Research And Application Of Parallel Genetic Algorithms Based On MapReduce

Edge is one of the most core features of images and contains a lot of useful information and data. Edge information technology utilizes image edge detection for image segmentation to extract the target image, which is the basis of image analysis. Therefore, it is widely used in image segmentation, object tracking, pattern recognition etc. As a kind of Intelligent Optimization Algorithm, Genetic Algorithms can optimize edge detection to extract the image edge.When processing image applications with high computational complexity and large amount of data, Genetic Algorithms will appear the problem of poor quality of solution, slow convergence speed etc. In order to improve the speed and capacity of processing large amounts of complex data, this paper studies the implementation framework of the Parallel Genetic Algorithms based on the Map Reduce model. Based on this framework, some improvements studied on the Parallel Genetic Algorithms can improve the quality of solution.The main creative results are as follows: 1.In order to overcome the premature problem of Genetic Algorithms, this paper puts forward an improved method for evaluating the degree of prematurity. By calculating the ratio of the average fitness of the better individuals to the maximum fitness of the population, this method can assess the degree of premature. Compared with other methods, this method is simple in calculation and normalized. 2. Using the above method for evaluating the degree of premature, this paper improves the migration operation of the Parallel Genetic Algorithm. This paper presents a method that can adjust the sub population migration cycle dynamically according to the degree of prematurity of sub population, which is helpful to improve convergence rate and the quality of the solution. For the improved Parallel Genetic Algorithms, this paper designs the implementation framework of Parallel Genetic Algorithm based on the Map Reduce programming model. 3.According to the theory that in high dimension space the extreme points tends to distribute in the boundary of space, this paper puts forward the concept of boundary membership of points. This concept that reflects the degree of boundary membership of points can be used to evaluate whether those points converge to the extreme point. In the selection and mutation operations of Genetic Algorithms, the solution boundary membership is used to make individuals adjust dynamically crossover and mutation probability, which can improve the convergence of the algorithm. 4. When the improved Parallel Genetic Algorithms presented above is applied to edge detection, the experiment has achieved good detection results and the processing speed of the algorithm has been greatly improved.