CountShoots: Automatic Detection and Counting of Slash Pine New Shoots Using UAV Imagery

Area of Science:

• Forestry Science
• Computer Vision
• Artificial Intelligence

Background:

• New shoot density in pine trees is crucial for assessing growth and photosynthetic capacity.
• Traditional manual and destructive methods for shoot density measurement are labor-intensive and limit research scope.

Purpose of the Study:

• To develop user-friendly software (CountShoots) for easy and convenient extraction of new shoot density using unmanned aerial vehicles (UAVs).
• To introduce novel AI models for accurate tree species identification and automated new shoot counting in slash pine.

Main Methods:

• A modified YOLOX model was employed for tree species and location identification from RGB images, achieving high recognition accuracy.
• A Slash Pine Shoot Counting Network (SPSC-net), based on the CCTrans network, was developed for automated new shoot counting.
• The performance of the developed models was compared against existing state-of-the-art models like YOLOv5, Efficientnet, Faster-RCNN, DM-Count, CSR-net, and MCNN.

Main Results:

• The modified YOLOX model achieved high recognition accuracies of 99.15% and 95.47%, outperforming other object detection models.
• The SPSC-net demonstrated superior performance with the lowest mean squared error (2.18) and mean absolute error (1.47) compared to other counting models.
• This research represents the first automated system for identifying tree crowns and counting new shoots in slash pine.

Conclusions:

• The CountShoots software provides a highly efficient and rapid user-interactive system for pine tree new shoot detection and counting.
• The developed AI models offer a non-destructive and convenient alternative to traditional methods, supporting tree breeding and genetic research.