"MulberryLab - Identification and Characterization of a Unique Mulberry (Morus sp.) Variation Endemic to Sri Lanka: A Study on Morphological and Genetic Traits"

Abstract

Mulberry cultivation plays an important role in agriculture and traditional medicine in Sri Lanka. A unique red mulberry variety, native to Sri Lanka, possesses significant nutritional and medicinal value. However, the visual similarity between this local variety and foreign mulberry types makes manual identification difficult, especially for farmers and non-experts. This often leads to misclassification, reduced market value, and improper agricultural decisions.

This project proposes an intelligent web-based system for automated mulberry variety identification and plant health analysis using deep learning and computer vision techniques. The system integrates two core models: a YOLOv10-based object detection model for identifying and counting Sri Lankan and foreign mulberries within an image, and a ResNet50-based image classification model for detecting common mulberry leaf diseases. The detection model was further enhanced by modifying the backbone with ResNet18 and incorporating the Convolutional Block Attention Module (CBAM) to improve feature representation and accuracy.

Users can upload images through a React-based frontend, while a Flask backend processes the requests and returns annotated images, variety classification, disease predictions, and descriptive information including nutritional benefits and usage guidance. The system was evaluated using both quantitative metrics such as accuracy and qualitative feedback from domain and technical experts. Experimental results demonstrate a noticeable improvement in detection performance, with accuracy increasing from 64.0% to 70.3% after architectural enhancements.

The proposed solution supports farmers, researchers, and agricultural stakeholders by providing a fast, cost-effective, and accessible tool for mulberry identification and health monitoring. This system contributes to the application of artificial intelligence in precision agriculture and offers a scalable foundation for future enhancements, including yield prediction and mobile deployment.