Automated Pest Classification and Infestation Detection using CNN and Transfer Learning Techniques

Traditional agricultural practices have historically been the bedrock for maximizing crop production in the agricultural sector, which now plays a crucial role in maintaining food supply. Deep le arning approaches, however, have become more well-known in the context of modern research due to their success in classifying a variety of image data. In this study, we use Convolutional Neural Networks (CNNs) to identify and categories nine different pest groups that are frequent in common crops. We carefully use image augmentation techniques to solve the difficulties of handling large datasets. We rigorously assess the accuracy and performance of the many convolutional neural network designs that are incorporated into the architectural configurations of our CNN models. Additionally, we use transfer learning strategies into the framework for the study of the pest dataset in order to improve our strategy. A comparison of the CNN-based approach and transfer learning, specifically using pre-existing architectural models, highlights an important conclusion, which is that the combination of CNN and transfer learning results in a remarkable testing accuracy of
$98{{\% }}$
. This outcome, which was attained by training on
$80{{\% }}$
of the data and testing on
$20{{\% }}$
of the data, highlights the method's strong potential.