Application of MobileNets Convolutional Neural Network Model in Detecting Tomato Late Blight Disease

Abstract

Late blight (LB) disease causes significant annual losses in tomato production. Early identification of this disease is crucial in halting its severity. This study aimed to leverage the strength of Convolutional Neural Networks (CNNs) in automated prediction of tomato LB. Through transfer learning, the MobileNetV3 model was trained on high-quality, well-labeled images from Kaggle datasets. The trained model was tested on different images of healthy and infected leaves taken from different real-world locations in Mbeya, Arusha, and Morogoro. Test results demonstrated the model's success in identifying LB disease, with an accuracy of 81% and a precision of 76%. The trained model has the potential to be integrated into an offline mobile app for real-time use, improving the efficiency and effectiveness of LB disease detection in tomato production. Similar methods could also be applied to detect other tomato infections.

Keywords:  MobileNets; convolutional neural networks; plant diseases detection; image classification; transfer learning