Crop Disease Detection Using Lightweight Deep Learning Model for Smartphone

This research presents the design and implementation of a smartphone-based application for the detection of crop diseases using deep learning techniques, aiming to provide farmers with an accessible, real-time solution for plant health monitoring. A customized version of the PlantVillage dataset, which includes additional manually curated and augmented images, was used to train a DenseNet201 convolutional neural network (CNN) on Google Colab. The trained model achieved an impressive 96% accuracy in classifying multiple types of crop diseases. To ensure accurate input validation, a secondary model was developed using Google’s Teachable Machine to distinguish between leaf and non-leaf images, achieving 99% accuracy. Both models were converted into the TensorFlow Lite (.tflite) format to optimize performance on mobile devices. The application was developed using the Flutter framework in Visual Studio Code and deployed on Android, allowing users to capture or upload images for instant disease diagnosis without requiring a constant internet connection. This dual-model approach ensures robustness and improves user confidence by pre-validating input images before disease predictionThe system's exceptional accuracy, usability, and deployment ease make it especially beneficial for smallholder farmers and agricultural extension professionals. In line with the objectives of digital farming and precision agriculture, this approach provides better validity, usefulness, and accessibility than previous studies in the sector.