Combining two-trace two-dimensional correlation analysis and convolutional autoencoder-based feature extraction from an entire correlation map to enhance vibrational spectroscopic discrimination of geographical origins of agricultural products