Genetic Optimization Of SOFM Neural Network In Image Segmentation

It’s a hot problem in computer vision research area that how to organize and combine the image data reasonably and effectively and how to use artificial intelligent and knowledge discovery in the images classification. Self-organizing Feature Map(SOFM, Self-organization Feature Map) Neural Network is a important self-organizing competitive learning neural network model. The SOFM network can project multi-dimensional data on a low-dimensional regular grid, so than it can be utilized to explore properties of the large data. But the process of weight updated to SOFM neural network self-learning lacks of global optimal. The evolution of optimization method by traditional GA is slow and has premature phenomenon. The genetic algorithm based on Lamarckian learning has been purposed in this paper can be used in optimization of updating of SOFM network.. The optimized SOFM is used in image segmentation. The main work of this paper is as follows:By using the genetic algorithm based on Lamarckian learning (LGA, Lamarckian Genetic Algorithm) to update the parameters of the SOFM neural network weights. Its chief points including improvement of updating process of network weights, in the process. Quantizing error and Pearson correlation coefficient are used in this paper as fitness function of the genetic operation. The optimized SOFM network is applied for texture image segmentation and the results of segmentation is better than the traditional self-organizing feature map neural networks.It is achieved that the SAR image segmentation based on combination of watershed algorithm and LGA-SOFM network. It will spend too much time and resources that using directly LGA-SOFM network for large amounts of data cluster. To remedy this, the watershed algorithm is used for pre-process of image, then data of combination pre-process of image and the feature through Contourlet transform is clustered by using LGA-SOFM.