Remote sensing image classification by the Chaos Genetic Algorithm in monitoring land use changes

Abstract

In order to improve the accuracy of monitoring land use changes, the Chaos Genetic Algorithm was proposed. The Chaos Genetic Algorithm has the capability of self-learning; hence through the input samples the global optimization clustering center was found. And then the clustering center was employed to classify the view figure of the remote sensing image. In this process, the ergodic property of chaos phenomenon is used to optimize the initial population; so it can accelerate the convergence of Genetic Algorithms. Chaotic systems are sensitive to initial condition system parameters. In order to escape from local optimums, the chaos operator was applied to optimize the individuals after the process of selection operator, variation operator and crossover operator. The Chaos Genetic Algorithm was applied to classify the TM image of Huainan. Moreover, the classification of the Parallelepiped and Maximum likelihood and Standard Genetic Algorithm methods are contrasted with it through the confusion matrix. The confusion matrix demonstrated that the overall accuracy and the Kappa coefficient of Parallelepiped, Maximum likelihood and Standard Genetic Algorithm methods are respectively 70% and 0.625%, 76.53% and 0.707%, and 82.13% and 0.777%. It also showed that the Chaos Genetic Algorithm was superior to the two traditional algorithms and the Standard Genetic Algorithm method, whose overall accuracy and Kappa coefficient reach 88.26% and 0.853% respectively.