Abstract:
"PotatoLens addresses the crucial issue of detecting blight diseases in potato crops, a significant threat to agricultural productivity and food security. Focusing on Sri Lankan agriculture, where potatoes are a staple crop, this project highlights the urgency of efficient, accurate early detection methods. Traditional practices fall short in timely and reliable disease identification, prompting a demand for technological innovation. Existing solutions, mainly relying on Convolutional Neural Networks (CNNs), offer limited understanding to end-users, underscoring a trust gap between AI disease detection models and farmers.
To resolve these challenges, PotatoLens incorporates a sophisticated AI model using Deep Learning and Explainable AI (XAI) techniques. A comprehensive dataset, featuring various stages of potato blight, underpins the model, ensuring robustness and representative learning. The project adopts an innovative approach by integrating CNNs for image-based disease detection and XAI for transparent decision-making processes. These technologies work in tandem to enhance the model’s predictive capabilities while making its outcomes understandable to farmers. By adding layers of interpretation and transparency, PotatoLens not only detects disease effectively but also bridges the trust gap in AI solutions.
Evaluation of PotatoLens involved rigorous testing using data science metrics such as accuracy, F1 score, precision, and recall. The model demonstrated high accuracy (over 91%) in disease detection, affirming its potential as a reliable tool for early blight identification. The inclusion of XAI further empowered users, offering insights into the AI's decision-making process, which proved crucial for user acceptance and practical application. These results underscore the system’s effectiveness in not only identifying disease with precision but also in fostering user trust through transparency."
