Paddy Plant Stress Identification Using Few-Shot Learning Framework

The efficient identification of stress in paddy plant leaves is paramount for optimizing agricultural resources and ensuring robust crop yields in smart agriculture practices. This research explores the potential of few-shot learning (FSL) to address the inherent challenges posed by limited training data in stress identification. Three distinct FSL approaches - Siamese network, Matching network, and Model-Agnostic Meta Learning (MAML) - are evaluated for their accuracy in stress detection. The study introduces the importance of stress detection in smart agriculture and the challenges of limited training data. It then covers the methodology in five stages: dataset description, data pre-processing, FSL implementation, accuracy evaluation, and reporting. Among the FSL models, the Matching Network stands out with an impressive accuracy of 86% for 6-way 1-shot learning. This surpasses the performance of a Convolutional Neural Network (CNN) tested with a larger shot size (270-shots), which achieved an accuracy of 81%. These comparative results underscore the potential of FSL techniques in achieving precise stress identification, even when working with limited data. The objective of this research is to contribute valuable insights towards enhancing the efficiency of stress prediction in paddy leaves, thereby fostering healthier and more productive paddy crop production in smart agriculture. The findings presented here aim to inform the development of effective stress detection systems and advance the field of precision agriculture.