Agricultural pests' recognition using deep learning and ChatGPT

Abstract

The issue of global food security has become a pressing concern, with the growing population and the increase in demand for food. However, agricultural crop losses due to pest infestation pose a significant challenge to the sustainability of food security. To address this challenge, the adoption of smart agriculture practices stands as the optimal strategy for farmers. By leveraging on the artificial intelligence techniques alongside modern information and communication technology, farmers can effectively combat pest infestations and mitigate crop losses. Furthermore, the emergence of generative AI, exemplified by ChatGPT, represents a rapid advancement in technology. These systems have the capability to offer natural language explanations and tailored suggestions for users.

Hence, this paper introduces AgriPest, a mobile application designed to assist farmers with pest identification and management in agriculture. The application will explore the integration of computer vision techniques and natural language processing. In this paper, multiple Convolutional Neural Network (CNN) models, specifically DenseNet, MobileNetV2, EfficientNetV2B0 and Xception, were trained and compared to identify a suitable backbone model for the pest identification model. Additionally, analysis and fine-tuning processes were conducted on both the OpenAI GPT-3.5 turbo and the Gemini Pro Large Language Models (LLMs) to identify the most suitable candidate for constructing a chatbot application.

By utilising deep neural network, the application automates the classification of pests based on the input image. Each identified pest is populated with a selection of pesticides that are tailored to its specific characteristics, as well as other natural techniques for combatting infestation. Moreover, AgriPest is integrated with ChatGPT, to provide personalised and context-specific feedback to address targeted pest of interest.