Lite-Agro: Exploring Light-Duty Computing Platforms for IoAT-Edge AI in Plant Disease Identification

The Lite-Agro study aims to deploy deep learning neural network models for pear disease identification through tree leaf image analysis on TinyML device. A case study on pear leaves is conducted with publicly available pear disease dataset. Quantitative comparisons are made between different datasets. Lite-Agro is a light-duty image computing detection solution that is tested for deployment on a microcontroller. The novelty of Lite-Agro, lies in the export of a lightweight TinyML, Tensorflow Lite model that is geared for low power applications on battery powered hardware. The goal is to find the best model that is custom selected for the application and achieves the highest accuracy. The study emphasizes finding a balance between size, accuracy and performance. In future iterations of the study, Lite-Agro is to be mounted on an unmanned aerial vehicle to be powered with solar panels. Modern low powered microcontroller devices are to be a staple implementation in Smart Villages.