Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Deep Learning-based Segmentation of Thermal Images for Heat Loss Detection in Building Façade
This study addresses the heat loss problem in buildings by developing the heat loss detection network (HLDNet). The HLDNet is a novel machine learning model incorporating convolutional neural networks (CNN), attention modules, and squeeze and excitation (SE) blocks. A dataset comprising 200 fused thermal and RGB images was developed using data collected from a multiunit residential building in an extremely cold environment. The model showed a remarkable result in detecting regions of heat loss and windows within the input images, an essential feature for distinguishing between noise and façade walls. During validation, the mean intersection over the union (mIoU) metric increased from 0.801 to 0.843 with the help of data augmentation. A real-time performance of 120 fps was obtained from the developed HLDNet.
Deep Learning-based Segmentation of Thermal Images for Heat Loss Detection in Building Façade
This study addresses the heat loss problem in buildings by developing the heat loss detection network (HLDNet). The HLDNet is a novel machine learning model incorporating convolutional neural networks (CNN), attention modules, and squeeze and excitation (SE) blocks. A dataset comprising 200 fused thermal and RGB images was developed using data collected from a multiunit residential building in an extremely cold environment. The model showed a remarkable result in detecting regions of heat loss and windows within the input images, an essential feature for distinguishing between noise and façade walls. During validation, the mean intersection over the union (mIoU) metric increased from 0.801 to 0.843 with the help of data augmentation. A real-time performance of 120 fps was obtained from the developed HLDNet.
Deep Learning-based Segmentation of Thermal Images for Heat Loss Detection in Building Façade
Waqas, Ali (Autor:in) / Araji, Mohamad T. (Autor:in)
Technology|Architecture + Design ; 8 ; 370-379
02.07.2024
10 pages
Aufsatz (Zeitschrift)
Elektronische Ressource
Englisch
Building facade evaluation using instance segmentation on thermal images
DataCite | 2024
|