Image acquisition and deep convolutional neural network incorporated cost-effective automated nursery disease detection method for the tea industry

Abstract

Tea cultivation, a globally significant industry across over 60 countries, requires effective management of tea nurseries for robust plant growth and disease control. Implementing automated disease detection in tea nurseries using advanced technologies enhances efficiency, precision, and sustainability compared to manual inspection methods. An image acquisition system incorporating a Deep Convolutional Neural Network (DCNN) was developed as a cost-effective automated solution for nursery disease detection. First, images of Blister Blight (BB) disease in tea plants were collected primarily from various estates. These images were then preprocessed, labeled, and augmented to create a robust dataset. Two models based on the YOLOv5 architecture were trained, with the second model benefiting from dataset augmentation and enhanced labeling. The Improved Model achieved a mean Average Precision (mAP) of 80.6% at a 0.5 threshold, demonstrating its high accuracy in BB detection in tea nurseries. It was integrated with the image acquisition system to enable continuous monitoring of tea leaves in nursery plantations and facilitate the immediate detection of BB diseases. This system holds significant potential for further enhancement in disease management across the tea industry, promoting healthier crops, reducing costs, and minimizing environmental impact.