Hsin-Cheng ChenShih-Fang ChenShiou-Ruei LinTa-Te Lin2024-10-282024-10-282024https://www.scopus.com/record/display.uri?eid=2-s2.0-85206100316&origin=resultslisthttps://scholars.lib.ntu.edu.tw/handle/123456789/722507Monitoring the growth status of tea trees in a tea garden requires significant time and manpower to determine the temporal trend of tea shoot growth. In this study, we propose an approach to use tea canopy images acquired from field to calculate the Tea Shoot Density Index (TSDI) of the tea tree. The TSDI can be used to estimate the yield of tea leaves in tea gardens and assist in various management operations. By integrating deep learning models of YOLO V8 instance segmentation with DINO V2 depth estimation, tea shoot object recognition accuracy is enhanced, enabling the precise calculation of the TSDI. This leads to developing a tea shoot growth model that provides a quantified pattern for evaluating tea tree growth curve assessment. The proposed method eliminates the necessity of a complex 3D imaging system by leveraging just a standard monocular RGB camera for tea shoot detection. Through the integration of an embedding system designed for continuous image acquisition across the tea canopy, our monitoring framework facilitates comprehensive evaluation and analysis of field tea tree growth. This approach offers insights into both temporal and spatial dimensions, empowering tea garden managers to optimize cultivation practices and elevate the quality of tea products.falseDeep LearningGrowth Curve ModelingTea Shoot Densityconference paper10.13031/aim.2024003312-s2.0-85206100316