on the colony image samples to verify the convergence of each algorithm,the distribution of sample classification and its accuracy rate.By comparing the results,the performance of the proposed four algorithms is better than that of the classical GWO on the problem of colony images classification.Among them,SOGWO shows the optimal performance in terms of convergence and classification accuracy.