Plant Disease Prediction Using Deep Learning Techniques

With the rising importance of agriculture in ensuring food security, early disease detection is crucial to mitigate yield losses and economic impacts caused by crop diseases. In this paper, a lightweight mobile application, BotaniCare, is presented which aims at revolutionizing agricultural practices by using deep learning techniques to accurately detect and diagnose diseases in plants. The system has been trained on a dataset of images of both healthy plants and those affected by various diseases. Through the implementation of the model, the system could recognize patterns and anomalies in plant health, enabling precise identification of plant diseases with minimal human intervention. The mobile application also included remedial actions to health the plants from the identified disease. The development of the plant disease prediction system involved data collection, pre-processing, model selection, testing and evaluation. The model was trained on labeled data and evaluated using appropriate metrics to ensure the reliability of the model. Different CNN architectures were compared and evaluated to be able to choose the most suitable one. By using transfer learning, MobileNetV2 was used in BotaniCare and a training accuracy of 98.7% and a validation accuracy of 96.4% was achieved during the model development and evaluation process. BotaniCare was thoroughly assessed using real-life images and validated against expert diagnosis, demonstrating its high accuracy and reliability in disease prediction. It is anticipated that the mobile application will be widely used by farmers in Mauritius to identify the frequent diseases of the common plants and apply appropriate remedial actions.