Research And Application Of Kernel Extreme Learning Machine In Image Segmentation

Image segmentation is an extremely important part of the image engineering.However,there is no general theory of image segmentation so far,it is of great significance to find an excellent image segmentation algorithm.At the beginning of design,the core consideration of extreme learning machine which was a new machine learning method was: fast learning,reducing intervention and high accuracy.By introducing the kernel function into the extreme learning machine,a kernel extreme learning machine can be obtained.Compared to the extreme learning machine,kernel extreme learning machine have better performance for feature learning and multiple classification problems.At present,there is less research on the application of kernel extreme learning machine to image segmentation.In this paper,the following segmentation algorithm is applied to color image segmentation to get better segmentation results,and has a certain value of application.Firstly,This paper designs a kernel extreme learning machine image segmentation algorithm based on saliency map.The algorithm detects the saliency of the original color image to get the saliency map.Then the saliency map is binarized and corroded,and finally we can get the corroded foreground area,background area.Selecting the training sample points,extracting the attributes of attributes,training the kernel extreme learning machine and obtaining kernel extreme learning machine classifier can be used the segmentation of the original color image.The segmentation algorithm obtained a better segmentation results,which also illustrates the kernel extreme learning machine as an excellent classifier for image segmentation has a broad prospect.Secondly,a kernel extreme learning machine method based on fuzzy C-means(FCM)pre classification is introduced.Considering that the process of selecting the sample points by the above segmentation algorithm is slightly complicated and can not be self-selected,the FCM pre classification training sample is adopted,Three FCM-KELM algorithms are implemented.At the same time,these three algorithms are compared with the FCM pre classification BP neural network,least square support vector machine,Gaussian kernel support vector machine.By doing comparative experiment analysis,The experimental results show that the FCM-KELM image segmentation algorithm is an efficient image segmentation algorithm,which not only can achieve fast segmentation,but also has high segmentation accuracy and has a certain application value.Finally,on the issue of improving the segmentation accuracy and the parameters no longer selected empirically,the corresponding optimization strategies are offered: the polynomial kernel and the Gaussian kernel are convex combined to form a multi-core extreme learning machine,and the particle swarm optimization algorithm is used to optimize the multi-core extreme learning Machine the best parameters.Then,the FCM pre classified particle swarm optimization(PSO)algorithm for multi core extreme learning machine image segmentation is obtained.Experimental design analysis shows that compared with the FCM-KELM algorithm,the proposed algorithm has a longer segmentation time and a higher segmentation accuracy and achieves a better segmentation result.