基于深度学习的居住建筑外墙面损伤检测研究

IMAGE DETECTION OF EXTERIOR WALL DAMAGE OF RESIDENTIAL BUILDING BASED ON DEEP LEARNING

以Tensorflow 2.0为平台,通过Faster RCNN算法框架建立深度学习模型。以1620张居住建筑外墙面受损照片为数据集。选取其中1296张为训练集,对模型进行有监督训练并测试模型训练深度,324张为测试集校检模型精度。测试结果表明,深度学习模型对居住建筑外墙的污染类损伤检测率为88.82%;裂缝类损伤检测率为90.21%;破损类损伤检测率为90.94%,检测平均耗时为每图0.23s。深度学习检测模型可有效反馈外墙面的主要损伤情况,提高建筑工程管WC率。

Tensorflow 2.0 is used as the framework to establish the deep learning model through faster RCNN algorithm.1620 damaged photos of the exterior wall of residential buildings were used as a data set.Among them,1296 photos are selected as training set for conducting supervised training on the model and testing the training depth of the model,and 324 photos are selected as the testing set for examining the detection precision of the model.Results show that the detection accuracy of the model of detecting dirt,crack and damage photos of exterior walls of residential buildings is about 88.82%,90.21%and 90.94%respectively.The average detection time is about 0.23 s per graph.The deep learning detection model can effectively feedback main damage and improve efficiency of construction management.