ANNOUNCEMENTS
Unmanned Aerial Vehicles (UAVs) are increasingly being used for agricultural crop monitoring, especially for assessing crop health, disease detection, phenotyping and crop counting. Multispectral imagery allows to extract useful information about the crops that are not visible to human eyes. Such information is used in precision agriculture in order to offer farmers with site-specific information about the crop health, this in turn helps them monitor, plan and manage their farms more efficiently. The study focuses on crop health assessment and image segmentation for ground cover classification. This study uses multispectral images of the crop pineapple for two dates captured by Micasense RedEdge-MX sensor to assess the crop health by deriving various vegetation indices such as Normalized Difference Vegetation Index (NDVI), Normalized Difference RedEdge Index (NDRE) and Optimised Soil Adjusted Vegetation Index (OSAVI). The RedEdge data from the raw images were converted into radiance and then finally to reflectance maps in order to generate different vegetation indices to assess the crop health. The RedEdge band, located between the Red and NIR bands captured by the sensor plays an important role in deriving the plant chlorophyll content accurately. Further this study also uses Machine learning algorithm to perform semantic segmentation of the ground cover to detect and segment the crop, shadow and soil pixels. Manual histogram thresholding, k-means clustering and machine learning algorithms Random Forest and Support Vector Machine have been used for image segmentation. The results show that the chlorophyll content captured using NDRE was dramatically lower when compared to NDVI. It also reflects that the crop health deteriorated in January when compared to November. Comparing the two algorithms for image segmentation, Random Forest Classifiers does well with 98% accuracy with AUC of 1.00 and 0.99, while SVM has the accuracy of 97% with AUC of 0.99 and 0.97.
Keywords – Precision agriculture, Crop health, Vegetation indices, RedEdge, Image segmentation, Machine learning.
