Recent advances in neural networks have made great progress in addressing the hyperspectral image (HSI) classification problem. However, the overfitting effect, which is mainly caused by complicated model structure and small training set, remains a major concern when applying neural networks to HSIs analysis. Reducing the complexity of the neural networks could prevent overfitting to some extent, but also decline the networks' ability to extract more abstract features. Enlarging the training set is also difficult. To tackle the overfitting issue, we propose an abundance-based multi-HSI classification method. By applying an autoencoder-based spectral unmixing technique, different HSIs are firstly converted from the spectral domain to the abundance domain. After that, the obtained abundance data from multi-HSI are collected to form an enlarged dataset. Lastly, a simple classifier is trained, which is capable to predict on all the involved datasets. Taking advantage of spectral unmixing, converting the data from the spectral domain to the abundance domain can significantly simplify the classification tasks. This enables the use of a simple network as the classifier, thus alleviating the overfitting effect. Moreover, as much dataset-specific information is eliminated after spectral unmixing, a compatible classifier suitable for different HSIs is trained. In view of this, a several times enlarged training set is constructed by bundling different HSIs' training data. The effectiveness of the proposed method is verified by ablation study and comparative experiments. On four public HSIs, the proposed method provides comparable classification results with two comparing methods, but with a far more simple model.