融合YOLOv10与立体视觉的小目标三维检测OA

In civil infrastructure monitoring,seepage is a major hazard to dam stability.Small objects are notoriously difficult to detect due to their limited scale,blurred boundaries,and complex backgrounds,leading to suboptimal localization accuracy.In view of this,the paper proposes a three-dimensional detection method that integrates a deep learning-based object detection model with stereo vision techniques to improve detection performance and spatial positioning accuracy.On the basis of the original detection network,a cross-scale edge enhancement module is introduced to strengthen boundary feature representation,and a lightweight attention mechanism is incorporated to enhance feature extraction efficiency.Leveraging binocular imagery for disparity estimation,this method facilitates the recovery of spatial coordinates for small objects.Validation was conducted based on actual engineering monitoring data.The results demonstrate that the proposed method achieves competitive performance in terms of average precision(AP),spatial localization error,and inference time,offering good real-time performance and adaptability.The proposed method is applicable to complex engineering scenarios such as dam seepage monitoring.Furthermore,it can be readily extended to related tasks in industrial inspection and automated sensing.