TEADIS: Deep learning based tea leaf disease classification and segmentation via field observation data

Abstract

Globally, tea leaf diseases significantly impact economic growth, production, and crop quality. This research proposes a novel Deep learning-based TEADIS framework to identify the tea leaf diseases. Initially, visual and digital data are gathered from the Internet of Things (IoT) devices and stored through Sigfox in the cloud environment. The visual data are pre-processed utilizing a scalable range-based adaptive bilateral filter (SCRAB) to remove noise from tea leaf images. The GoogLeNet with GeLu activation function is employed to classify the tea leaves into normal and abnormal leaves. The digital data from the environmental field are used to detect the occurrence of tea diseases for segmentation. Attention V-Net is used to improve segmentation accuracy with attention mechanisms to focus on relevant regions for enhancing the precision of identifying the affected areas. The proposed TEADIS model attained an accuracy of 97.77% based on the gathered the data from the year 2021-2023.