Abstract
Recent advances in neural networks have made great progress in addressing thehyperspectral image (HSI) classification problem. However, the overfittingeffect, which is mainly caused by complicated model structure and smalltraining set, remains a major concern when applying neural networks to HSIsanalysis. Reducing the complexity of the neural networks could preventoverfitting to some extent, but it declines the networks' ability to extractmore abstract features. Enlarging the training set is also difficult. To tacklethe overfitting problem, we propose an abundance-based multi-HSI classificationmethod. By applying an autoencoder-based spectral unmixing technique, differentHSIs are firstly converted from the spectral domain to the abundance domain.After that, the abundance representations from multiple HSIs are collected toform an enlarged dataset. Lastly, a simple classifier is trained, which makespredictions over all the involved datasets. Taking advantage of the spectralunmixing, transforming the spectral features to the abundance featuressignificantly simplifies the classification tasks. This enables the use of asimple network as the classifier, thus alleviating the overfitting effect.Moreover, as much dataset-specific information is eliminated by the spectralunmixing, a compatible classifier suitable for different HSIs is trained. Aseveral times enlarged training set is constructed by assembling the abundancesfrom different HSIs. The effectiveness of the proposed method is verified bythe ablation study and the comparative experiments. On four datasets, theproposed method provides comparable results with two state-of-the-art methods,but using a much simpler model.