Improvement Adaptive Gentic Algorithm Hyperplane Classificaton For Remote Sensing Image

As the foundation of remote sensing data application, classification techniques for remote sensing has theoretical and practical meaning to acquire the information of land surface environmental change rapidly and accurately. Due to the complexity of data, limitations of classifiers and some other factors, traditional classification methods still hard to obtain a higher classification accuracy. While the artificial intelligence methods,which characterized as finding the global optimum, learning from itself results, self-adapting, etc, have obvious advantages in dealing with the remote sensing complex data. Gentic algorithm hyperplane classification is a typical artificial intelligence method. We found that there were some problems in processing remote sensing data with this method. To solve the problems in this algorithm, the paper proposed an improvement adaptive gentic algorithm hyperplane (HP-IAGA) classification method. We verified and analyzed the new algorithm with the Landsat TM image data of Changsha-Zhuzhou-Xiangtan(CZT) urban agglomeration, and obtained main conclusions are as follows:(1) Proposed a new classification algorithm-Improvement Adaptive Gentic Algorithm Hyperplane(HP-IAGA). Sigmiod function was introduced into this algorithm, letting genetic parameter to nonlinear adaptive adjustment. Repeated training setting was not required and operation of algorithm was simplified. Meanwile, positive feedback mechanism of adaptive value made outstanding individuals’ viability more strong, algorithm’s global optimization ability and the rate of evolution convergence was enhanced. The improved algorithm is able to mine the rules of classification more quickly, more stably and more reliably.(2) Verified the new algorithm by use the mult-phase remote sensing data of CZT urban agglomeration core area. The classification accuracies of each phase of remote sensing data are all above85%. Compared with simple gentic algorithm hyperplane, decision tree classification and maxium likelihood, the accuracy was improved evidently.(3) Got the landcover types of CZT urban agglomeration effectively in the application study of HP-IAGA., and revealed the changing process in the study area. In the last fifteen years, urban districts of CZT three cities have been continuously sprawling, construction land has increased; areas of unexploited land, as a transition state between other land types to shift into construction land, fluctuated relatively markedly. Woodland decreased rapidly in the begining period, then slowed down in the later period. This reflects that ecological protection has achieved initial success.