One-class classification for ginger plant growth monitoring

Tan, Ernest Cong Ying (2025) One-class classification for ginger plant growth monitoring. Final Year Project, UTAR.

PDF Download (14Mb)

Abstract

This project introduces a ginger plant monitoring system, that trace the ginger plant growth from week 1 to week 20 which is the lifespan of the ginger plant growth in yield. I present a anomaly detection framework for ginger plant growth monitoring, integrating precise segmentation via YOLOv8n-seg with a prototypical few-shot learning model for one-class classification. To address the challenges of limited labelled anomalous data and environmental variability, the system focuses on learning representations of healthy plant growth and identifying deviations without the need for extensive anomaly datasets. The preprocessing stage isolates plant regions to reduce background interference, enhancing feature extraction for the classification task. The prototypical network, trained exclusively on normal samples, enables anomaly detection by measuring feature space distances to learned prototypes, facilitating the identification of subtle growth irregularities such as stress or disease. A Python Flask-based deployment platform allows for both manual image uploads and real-time monitoring, providing immediate feedback for agricultural interventions. Experimental results demonstrate that the proposed method outperforms traditional convolutional neural network classifiers and unsupervised clustering approaches in terms of robustness and accuracy under varying environmental conditions. This work contributes a scalable, efficient, and practical solution for improving crop yield, reducing labour dependency, and advancing sustainable precision agriculture.

Actions (login required)