Robust Estimation of Plucking Points Across Seasonal Flushes in Tea with UAV Imaging

Seasonal flushes of the tea crop represent different periods of the year during which leaves are picked for manu-facturing. A significant component of the quality and flavour of tea produced in India is associated with the flush type, where each flush has distinct characteristics. Using overhead images captured with Unmanned Aerial Vehicles (UAVs), we propose in this paper deep learning-based models for (a) robust plucking points in tea and (b) classifying them into flush types. Our approach was developed and tested with drone-based data gathered for a garden in the north-eastern tea-growing belt of the country over an entire season from late-February to mid-November. An mAP score of 0.91 was obtained for the detection of plucking points, and flush-type classification was achieved with an accuracy of 87%. Results showed that the UAVs integrated with tea operations can be quite effective for precision surveillance to help guide field decisions for better harvest and production efficiencies.