<![CDATA[With a projected global population of 9.7 billion by 2050, agriculture faces significant challenges in ensuring food security. One major obstacle is plant diseases that reduce crop yields by 40% per year. Previous research is often limited to disease detection in a single plant species, thus poorly reflecting multi-species needs in real agricultural practices. This research aims to develop and evaluate deep learning-based plant disease detection system using Convolutional Neural Networks (CNN) applied to two plant families, Solanaceae and Rosaceae. The dataset used was PlantVillage, containing 54,306 leaf images in JPEG format downloaded from GitHub, with data outside two families discarded during pre-processing. Three deep learning models were tested: transfer learning with InceptionV3 architecture and two custom CNNs (DFE and LCNN). LCNN model showed the best performance with training, validation, and testing accuracies of 99%, 99%, and 95%, respectively. In contrast, InceptionV3 achieved 96% training, 98% validation, and 92% testing accuracy, while DFE with 86% training, 94% validation, and 82% testing accuracy. Confusion matrix analysis showed difficulty distinguishing between healthy potatoes and potatoes with late blight, as well as cedar apple rust. These results highlights importance of developing specific model architectures rather than complex models for accurate multi-crop disease detection.]]>
Copyrights © 2024
