Enhancing Plant Disease Detection with CNNs and LLMs: A Comprehensive Approach to Diagnosis and Mitigation

Enhancing Plant Disease Detection with CNNs and LLMs: A Comprehensive Approach to Diagnosis and Mitigation

Paper in proceedings (conference paper)

This study presents a novel approach to plant disease detection by integrating a Convolutional Neural Network (CNN) with an open-source Language Model (LLM) within a user-friendly web application. From a custom-built dataset assembled using open access sources, consisting of 48 classes representing various plant diseases and healthy specimens, the CNN model achieves an impressive accuracy of 99.73% on the test set. The framework employs a robust experimental setup, including meticulous data partitioning and hyperparameter tuning, to ensure effective model training and evaluation. While CNN demonstrates exceptional performance in detecting well-represented diseases, challenges in accurately classifying underrepresented classes are identified, emphasizing the need for data augmentation strategies to enhance model robustness. The integrated LLM enhances user interaction by providing real-time insights and actionable recommendations based on CNN predictions, making the tool accessible to users with varying agricultural expertise. Further work aims to refine the system through dataset expansion and advanced training techniques, ultimately positioning this tool as an asset for sustainable agricultural practices.

This study presents a novel approach to plant disease detection by integrating a Convolutional Neural Network (CNN) with an open-source Language Model (LLM) within a user-friendly web application. From a custom-built dataset assembled using open access sources, consisting of 48 classes representing various plant diseases and healthy specimens, the CNN model achieves an impressive accuracy of 99.73% on the test set. The framework employs a robust experimental setup, including meticulous data partitioning and hyperparameter tuning, to ensure effective model training and evaluation. While CNN demonstrates exceptional performance in detecting well-represented diseases, challenges in accurately classifying underrepresented classes are identified, emphasizing the need for data augmentation strategies to enhance model robustness. The integrated LLM enhances user interaction by providing real-time insights and actionable recommendations based on CNN predictions, making the tool accessible to users with varying agricultural expertise. Further work aims to refine the system through dataset expansion and advanced training techniques, ultimately positioning this tool as an asset for sustainable agricultural practices.

CNN; LLM, Plant Disease Detection

CNN; LLM, Plant Disease Detection

SHARMA, S.; TIWARI, D.; GARG, A.; KAUSHAL, A.; DUTTA, M.; MEZINA, A.; FROLKA, J.

26.11.2024

Meloneras, Gran Canaria, Spain

978-3-8007-6544-7

ICUMT 2024; 16th International Congress on Ultra Modern Telecommunications and Control Systems and Workshops

13

18

6