Automatic Classification and Detection of Insect Pests Using Deep Transfer Learning

The growth of most significant field crops, including rice, wheat, maize, soybeans, and sugarcane, is inhibited by pest attacks, and numerous insect species decrease crop productivity. Additionally, manually recognizing and identifying pests takes time and effort. Therefore, early analysis and automatic detection and classification of insect pests are necessary to save crops from different insect pests. For this purpose, machine learning (ML) approaches have recently been widely used to classify and detect various insect attacks on crops. Misclassification of insect pests may lead to using the wrong pesticides, causing harm to agricultural yields and the surrounding environment. In this work, we employ nine deep learning (DL) pre-trained algorithms to record their performance in identifying, detecting, and classifying various insect attacks on crops. For this purpose, we adopted two popular ten-class crop pest datasets (small-scale and large-scale) to evaluate the different transfer learning (TL) algorithms. The TL algorithms selected for identifying and classi-fying insect pests are: XceptionNet, InceptionNetV3, GoogleNet, ResNet-50, ResNet-lOl, ShuffieNet, DarkNet-53, DenseNet201, and MobileNetV2. Our experimental results demonstrate that the XceptionNet and ResNet-50 TL algorithms performed ex-ceptionally well in identifying agricultural pests with 99.70% and 98.70 % accuracy, respectively, outperforming other TL algorithms.