Research On Remote Sensing Image Clustering Based On Particle Swarm Optimization

The development of remote sensing technology has played an important role for people to discover and observe the earth. Satellite digital images provide abundant observational data to about earth observation. In order to utilize remote sensing data effectively, it is need to transform the spectral information of remote sensing image to the classification information of users. Image analysis, classification and interpretation are needed in the transformation process. In the field of remote sensing image classification, the process of the unsupervised classification namely clustering requires less iteration and only asked to find natural groups of images. It is a challenging research area of remote sensing image automation clustering by computer. With the development of artificial intelligence technology, kinds of intelligent models and algorithms are applied into the exploration of remote sensing image clustering, which improve the accuracy of clustering a lot.Particle swarm optimization is one of the swarm intelligence algorithms, which has good adaptive self-organization, simple and efficient optimization capacity of location of groups. This thesis mainly focuses on the remote sensing image clustering based on particle swarm optimization. There are always mixed pixels in remote sensing image in view of the sensor resolution and complex terrain reasons. In order to overcome this problem to improve the accuracy of clustering, an improved mixed pixel decomposition method based on maximum entropy is proposed and it works well not only for the linear mixed end-data but also the nonlinear ones. The abundance distribution is estimated in this model in addition to the end-data determined. The estimation provides the basis of membership for the fuzzy particle swarm clustering, and it avoids the bad influence of the hard classification to the result. What's more, the quantum computation is introduced in this paper. The particles are encoded with quantum bits and updated with quantum rotating gate. Besides, a mutation operator is realized by quantum not-gate in the evolving process to avoid the premature convergence of the algorithm. In the section of experiments, the LANDSAT multispectral remote sensing images are adopted. Firstly, eigenvector of image is extracted through the principle component analysis, wavelet decomposition and gray level co-occurrence matrix method. Subsequently, the basic PSO and QFPSO algorithm are applied in the experiment of remote sensing image clustering. Paper compares the result of the two algorithms and it proves that the QFPSO gets the better clustering result.