Enhancing training performance of convolutional neural network algorithm through an autoencoder-based unsupervised labeling framework for mineral exploration targeting