The Improved Ant Colony Rules Mining Algorithm And Its Application Of Remote Sensing Classification

Remote sensing image classification method has been an important part of the field of remote sensing research. When the study area is larger, more complex land cover types, the current classification methods are often difficult to obtain a higher classification accuracy, while the artificial intelligence methods and techniques in dealing with complex issues has obvious advantages. So, it is necessary to introduce the artificial intelligence methods to improve the remote sensing classification accuracy. As one of swarm intelligent optimization algorithms, the ant colony algorithm has been successfully applied in the remote sensing data classification, provides a broader idea for remote sensing image classification. However, the convertional Ant-Miner algorithm takes long computation time with a slow convergence in mining remote sensing image classification rules. We proposes an new ant colony algorithm based on conversional Ant-Miner algorithm, this study selects the Changsha-Zhuzhou-Xiangtan urban agglomerations as a case study area and use Landsat TM image in1996-2006, get the land cover information and reveal the changes. The main conclusions are as follows:(1) Ant colony rule mining algorithm can be better found the optimal solution, is robust, but Ant-Miner algorithm also has some deficiencies caused by the ant colony system defects, such as the streamlining of the classification rules set poor, vulnerablelocal optimal solution, the calculation time is longer.We proposes an new ant colony algorithm based on conversional Ant-Miner algorithm.Firstly,Improve and perfect the formula is not defined and the definition of unreasonable rules validity measure formula. Then, the convertional Ant-Miner algorithm is modified about the strategy of pheromone update by using new pheromone concentration update item and pheromone evaporation coefficient. Lastly, the mutation operator is introduced in algorithm solve. The new ant colony algorithm accelerates effectively the evolutionary process and shorts the calculation time.(2) In order to to verify the new algorithm, this study selects the Changsha-Zhuzhou-Xiangtan urban agglomerations as a case study area and use Landsat TM image in1996-2006. The results indicate that the new algorithm obtains the simpler forms of classification rule and reduce the computation time. Improved Ant-Miner algorithm classification results of the overall accuracy can reach more than85%, Kappa coefficient above0.81, the remote sensing image classification using improved Ant-Miner algorithm is more accurate and efficient than conversional Ant-Miner algorithm、 decision tree method and maximum likelihood method.(3) Based on the improved Ant-Miner algorithm,we get the remote sensing image classification of Changsha-Zhuzhou-Xiangtan urban agglomerations core area, reveals the Changsha-Zhuzhou-Xiangtan urban agglomerations core area the complex land cover changes. The results showed that:from1996to2006, the study area around Changsha、 Zhuzhou and Xiangtan City land cover change are most obvious,construction land has maintained a high growth state, its dynamic degree is active; forest land, farm land and water has little change, high weak of dynamic degree, forest land and farm land is the main source of construction land and bare land; bare land area has large fluctuation, are faster in the transfer and new increase, has the most strongly dynamic degree, the minimum stability and the highest degree of development and utilization.