Research on UAV aerial imagery detection algorithm for Mining-Induced surface cracks based on improved YOLOv10

Area of Science:

• Geospatial analysis
• Artificial intelligence
• Geological engineering

Background:

• UAV-based aerial imagery is crucial for detecting mining-induced surface cracks, aiding geological disaster early warning and ensuring safe production.
• Challenges include small crack size, complex morphology, scale variation, and uneven distribution, compounded by limited UAV computational power.

Purpose of the Study:

• To develop an efficient and lightweight small-target detection model (YOLO-LSN) for UAV-based crack detection in mining areas.
• To optimize crack feature extraction and enhance the detection of small cracks in complex environments.

Main Methods:

• Introduced a Lightweight Dynamic Alignment Detection Head (LDADH) for multi-scale feature fusion and dynamic receptive field adjustment.
• Developed a Small Object Feature Enhancement Pyramid (SOFEP) to improve small crack detail representation.
• Proposed a weighted loss function combining Normalized Wasserstein Distance (NWD) and IoU for improved localization accuracy.

Main Results:

• Achieved a 12% mAP@0.5 improvement and a 17% parameter reduction on a custom mining crack dataset.
• Demonstrated effectiveness on the VisDrone2019 dataset with a mAP@0.5 of 0.422 (+11.6%).
• Validated the model's efficiency and reliability for UAV-based small-object detection.

Conclusions:

• YOLO-LSN provides an effective and computationally efficient solution for detecting small surface cracks using UAV imagery.
• The model enhances geological hazard warning systems and contributes to safer mining operations.
• The proposed methods address key challenges in UAV-based small-object detection for environmental monitoring.