SMOTE-Based Weighted Deep Rotation Forest for the Imbalanced Hyperspectral Data Classification

Conventional classification algorithms have shown great success in balanced hyperspectral data classification. However, the imbalanced class distribution is a fundamental problem of data, and it regarded as one challenges tasks. To solve this problem, non-ANN based deep learning, namely SMOTE-Based Weighted Deep Rotation Forest (SMOTE-WDRoF) proposed paper. First, neighboring pixels instances are introduced spatial information datasets created by using SMOTE algorithm. Second, these fed into WDRoF model that consists rotation forest multi-level cascaded random forests. Specifically, used to generate feature vectors, which input subsequent cascade forest. Furthermore, output probability each level original stacked dataset next level. And sample weights automatically adjusted according dynamic weight function constructed results Compared with traditional learning approaches, method consumes much less training time. The experimental on four public demonstrate can get better performance than support vector machine, forest, combined convolutional neural network, rotation-based multiclass imbalance learning.