Deep Learning Based Detection of Rhinoceros Beetle Infestation in Coconut Trees Using Drone Imagery

This paper reports an end-to-end pipeline for detecting a specific infestation among coconut trees, caused due to the attack of rhinoceros beetle, which can critically harm the productivity, using drone imagery. The advantage of drone imagery lies in giving a bird’s eye view of the plantation which cannot be seen from the ground level. However, the challenge in processing drone imagery stems from the lack of depth information. The main objective is to detect and extract the individual tree-crown from an image that might contain up to 30 tree-crowns, to enable further analysis. The challenges lie in separating the crown in the presence of textured soil, shadows, companion plants. The data-set generated is composed of 1212 drone images containing 9727 individual tree crowns. In this work, we use Invariant Risk Minimization (IRM) for the first time in object detection and classification approach. Faster R-CNN model was used to extract candidate regions containing all individual crowns in the drone image while VGG-16 model was used to classify the detected crowns based on their health. For crown detection, the precision and recall score obtained was 97.30% and 92% respectively, while the classification accuracy obtained was 84.64%. This illustrates that analyzing drone images can be effectively used to monitor the well-being of a large plantation.