Remote sensing imagery classification using AdaBoost with a weight vector (WV AdaBoost)

AdaBoost is an excellent ensemble learning algorithm, which can be combined with many weak classifiers to form a strong classifier and thus improves the classification accuracy. However, the combination using AdaBoost focuses on the performance of base classifiers at overall level and ignores the performance at per-class level. This limits the improvement of classification accuracy when using AdaBoost and results in the problem of overfitting in later iterations. To minimize such disadvantages and inherit the benefits of AdaBoost, an improved AdaBoost algorithm with weight vector (WV AdaBoost) is proposed in this letter. Using weight vectors, each individual class is assigned a weight to represent the recognition power of base classifiers. C4.5 decision tree, Naïve Bayes, and artificial neural network (ANN) are used in AdaBoost and WV AdaBoost base classifiers training. The classification experiment performed for remote sensing (RS) imagery shows that WV AdaBoost can significantly minimize overfitting; it outperforms AdaBoost by yielding a much higher classification accuracy. WV AdaBoost can improve classification accuracy to the highest degree within a small number of iterations.