Scalable End-to-End RF Classification: A Case Study on Undersized Dataset Regularization by Convolutional-MST