2:00 PM - 2:15 PM
[U04-02] Application of Remote Sensing, Machine Learning, and Spectral Measurement for Coffee Leaf Rust Early Detection and Classification in Rwanda
Keywords:Remote Sensing, Spectral Libraries, Coffee Plants, Machine Learning
Over the last decade, coffee farmers have been experiencing the problem of coffee leaf rust disease (Hemileia vastatrix) that has been diminishing their harvest in Rwanda, due to the unmatured defoliation caused by its pathogenic fungus. Different approaches were put in place but until now no effective approach is in place to tackle rust disease at an early stage.
We developed an algorithm to detect and classify coffee leaf rust by using Sentinel-2 images and spectrometer data. We applied a Decision Tree DT and Support Vector Machine SVM algorithms on Sentinel-2 images and spectrometer data respectively to classify coffee rust disease into 4 classes: healthy coffee, rust disease at a low level, medium, and high level. Band8, Band8A, and Band11 were selected for Sentinel-2 images, and we found that the following 10 bands 579.0, 578.5, 579.5, 578.0, 580.0, 697.0, 696.5, 697.5, 698.0, 696.0 of spectrometer data are sensitive to healthy and diseased coffee plants at an elevation angle of -900 (Nadir) and height of 1.5m.
In addition, we found that fertilizer, soil moisture, and water contribute tremendously to the spectral reflectance of healthy and diseased coffee plants. Another advantage of our algorithms is that we can distinguish coffee from other crops and forests.
We developed an algorithm to detect and classify coffee leaf rust by using Sentinel-2 images and spectrometer data. We applied a Decision Tree DT and Support Vector Machine SVM algorithms on Sentinel-2 images and spectrometer data respectively to classify coffee rust disease into 4 classes: healthy coffee, rust disease at a low level, medium, and high level. Band8, Band8A, and Band11 were selected for Sentinel-2 images, and we found that the following 10 bands 579.0, 578.5, 579.5, 578.0, 580.0, 697.0, 696.5, 697.5, 698.0, 696.0 of spectrometer data are sensitive to healthy and diseased coffee plants at an elevation angle of -900 (Nadir) and height of 1.5m.
In addition, we found that fertilizer, soil moisture, and water contribute tremendously to the spectral reflectance of healthy and diseased coffee plants. Another advantage of our algorithms is that we can distinguish coffee from other crops and forests.