Red Palm Weevil (RPW), Rhynchophorus ferrugineus (RPW), is recognized globally as a serious pest in many varieties of palm trees. The pest damages ornamental palms, endangering pedestrians and traffic, palm oil plantations in Southeast Asia, coconut palms across the Pacific islands, as well as date orchards throughout the Middle East. Economic damage due to this pest in the Middle East Gulf States is estimated at up to $25 million annually. The pest first appeared in Israel in 1999, in date orchards near the Dead Sea, and has since spread throughout the country.
Early detection of RPW infestation is challenging since no obvious signs appear until the tree crown collapses. When damage becomes serious, the crown suddenly dries up, and eventually leads to collapse of the whole plant. Existing early detection methods are expensive, labor intensive, and not fully accurate. These methods include pheromone trapping, manual inspection of trees to identify infection locations, and trained sniffer dogs to locate RPW infested trees. More recently acoustic sensors that pickup sounds of larvae in the tree trunks have proven moderately successful. However that method is costly, labor intensive, and requires damaging the tree to insert the sensor.
This work proposes a new approach for early detection of RPW based on remote sensing and Machine Learning. Preliminary examination of drone acquired imagery indicates that reflectance in the thermal band, with wavelengths longer than near infra-red, is stronger in infested trees than in clean trees. This effect is most pronounced in the central, sensitive frond (“lulav”) region of the tree canopy. After collecting a large data-set of both drone imagery and acoustic sensor data, a Deep Convolutional Neural Network classification model will be trained and validated in an attempt to recognize infested trees.