基于机器视觉深度学习的电站渗水检测识别技术研究

Research on power plant water penetration detection and identification technology based on machine vision deep learning

针对传统的人工检测水库库底排水巡检区域渗水、裂缝的方法存在对工作人员要求过高,检测结果误差过大,检测区域限制过大等缺点,文中提出了一种基于Canny算法与卷积神经网络的裂缝检测识别技术。首先,利用轮式机器人对水电站中可能存在的裂缝进行图像采集,接着借助Canny算法对图像进行预处理并制作成对比数据库,通过数据库训练出能够识别含有裂缝图像的卷积神经网络。最终,将卷积神经网络迁移至机器人的微主板中,使得机器人在巡检过程中可以对渗水、裂缝等异常现象及时报警。实验结果表明,基于本方案的裂缝图像识别率达98.33%,在实际巡检工作中能够发现绝大多数的渗漏危险并给予报警。

In view of the shortcomings of the traditional method of artificial detection of water seepage and cracks in the inspection area of reservoir bottom drainage,such as too high requirements for the staff,too large error in the detection results,and too large limit in the detection area,this paper proposes a crack detection and recognition technology based on Canny algorithm and convolution neural network.Firstly,the wheel robot is used to image mining the possible cracks in the hydropower station Then,the image is preprocessed by Canny algorithm,and a contrast database is made.The convolution neural network which can recognize the image with cracks is trained by the database.Finally,the convolutional neural network is transferred to the micro main board of the robot,which makes the robot alarm the abnormal phenomena such as water seepage and cracks in the process of inspection.The experimental results show that the crack image recognition rate based on this scheme is up to 98.33%.In the actual inspection work,most of the leakage risks can be found and alarm can be given.