Enhanced Detection of Whiteflies on coconut leaves through TERE and MASK R-CNN Integration

The infestation of whiteflies is very extensive to the cultivation of coconut trees and causes tremendous loss in agricultural production. Therefore, it is very important to identify whiteflies precisely and efficiently for timely pest management. In this work, a novel method for the detection of whiteflies on coconut tree leaves is proposed, which integrates Threshold-Enhanced Region Expansion (TERE) with Mask R-CNN. TERE uses Otsu's method to ensure good initial thresholding, followed by region growing to further refine the segmentation of possible whitefly regions. This will isolate areas of interest with least noise. Subsequently, instance segmentation and classification through Mask R-CNN are performed for the identification and localization of whiteflies within the segmented regions. In this work, experiments are conducted on a comprehensive dataset of coconut leaf images, where the accuracy achieved is 96% with a loss of 0.12, thus proving to be highly precise and reliable in whitefly detection with no over fitting.