UAV Multispectral Imaging
The project focuses on the development of a UAV platform-based multi-sensor system for early detection and monitoring of powdery and downy mildew in cucurbit crops.
Cucurbit Downy Mildew (DM) and Powdery Mildew (PM) are some of the most important diseases of cucurbits worldwide, causing severe reductions in yield and loss of fruit quality. In addition to employing host plant resistance, fungicide applications are used for crop protection, and initiated when those diseases have been detected in a field or neighboring county or state. Due to the virulence of these pathogens, scouting and treatment is essential to reduce marketable losses. Traditional scouting requires walking fields and manually inspecting plants for symptoms and signs of infection. However, this approach is very labor intensive, particularly on large-scale farms, and relies on the experience of the scout to be able to recognize signs of those diseases.
The goal of this project is to design a UAV platform that uses a sensor array to detect and pinpoint signs of DM/PM diseases on cucumber and pumpkin crops. The use of this technology will have several advantages. First, different sensors (RGB, IR, Multi-Spectral, etc.) will allow for early disease detection, possibly even before they are noticeable to human eyes. Secondly, the UAV will require minimal human supervision, and be able to scout crops more frequently and thoroughly than before. Lastly, after initial disease detection, the UAV will remain useful by monitoring crop health and helping to evaluate fungicide efficacy and optimize sprayer operation and coverage. The proposed technology could also be adapted in the future to identify and quantify damage caused by diseases, insects, and weeds on different crops.
Cucumber abd Pumpkin crops have been made available for the project at the OSU Waterman Farm and at the OSU Western Agricultural Research Station. MS students have collected multispectral images of the crops under healthy and unhealthy considtions (plants affected by PM and DM). These images are being used to derive classifiers capable of detecting the diseases.
Students are working on extracting significant features from the immages that classifiers can use to cathegorize the status of the crop (healthy/unhealthy). In particular, for feature extraction, Local Binary Pattern (LBP), Gray-level Co-occurrence Matrix(GLCM) are some of the methods under investigation. On the other hand, K-nearest neighbors (KNN), Logistical Regression, Support Vector Machine(SVM), Neural Networks and Random Forest algorithms are being used as classifiers. Using the set of images previously described (half for training the algorithms and half for validation) a 90% accuracy was achieved in the crop healthy detection.
Current and Future Work
So far, only few features and hyperspectral bands have been used to build the classifiers, so students are working on improving the methodlogy by considering more bands, combination of bands (e.g. vegetation indeces) and applying more methods for feature extraction.
If you are interested in
- feature extraction;
- object classification;
- deep learning and machine learning;
- hyperspectral imaging;
please contact me to join the team.
Lisa Fiorentini, Dept. of Electrical and Computer Engineering, OSU
Wladimiro Villarroel, Dept. of Electrical and Computer Engineering, OSU
James Jasinski, Dept. of Extension, OSU
John Fulton, Dept. of Food, Agricultural, and Biological Engineering, OSU
Sally Miller, Dept. of Plant Pathology, OSU