Hyperspectral image classification method based on neighborhood information deep learning

Abstract

The invention discloses a hyperspectral data classification method based on neighborhood information deep learning. The method comprises the steps of randomly dividing hyperspectral image data into atraining set and a test set; taking class attribution of the training set samples in the n * n neighborhood of each pixel point and the score of a front one main component of all the samples in the n* n neighborhood as the spatial information of each sample, putting the spectral information and the spatial information of each sample of the training set into a convolution neural network for modeltraining, putting the spectrum information and the space information of each sample of the test set into the model for classification result prediction. According to the method, the class attributionof the training set of each pixel point in the n * n neighborhood in a picture and the main component distribution of all the samples in the neighborhood serve as spatial information, spatial featureextraction is carried out on the neighborhood image through a two-dimensional convolution neural network, and fusion with the spectral dimension information is then conducted. The classification precision can be remarkably improved. The method has a good application prospect in the field of hyperspectral data classification.

Description

technical field [0001] The invention relates to the technical field of hyperspectral image processing, in particular to a hyperspectral data classification method based on deep learning of neighborhood information. Background technique [0002] Hyperspectral remote sensing data can simultaneously acquire spectral information and spatial information to form a three-dimensional data block, which has broad application prospects in many fields such as surface target detection, agricultural and forestry breeding guidance, and mineral development and detection. Traditional hyperspectral image classification generally uses spectral dimension information to classify pixels. Commonly used methods include: support vector machine (SVM), K-nearest neighbor algorithm (K-NN), neural network (ANN), decision tree (DT ) and Random Forest (RF) etc. [0003] The previous classification methods were mainly aimed at the classification of spectral information. However, this type of classificati...

AI Technical Summary

Problems solved by technology

However, these methods mostly involve complex hyperspectral image segmentation strategies, spectral filtering, and classification voting strategies. The classification results are greatly affected by parameter settings and the algorithm design is complex. Therefore, how to effectively use hyperspectral spatial neighborhood information needs further research. research and improvement

Images