Low-Cost Smartphone-Based Plant Disease Diagnosis for Zimbabwean Farmers Using Transfer Learning and Crowdsourced Image Data

Timely and accurate identification of crop diseases is vital for improving food security and farmer livelihoods, particularly in low-resource agricultural settings.This study presents a lowcost, smartphone-compatible plant disease diagnosis system designed specifically for Zimbabwean farmers.The system integrates transfer learning with the MobileNetV2 architecture and leverages a hybrid dataset composed of curated PlantVillage images and 400 crowdsourced leaf images collected from smallholder farmers in Zimbabwe.Following preprocessing and augmentation, the data was used to train a lightweight convolutional neural network via a two-stage transfer learning approach.The model achieved a test accuracy of 91.0%, with strong precision, recall, and F1-scores across six classes.A web-based prototype was developed using Streamlit and deployed via Ngrok, allowing real-time disease prediction through browser-based image uploads, simulating field use on mobile devices.Compared with previous studies, this work demonstrates competitive accuracy while emphasizing practical deployability and contextual relevance.The inclusion of locally sourced images significantly improved real-world performance.The approach empowers Zimbabwean farmers with rapid, accurate, and actionable plant disease diagnosis, supporting sustainable agriculture and food security.