UAV imagery, advanced deep learning, and YOLOv7 object detection model in enhancing citrus yield estimation

Abstract

Accurate citrus fruit yield and estimation is of utmost importance for precise agricultural management. Unmanned aerial vehicle (UAV) remote-sensing systems present a compelling solution to this problem. These systems capture remote-sensing imagery with both high temporal and spatial resolution, thus empowering farmers with valuable insights for better decisionmaking. This research assessed the potential application of UAV imagery combined with the YOLOv7 object detection model for the precise estimation of citrus yield. Images of citrus trees were captured in their natural field setting using a quadcopter-mounted UAV camera. Data augmentation techniques were applied to enhance the dataset diversity; the original YOLOv7 architecture and training parameters were modified to improve the model’s accuracy in detecting citrus fruits. The test results demonstrated commendable performance, with a precision of 96%, a recall of 100%, and an F1-score of 97.95%. The correlation between the fruit numbers recognized by the algorithm and the actual fruit numbers from 20 sample trees provided the coefficient R2 of 0.98. The strong positive correlation confirmed both the accuracy of the algorithm and the validity of the approach in identifying and quantifying citrus fruits on sample trees.