Study Of Classification From Remote Sensing Image Based On Ant Colony Optimization

Image classification is a process which simplifies the complex phenomenon to a small number of general categories. And it is an important way to extract useful information for the target recognition. Today the volume of data is growing rapidly. In particular, it is important to extract the target and some other interesting parts from the background image effectively and improve the efficiency and effectiveness of image classification. The other, because remote sensing images usually have these properties: large gray-class, great information, fuzzy border, complex structure of target and so on. These make it be very difficult to classify remote sensing images accurately. So the research of remote sensing image classification has become an important topic which has the theoretical and practical value.Firstly, this paper introduces the background and development of the classification of remote sensing images. Later it expounds the origin of Ant Colony Optimization (ACO), the basic principle, the flow of algorithm and some typical improved forms. On this basis, an improved Ant Colony Optimization based on dynamic control of solution construction and mergence of local search solutions is put forward. Then it introduces the Markov nature of digital image. Afterwards it founds the fitness function according to Markov nature. The fitness function involves the characteristic of images (including gray, variance, entropy, energy and fractal) and the measurement of similarity (including direction of the gradient, local relevance, distance of pheromone). In the end, above the classification rough result of K-means clustering, it establishes the model of precise classification of remote sensing image using Ant Colony Optimization.This paper has some innovation in overcoming slow convergence and easiness to fall into the local optimal solution of ACO. Experiment results show that the proposed method has higher precision of image classification comparing with the traditional algorithms. In this paper, the ant algorithm research will expand the range of the handling and application of some other problems, such as path planning, combinatorial optimization questions and so on.