Remote sensing image classification based on adaptive ant colony algorithm

Aimed at solving the existing problems in most traditional classification algorithms applied in hyper-spectral classification, such as low speed, low accuracy, and difficult convergence, a hyper-spectral image classification based on an adaptive ant colony algorithm (ACA) is proposed in this paper. First of all, training samples are used to construct the paths selected by ants, and rule pruning is used to prune the paths irrelevant to classification and adaptively update paths. Then, the constructed paths are used to classify the hyper-spectral data. Finally, an accuracy evaluation of the classification is conducted. The hyper-spectral remote sensing data of Washington, D.C., was used to test the method. The method was also compared with principal component analysis (PCA), radial basis function (RBF) neural network, semi-supervised sparse discriminate embedding (SSDE), and adaptive sparse representation (ASP). Results show that the overall classification accuracy increases by about 7.6% using the adaptive ant colony algorithm compared with other algorithms, which improves the classification accuracy of hyper-spectral images efficiently.