Classification of Insect Pest Species using CNN based Models
Nikita Agarwal, Tina Kalita, Ashwani Kumar Dubey
Abstract
The group of animals that is considered to be the most diverse on this planet are insects, with an estimate of 5.5 million to 10 million species. Insects falling under the category of pests are the focus of this paper. Pests transmit and carry various diseases that are harmful not only for the crops but also for the population that is consuming them. Hence, early recognition of these insect pests is essential. In this paper, IP102 dataset is being used, which is a collection of insect pest species. The IP102 dataset consists of 102 species, having more than 75,000 images. Five different pre-trained Deep Learning models, namely, DenseNet-169, MobileNetV2, ResNeXt-50, VGG-16, and VGG-19 are being used in this paper. The performance of these models is then being evaluated on IP102 dataset for the identification of insect pest species. The accuracy of each model achieved is further compared to obtain the one with optimal performance.