Improved YOLOv11 lightweight-based trunk detection method and system

CN120808200APending Publication Date: 2025-10-17JISHOU UNIVERSITY
View PDF 0 Cites -1 Cited by

Patent Information

Application Number
CN202510738746.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-06-04
Publication Date
2025-10-17

AI Technical Summary

Technical Problem

Existing tree trunk detection models have problems in complex forest environments, such as poor detection robustness, imbalance between accuracy and speed, and hardware deployment bottlenecks, making it difficult to meet the real-time inspection needs of drones.

Method used

By adopting the improved YOLOv11 lightweight method, building a self-built dataset, dynamically fusing multi-frame point cloud and image information, and combining the StarNet_Trunk lightweight network, C2DA module and EffiDet detection head, the allocation of computing resources is optimized, and the segmentation accuracy and system robustness are improved.

Benefits of technology

Significantly reduce the number of model parameters and calculations, improve detection speed, achieve centimeter-level real-time obstacle avoidance, support the safe cruising of drones in complex woodlands, and reduce hardware costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120808200A_ABST
    Figure CN120808200A_ABST
Patent Text Reader

Abstract

The invention discloses a trunk detection method and system based on improved YOLOv11 lightweight, and belongs to the technical field of unmanned aerial vehicle target detection. The method comprises the following steps: constructing a real acquisition and synthesis enhancement data set (1: 1 proportion), and simulating a complex scene through background segmentation and random scaling superposition; a YOLOv11-TrunkLight lightweight model is established, a backbone network of the YOLOv11-TrunkLight lightweight model adopts a StarNetTrunk lightweight module, an SPPF multi-scale feature aggregation module and a C2DA module containing a Dattention mechanism, and a head of the YOLOv11-TrunkLight lightweight model adopts an EffiDet detection head of a dynamic anchor frame; training parameters are set to train the model, and evaluation is carried out through indexes such as precision, recall rate and mAP at 0.5. According to the method, the parameter quantity is reduced by 34.6%, the calculated quantity is reduced by 39.7%, the FPS is improved to 275.37, and real-time accurate obstacle avoidance of the unmanned aerial vehicle in a dense forest environment is realized.
Need to check novelty before this filing date? Find Prior Art