UAV Remote Sensing for Disease Detection in Potato Leaf

Disease detection is one of the most common applications of precision agriculture since early detection is vital to prevent the spread of diseases and minimize the potential damage to agricultural production. Drone-based remote sensing can monitor large agricultural areas and identify anomalies in crop health, such as changes in vegetation index or thermal patterns. This study examines the use of UAV images (Mica Sense Red Edge-M sensor) to detect virus-affected (diseased) potato leaves at SASA Gogarbank Farm in Scotland using multispectral bands: Red, Green, Blue, Near-infrared, and Red-edge. Although the resolution is high due to the low flight distance in the field where potato crops are located, it is almost impossible to visually distinguish between healthy and diseased ones. Therefore, vegetation indices (VIs), widely used in agronomic fields, were calculated using the plants' reflectivity profiles. As a result of these calculations, it has been observed that SAVI emphasizes soil pixels, and VARI emphasizes flower pixels. In this direction, these two indexes, SAVI and VARI, are used to delete non- leaf pixels in the data before classification. In addition, it has been determined that NDVI has the highest standard deviation, and the band stack has been made with this index, and NDVI has been added as the 6th band. ROIs were determined by selecting 6 different potato plant species in the study area since the number of viruses detected in each plant species and the type differences between varieties in the study area where there were different potato varieties. According to the laboratory results provided by the SASA's guidance, it is known which plant is diseased and which is healthy. Accordingly, an object-based supervised, K-nearest neighbor (KNN) classification was performed. As it is supervised, the guideline provided by SASA is used for the training data process and the accuracy assessment. As a result of the classification, an accuracy of over 85% was generally obtained for each ROI. The results show that the UAV remote sensing and computed vegetation indices can distinguish between diseased and healthy leaves. Also, since the difference in reflectance between the diseased and healthy leaf is observed the most in the Red-edge band, the Red-edge sensors tend to give better results in differentiating the leaves.