基于卷积神经网络的建筑表面裂纹识别技术

Detection Technology of Building Surface Cracks Based on Convolution Neural Network

抹灰工程的施工质量直接影响用户体验,其中建筑墙体表面裂纹是工程中常见的质量问题。为实现建筑表面裂纹的无人化识别及检验,采集了一系列建筑表面裂纹图像,并使用图像分割和数据增强方法对图像进行预处理,基于此,采用卷积神经网络搭建了建筑表面裂纹识别模型,同时,提出一项双尺度裂纹识别技术以提高裂纹识别的精准度,并将该识别模型内嵌至配置3D相机的行走式机器人,最终形成基于卷积神经网络的建筑表面裂纹识别技术。经实际应用检验,该技术具有较高的识别准确度,对肉眼难以发现的裂纹也有较好的识别效果。

The construction quality of plastering engineering directly affects the user experience,among which the surface crack of building wall is a common quality problem.To realize unmanned detection and inspection of building surface cracks,this paper adopted image segmentation and data enhancement methods to preprocess the crack images.After that,a building surface crack detection model was built by using convolution neural network.Thereafter,a dual scale crack identification technology was proposed to improve the accuracy of crack detection.Besides,the model was embedded into a walking robot with a 3D camera.The practical application shows that the proposed crack identification technology has high recognition accuracy and good recognition effect for cracks that are difficult to be found by the eye.