Objective Due to the large size variation and frequent occlusion of tree trunks in complex environments, tree trunk detection is susceptible to issues such as missed detections and false positives. To effectively address these problems, a tree trunk detection network named SSFYOLO, based on a single-stage target detection framework, was proposed.

Method Firstly, a spatial awareness module (SAM) was designed. The SAM module can efficiently process multi-scale and multi-resolution feature information, ensuring computational efficiency while fully integrating and accurately extracting various features, thereby improving the accuracy and efficiency of target detection. Secondly, a multi-scale feature enhancement adaptive network (FastScaleNet) was designed to replace the C2f structure in the YOLO model. By refining multi-scale feature fusion and optimization, and utilizing skip connections and feature fusion strategies, FastScaleNet effectively retained feature information at different levels, enhanced the model’s ability to express targets of different scales, and improved the model’s robustness and versatility. Finally, the WIoU mechanism was introduced to dynamically optimize the loss weights for small targets. Depending on the target size, the WIoU mechanism dynamically adjusts the loss weights, allowing the model to flexibly adjust parameters when facing small targets. This enabled the model to adapt to the detection requirements of targets of different sizes, further improving the accuracy and robustness of small object detection.

Result Detection experiments were conducted on a tree trunk dataset in complex scenarios. Compared with the mainstream detection algorithm YOLOv8, the proposed SSFYOLO algorithm achieved better detection accuracy with reduced parameter volume. Specifically, the parameter volume was reduced by 20%, mAP and the recall rate represented improvements of 1.6 and 0.7 percentage points, respectively.

Conclusion We successfully designed the SSFYOLO algorithm for detecting tree trunks in complex forest scenarios. The proposed algorithm demonstrates excellent performance in tree trunk detection in complex environments, and has broad application prospects.