Remote sensing for precision forestry: detection and counting of Eucalyptus saligna trees using multispectral RPAS imaging and the YOLOv8n computer vision algorithm

Abstract

The integration of intelligent systems for processing forest data is advancing, offering a promising approach to enhance the efficiency and sustainability of the forestry sector. However, there is currently a lack of rapid and accurate methods for tree counting during the initial phase of Eucalyptus plantations. Therefore, this study aimed to evaluate the use of the You Only Look Once version 8 nano (YOLOv8n) algorithm to detect and estimate the count of Eucalyptus trees 180 days post-planting, using high resolution multispectral imagery. The study area consisted of a 7.09 ha commercial Eucalyptus saligna plantation located in southern Brazil. The identification and delimitation of E. saligna tree crowns for training and validation of the model were performed using multispectral images obtained by a Remotely Piloted Aircraft System (RPAS). For detection and count estimation, the YOLOv8n algorithm was used. For model evaluation, 70% of the dataset was allocated for training and 30% for validation, achieving a recall of 0.93 and an mAP50 of 0.92. Subsequently, the model was employed to predict and estimate the tree count in an independent test area (4.45 ha). The final validation metrics yielded a recall of 0.915, mAP50 of 0.722, true positive rate of 78.15%, and false positive rate of 13.61%, with a relative error of 4.78%. The YOLOv8n model proved to be promising in the detection and count of E. saligna trees during the initial phase of the plantation, offering potential support for assessing survival rates, as well as for monitoring growth and estimating wood volume.