Development Of Ground-based Sensor System For Automated Agricultural Vehicle To Detect Diseases In Citrus Plantations
1S. Sankaran, 2R. Ehsani, 2A. Mishra, 3C. Dima
1.
2. University of Florida
3. Carnegie Mellon University
An integrated USDA-funded project involving Carnegie Mellon University, University of Florida, Cornell University and John Deere is ongoing, to develop an autonomous tractors for sustainable specialty crop farming. The research teams have come together to develop an automated system for detecting plant stress, estimating yields, and reducing chemical usage through precision spraying for specialty crops. The goals of the automation process are to reduce the tractor-related labor costs, reduce the scouting labor costs, to improve equipment utilization, and reduce chemical usage. These systems can be used for a broad range of specialty crops.
The research work described in this paper was aiming towards detecting diseases in citrus trees. A platform was developed to monitor the health of citrus trees through multiple-sensing techniques. Currently, spectroscopic based techniques have been applied for disease detection. The spectral reflectance data from the diseased and healthy citrus trees were collected in the spectral range 350 to 2500 nm, with a resolution of about 3 nm. The spectral reflectance data will be analyzed using statistical models to classify the diseased samples from those of healthy samples.
Preliminary studies were conducted to classify diseases trees from those of healthy trees based on the spectral reflectance signatures. The principal component analysis was performed on the dataset containing the raw data, first derivative, and second derivative of the raw data (Savitzky -Golay method). The results indicated that when the principal components derived from the spectral reflectance datasets were used as the input features in the classification algorithm, the quadratic discriminant analysis yielded an overall classification accuracy of about 90% during the classification of diseased trees from those of healthy trees. The results signify the potential for utilizing multi-band spectroscopic techniques for disease detection in specialty crops. Further analysis on a larger dataset is ongoing to validate these findings.
