摘要
在高层建筑平面审查过程中,主要通过基于ResNET(残差学习网络)的图片分类模型进行平面凹凸不规则识别,存在数据样本不均衡的问题,使得最终识别结果曲线下面积(AUC)值较低。针对此类问题,提出建筑信息模型(BIM)下的高层建筑平面凹凸不规则识别方法。利用布尔交运算方法,分析高层BIM模型中所有结合对象之间的关系,获取建筑平面信息。通过无人机搭载图像采集系统获取建筑实景图像,并应用多尺度局部直方图均衡化算法实现图像增强。建立以最大类间方差法为基础的图像分割方案,结合多策略融合未来搜索算法确定最优分割阈值,得到高层建筑图像目标前景区域。最后,运用深度学习网络构建基于潜在特征空间的图片异常检测模型,将分割后图像输入训练好的模型中自动学习,即可得到建筑平面凹凸不规则识别结果。实验结果表明:该方法识别结果的AUC值为0.9,更符合高层建筑平面审查工程的要求。
In the process of high-rise building inspection,the image classification model based on the ResNET(residual learning network)is mainly used to identify the irregular concave and convex plane,but it faces the problem of unbalanced data samples,resulting in lower AUC values in the final outcome.In order to solve these problems,a method for identifying the irregular concave and convex plane of high-rise building based on building information modeling(BIM)was proposed.The relationship between all combined objects in the BIM model of high-rise building was analyzed by using Boolean intersection algorithm,and the building plane information was obtained.The real image of the building was obtained by the image acquisition system of the UAV,and the multi-scale local histogram equalization algorithm was applied to realize image enhancement.An image segmentation scheme based on maximum inter-class variance method was established,and the optimal segmentation threshold was determined by combining multi-strategy and future search algorithm to obtain the target foreground area of high-rise building image.Finally,a deep learning network was used to construct an image anomaly detection model based on latent feature space.The segmented image was input into the trained model for automatic learning,and the recognition results of irregular concave and convex building plane can be obtained.The experimental results show that the AUC value of the recognition result of this method is 0.9,which is more in line with the needs of high-rise building plane inspection engineering.
作者
郑平芳
袁玲
ZHENG Pingfang;YUAN Ling(Department of Transportation Engineering,Fujian Forestry Vocational Technical College,Nanping 353099,China)
出处
《成都工业学院学报》
2024年第5期6-10,共5页
Journal of Chengdu Technological University
基金
福建省中青年教师教育科研项目(JAT210764)。
关键词
建筑信息模型
高层建筑
潜在特征空间
图像增强
图像分割
不规则识别
Building Information Model(BIM)
High-rise Building
latent feature space
Image Enhancement
Image Segmentation
irregular recognition
