An Improved Deep Learning Model Implementation for Pest Species Detection

Abstract

Pests account for more than half of all known animal species and can be found in all types of environments. These pests are one of the main reasons behind the decline in crop yield. Accurate recognition of pests must be done so that timely measures can be taken according to the type of pest and the losses are reduced over time. As a result, Deep Learning has been utilized to identify pest species more quickly and accurately. In this paper, a Vision Transformer (ViT) based model is being used to detect the pest species more accurately. A large data set is being used containing 50 different species of insect pests. It incorporates well over 34,089 photos that are divided into 50 categories and have a naturally long-tailed distribution. A comparison of the performance of ViT is done with the ResNet model on the same data and it is found that the ViT model performs better than the ResNet model.