סמינר מחלקה של תומר פניני - זיהוי מוקדם של ToBRFV בזרעי עגבניות באמצעות מערכת אוטומציה חכמה המבוססת על אלגוריתמים של HSI ולמידת מכונה
SCHOOL OF MECHANICAL ENGINEERING SEMINAR
Monday 22.07.2024 at 14:00
Wolfson Building of Mechanical Engineering, Room 206
Early Detection of ToBRFV in Tomato Seeds Using an Intelligent Automation System Based on HSI and Machine Learning Algorithms
Tomer Pnini
MSc. student under the supervision of Prof. Noam Koenigstein
School of Mechanical Engineering, Tel Aviv University, Tel Aviv, Israel
This study introduces an innovative method for early detection of Tomato Brown Rugose Fruit Virus (ToBRFV) in tomato seeds using hyperspectral imaging (HSI) combined with machine learning algorithms. The experiment involved gathering a comprehensive database of 30,000 seeds, meticulously captured with both Visible and Near-Infrared (VNIR) and Short-Wave Infrared (SWIR) cameras. Custom sample trays, an illumination arrangement, and a robotic arm for precise tray scanning were employed in the experimental setup. A key focus was on devising a methodology for band selection, crucial for developing cost-effective multispectral cameras. Seed samples were deliberately contaminated with ToBRFV, and hyperspectral imaging acquisition utilized custom trays and synchronized sensors. Data preprocessing involved segmentation, normalization, and construction of a tabular database.
Machine learning algorithms, particularly XGBoost, were evaluated for classification tasks, with feature selection using a Sequential Forward Selection (SFS) approach and statistical feature selection to enhance accuracy rates. Results demonstrated the effectiveness of hyperspectral imaging coupled with the XGBoost algorithm, achieving an accuracy of 98.6% when utilizing the entire spectral range. Furthermore, employing the XGBoost-based SFS approach for band selection highlighted the significance of specific spectral regions. Through iterative analysis, discrete bands were identified, focusing on channel minimization. Ultimately, retaining only 5 VNIR bands (455.6 nm, 880.4 nm, 545.8 nm, 767.0 nm, 535.2 nm) together with 4 SWIR bands (2023 nm, 1461 nm, 2142 nm, 1008 nm) achieved an impressive accuracy of 98.0%. Additionally, combining the 5 VNIR bands with only 1 SWIR band resulted in an accuracy of 96.4%.
These findings have significant implications for advancing multi-spectral cameras and improving disease detection in agricultural crops. Additionally, the research holds promise for detecting diseases in various other crop seeds, thereby enhancing the fight against plant diseases. Furthermore, it contributes to cost savings and has the potential to mitigate global food wastage.
