Empoasca Pest Attack Classification on Tea Plantations Using Multispectral Imaging and Deep Learning

Empoasca Pest Detection Multispectral Imaging Green Spectrum Analysis AlexNet ResNet-50 Tea Plantation Pest Management

Authors

Downloads

This study aims to enhance the management of Empoasca pests in tea cultivation, a critical sector for Indonesia’s economy, by developing an innovative detection method. The challenge of pest infestations may significantly reduce tea production yields, and the misuse of chemical pesticides further compromises tea quality. We propose a novel approach that integrates multispectral imaging with Convolutional Neural Networks (CNN), specifically employing ResNet-50 and AlexNet architectures to accurately detect Empoasca infestations. We begin with the data collection process, followed by the development of the preprocessing model and evaluation of its performance. We classify tea leaves affected by Empoasca pests using spectral data obtained from a multispectral camera operating across Green, NIR (Near Infrared), REG (Red Edge), and RED channels. We evaluated various spectral channels and identified the green spectrum as the most effective for revealing visual characteristics, such as curled leaves associated with Empoasca damage. Experimental results demonstrated that ResNet-50 outperformed AlexNet, achieving a remarkable accuracy of 99% on the green channel, while AlexNet showed notable accuracy declines on other channel combinations. These findings underscore the effectiveness of the green spectrum and the superiority of ResNet-50 in achieving precise pest detection, offering a reliable technological solution for modern tea plantation management.