Butterfly Species Identification Using Convolutional Neural Network (CNN)

Abstract

Butterflies are important from aesthetic, ecosystem, educational, health, economic, scientific and intrinsic value in Malaysia. One of the popular significance of butterflies is as a model to understand the effect of habitat loss and environmental change. Current approach in image processing for butterfly identification is not efficient due to complicated butterfly shapes. Gathering, recognizing and archiving specimen images physically is tedious and costly for entomologist. Hence, the need to have an application that can accelerate the process using a technique that easy to understand will definitely solve the problems. Nowadays, dataset consist a lot of noise or too small to suit to latest application. Thus, a study of butterfly species identification using image processing technique and Convolution Neural Network (CNN) is proposed. This research focuses on GoogLeNet a pre-trained model of CNN architecture. Four species of butterflies which that are commonly found in Asia which is Black Veined Tiger, Chocolate Grass Yellow, Grey Pansy and Plain Lacewing used in this research. The testing conducted reflected 97.5% overall identification accuracy on one hundred and twenty images of four types of butterflies.