Design and Validation of an AI based Plant Stress Detection System in the EDEN ISS Environment

Abstract

This thesis explores the detection of abiotic stress in plants using the EDEN ISS dataset. It aims to develop algorithms for image preprocessing, identify key features for stress detection, and implement suitable anomaly detection methods. The primary objective is to enhance the accuracy and reliability of stress detection in controlled environments, particularly for applications in space missions where sustainable food production is essential. The report identifies the yellow pixel count and optical flow angle as critical indicators of plant stress. These features proved effective in distinguishing between healthy and stressed plants. The Green Threshold method emerged as the most efficient segmentation approach, while the Local Outlier Factor (LOF) method demonstrated high accuracy and reliability in anomaly detection. Extensive experimental tests validated the developed algorithms and models, confirming their effectiveness. The study also highlights the need for more extensive, diverse datasets. Future research should focus on expanding the dataset, improving algorithms, and automating preprocessing steps to enhance efficiency and applicability. The developed methods offer practical solutions for real-time monitoring and management of plant health in controlled environments and are particularly relevant for long-duration space missions. This research contributes significantly to controlled environment agriculture and lays the foundation for future innovations in plant stress detection.