Estimation of Plucking Points with Overhead Imaging in Tea - A Case Study

Tea is a perennial crop that grows well in elevated regions and is usually rainfed. It is a widely consumed beverage in India and the world over. The quality of made tea is significantly influenced by the quality of harvested leaves. Identifying and classifying young shoots, therefore, is critical to planning the harvest process and enabling timely plucking of quality leaves to help produce the best grades of made tea. Tea gardens in India are very labour-intensive when it comes to operations for pesticide application, manuring, and harvesting (aka plucking). Naturally, a fraction of the same workforce does the bulk surveillance as well, whether to spot pest flare-ups for control measures or assess a flush of young shoots ready for harvest. With labour costs on the rise and edge-IoT ecosystems in the form of drones, smart devices, and high-resolution cameras becoming more accessible and affordable, there is a big opportunity to do more (and precise) surveillance with fewer people. With a focus on estimating quality leaves with overhead imaging that can be scaled, we present in this paper our experiences and learnings from the 2021 tea season in North East India. Quality leaf-output comes from plucking two young shoots and a bud which is not trivial to identify in a tea bush where all healthy leaves are in shades of green. Moreover, overhead imaging only gives a selective (top-down) viewpoint of the bush where camera height could be adjusted to capture the identifiable features that could be replicated at scale as part of a larger process. We will discuss some of these aspects as part of our proposed approach. Drone-based imaging was extensively used in this study, and when combined with other modalities, we were able to detect the plucking points with an accuracy of 89.23%.