Enabling Consumer UAVs for Precision Agriculture Applications: A Case Study of Yield Estimation

Unmanned aerial vehicles (UAVs) equipped with various sensors and onboard processing capabilities have emerged as a promising means to acquire field data for precision agriculture applications. However, such UAVs are costly, restricting their deployment in small-To-medium-sized fields, particularly in developing countries. In contrast, consumer-grade UAVs have high-resolution RGB cameras and video streaming abilities at affordable prices. This paper presents an efficient processing pipeline to analyze video streams from consumer-grade UAVs on smartphones. The processing pipeline consists of preprocessing, object detection, and yield estimation. The object detector, being the most computationally expensive module, is invoked every nth frame due to video redundancy and the target platform's limited resources. The yield estimation task on a smartphone requires efficient and accurate fruit detection, which a modified YOLOv8n model achieved. We evaluate our pipeline on datasets of apple and peach trees and demonstrate that it can process UAV-captured images to collect yield-related statistics. We also discuss the lessons learned and outline future directions for consumer-grade UAV-based precision agriculture applications.