Performance effect analysis for insect classification using convolutional neural network

It is necessary to develop an automated insect classification method in order to create high added value of insects. Recently, image classification work applying deep learning shows the best result of classification performance. In this paper, we use convolutional neural network to classify insects and examine the effect of the classification performance with the varying number of kernels in the convolution layer and with the different image datasets. Experiments were conducted on 27 insect classes of ImageNet dataset with AlexNet. We confirmed that the classification performance of neural network improves where insects are well represented at the center of the image, and when the number of convolution kernels increases.