Research And Application Of Multi-threshold Image Segmentation Method Based On Swarm Intelligence Optimization

Currently,medical pathology image analysis and processing is the basic prerequisite for diagnosis of many diseases,and medical image segmentation is the key issue to determine whether medical images can provide a reliable basis for clinical diagnosis and treatment.Since the purpose of image segmentation is very different,which brings great challenges to the development of image segmentation technology,coupled with the fact that medical images contain a large amount of complex data information and the increase in the number of thresholds,the time complexity of traditional multi-threshold image segmentation methods has increased dramatically,so it is important to build a high-quality multi-threshold medical image segmentation model.In this paper,we construct multi-threshold image segmentation models based on nonlocal mean filtering,2D Kapur entropy,and 2D histogram with swarm intelligence optimization algorithm as the core,and then apply these models to the segmentation of various medical pathology images.We first discuss the shortcomings of the continuous ant colony optimization algorithm,the differential evolution algorithm,and the multi-verse optimization algorithm in terms of solution quality,convergence accuracy,and falling into local optimums.Finally,they are used to improve the traditional multi-threshold image segmentation methods respectively,and then we propose the multi-threshold image segmentation model based on the ant colony optimization with mutation mechanisms,the multi-threshold image segmentation model based on the differential evolution with slime mould foraging mechanism,and the multi-threshold image segmentation model based on the multi-verse optimization with learning strategies,and they are used for the pathology of COVID-19,breast cancer,and lupus nephritis respectively image segmentation,which greatly improves the segmentation quality and segmentation efficiency.The details are described as follows.(1)For the segmentation problem of COVID-19 pathological images,a multi-threshold image segmentation model based on ant colony optimization with mutation mechanisms is proposed.First,an improved ant colony optimization algorithm is proposed by introducing the Cauchy mutation mechanism in the continuous ant colony optimization algorithm to enhance the search ability of individual ants,as well as using the greedy Levy mutation to enhance the self-learning ability of the optimal individual.Experimental simulations are performed on 30 test functions from IEEE CEC2014,and the experimental results show that the proposed method is enhanced in terms of search ability and self-learning ability,and its solution quality is better when compared with its 3 variants and 10 peers.Further,we combine the algorithm with 2D Kapur entropy based on nonlocal mean filtering and 2D histogram,in order to build the segmentation model.Finally,the method is tested for segmentation on COVID-19 pathology images.The experimental results show that the model achieves better image segmentation results.(2)Aiming at the problem of breast cancer pathology image segmentation,a multi-threshold image segmentation model based on differential evolution of slime mould foraging mechanism is proposed.Firstly,the differential evolution algorithm is improved by employing slime mould foraging behavior to enhance the convergence accuracy and the ability to avoid falling into local optimum.The proposed method is compared with nine conventional algorithms,and nine improved algorithms by simulation experiments on 30 test functions of IEEE CEC2014,and the advantages of the proposed method in terms of convergence accuracy and ability to avoid falling into local optimum are well demonstrated.Further,the 2D Kapur entropy is used as the objective function of the method,and a multi-threshold image segmentation model is developed by combining the nonlocal mean filtering technique and the 2D histogram technique,so as to achieve the segmentation of breast cancer pathology images.Finally,the segmentation ability of the model was tested on pathological images of invasive breast ductal carcinoma from a public dataset and compared with other methods.The experimental results show that the method can obtain high-quality segmentation results when performing image segmentation on invasive ductal breast cancer.(3)We propose a multi-threshold image segmentation model based on multivariate universe optimization with learning strategy for lupus nephritis pathology image segmentation problem.First,the multi-verse optimization algorithm is improved to speed up the convergence rate of the algorithm by using the strong convergence property possessed by the proposed reflection learning strategy,and the orthogonal learning strategy is used to improve its convergence accuracy and the quality of the solution.The results are tested on 31 benchmark functions and show that the improved multi-verse optimization algorithm achieves a large improvement in both convergence and solution quality when compared with 9 peers.Further,we introduced the method into the 2D Kapur's entropy-based image segmentation method and dynamically optimized its optimal set of segmentation thresholds to obtain the optimal set of thresholds for segmentation.Finally,experiments on lupus nephritis pathology images show that the method can find a higher quality threshold set and obtain better results for segmentation of lupus nephritis pathology images.