Texture based Image Species Classification with Deep Convolutional Neural Network

Our goal of study, texture based butterfly image classification with considering two standard deep learning networks and their shallower versions. Butterfly images involve only low and mid level texture information, which is typically learnt in the lower and mid-layers of the networks. We explore, whether smaller versions of these networks are robust enough to maintain a reasonable performance, as compared to the complete network, on the butterfly classification task. To identify the homogeneous texture region with texture class it belongs. We also demonstrate the effect of masking the background under the above scenario. Our approach yields high quality best results for the butterfly classification task, and we draw some interesting observations from our experimental analysis.