Hyperspectral image classification based on pre-post combination process

Aiming at the problem of poor classification accuracy of traditional machine learning algorithms based on spectral information analysis, this paper proposes a hyperspectral image classification method based on pre-processing before classification and processing optimization combination after classification. Firstly, the original samples are subjected to Gaussian filter and Linear discriminant analysis for reducing noise and dimensions. Then, the data is initially classified by traditional machine learning algorithms such as k-nearest neighbor(KNN), Support Vector Machine (SVM), sparse representation-based classifier (SRC)or multiple logistic regression (MLR). Combining local pixel spatial information to determine the confidence of the prediction labels. Finally, the initial prediction label is corrected by a continuous multi-layer neighborhood optimization layers to obtain a final classification label. Comparative experiments were performed on multiple hyperspectral remote sensing databases such as Indian Pines. The experimental results show that the proposed method has obvious performance improvement in classification accuracy and time efficiency, which has a certain degree of robustness in the process of combining with different classifiers.