Classification of the Remotely sensed Images with Nearest Neighbor and Distribution of the Training Samples

In this paper a new method for classification of the hyperspectral and multispectral images with the nearest Neighbor (NN) training sample as well as the distribution of training samples in each class is proposed. In remotely sensed images classification, finding the best criterion for separating classes and assign an accurate label to the pixels is a major challenge. KNN (K nearest Neighbor) algorithm is a powerful supervised and nonparametric method for classification of the multi and hyerspectarl images. In this method Euclidian distance between unknown pixel and all training samples is calculated. Then, the K nearest training samples is separated. Finally, the unknown pixel classified in the class with the maximum training samples in nearest Neighbor. In the basic form of this method minimum distance is the main criterion for classification rather than other effective parameters. For instance, spectral distribution of the training samples near the unknown pixel can affect the classification. In this paper the distribution of the training samples in each class is added to the Euclidian distance criterion. Distribution is calculated based on the features of the training samples. This new factor can determine the most admissible growing direction in each class. Simulation and experimental results based on the hyperspectral images represent the effectiveness of this method.

Classification of the Remotely sensed Images with Nearest Neighbor and Distribution of the Training Samples Keywords:

Classification , hyperspectral images , KNN , Supervised , nonParametric