CNN-Based crop disease and pest detection systems: Enhancing accuracy and efficiency in agricultural sustainability

Abstract

This paper highlights the effectiveness of classifying plant diseases using different Convolutional Neural Network (CNN) models. By combining CNNs with classification techniques like Support Vector Machines (SVM), accuracy improved to an impressive 98-99%  across various plant disease categories. CNNs automate feature extraction and classification, outperforming traditional methods in complex tasks.The paper compares Local Binary Pattern (LBP) with a total of 19 CNN architectures, including ResNet-101, Google Net, DarkNet-19, and others, demonstrating that CNNs consistently exceed 99 % accuracy in datasets for rice, corn, and jute. Our model achieved 99.9 % accuracy for Rice, 99.87 % for Jute, and 99.01% for Corn datasets when utilizing the ResNet-101 and DarkNet-19 CNN models. In agricultural areas like Bangladesh, this emerging method has the capability to completely transform crop disease management by facilitating early identification and prompt response, which would increase crop yields and food security. The work also makes real-time illness prediction accessible to farmers by introducing a Graphical User Interface (GUI). The potential of CNN-based systems to revolutionize precision agriculture, maximize resources, lower expenses, and empower farmers is highlighted in this study.

J. Bangladesh Acad. Sci. 49(1); 73-85: June 2025