Hyperspectral Image Classification via a Joint Weighted K-Nearest Neighbour Approach

In this paper, we propose a simple yet effective classification framework to conduct hyperspectral image (HSI) classification based on K-nearest neighbour (KNN) and joint model. First, we extend the traditional KNN method to deal with the HSI classification problem by introducing its domain knowledge in HSI data. To be specific, we develop a joint KNN approach to solve the HSI classification problem by considering the distances between all neighbouring pixels of a given test pixel and training samples. Second, we exploit a set-to-point distance between neighbouring pixels and each training sample, and introduce this distance into the joint KNN framework. In addition, a weighted KNN method is adopted to achieve stable performance based on our empirical observations. Both qualitative and quantitative results illustrate that our method achieves better performance than other classic and popular methods.