基于运动识别的边坡工程坍塌检测研究

The Research on Slope Collapse Detection Utilizing Motion Recognition Technology

针对河防工程边坡坍塌监测工作中存在监测效率低、人身安全风险大、人力成本高等问题,提出了一种基于深度学习技术的河防工程边坡坍塌智能检测方法,该方法首先利用基于注意力机制的轻量化U-net图像分割模型对河防工程边坡区域进行精准分割,然后再利用光流法对边坡区域中产生的运动进行识别,最后设计一个轻量化卷积神经网络对识别到的运动目标进行分类,判断运动目标是否为边坡本身。通过在模拟河防工程坍塌数据以及实际现场数据上的实验结果表明所提出的算法可以有效地对工程边坡坍塌事件进行识别,具有较强的应用价值。

In response to the issues of low monitoring efficiency,significant personal safety risks and high labor costs present in the monitoring work for river defense project slope collapses,an intelligent detection method for these collapses based on deep learning technology has been proposed.The proposed approach initially employs the attention-based lightweight U-net image segmentation model to precisely delineate the slope areas of river defense projects.Subsequently,the optical flow method is employed to recognize any movement occurring within the slope areas.Finally,a lightweight convolutional neural network is devised to classify the detected motion targets and determine whether the moving target is the slope itself.The experimental results obtained from both simulated data of river defense project collapses and actual on-site data demonstrate the effective capability of the proposed algorithm in identifying events of engineering slope collapse,thus highlighting its significant practical value.