Automated olive grove classification and tree counting in very high resolution aerial imagery using deep learning

The economic significance of olive trees has fueled a growing interest in the development of advanced and efficient technologies specifically designed for olive cultivation. A key focus within this area has been the use of remote sensing technologies. By integrating remote sensing with Deep Learning we developed a robust algorithm for olive grove classification and tree counting. Our automated procedure uses Very High-Resolution RGB aerial imagery and includes a land classification step based on EfficientNet to distinguish olive groves from a generic land type, followed by an automatic tree counting process leveraging the YOLO algorithm. We evaluated our method on an open dataset of aerial VHR RGB images covering 443.8 km 2 in Southern Italy. For the classification task, the proposed model achieved a Cohen's κ of 0.982 and an overall accuracy of 99% at the parcel level. For tree detection and counting, YOLO reached a median F1 score of 91.5%, with a precision of 93.0%, recall of 90.5%, and a mean absolute error of 11.4 trees per hectare ( R 2 = 0.90 ). By leveraging state-of-the-art Deep Learning algorithms, our methodology marks a significant advancement in precision agriculture for olive orchards, offering notable improvements over current approaches. • EfficientNet-B1 Achieves Exceptional Performance in Olive Grove Classification. • YOLO Demonstrates Superior Tree Density Estimation and Precision Compared to Existing Methods.