Digital drawings, which provide a detailed graphical representation, play a crucial part by featuring instant feedback on a vast amount of information among users. Artificial intelligence (AI) adopts a routine to extract features from the digital drawings and classify these drawings and as a result, a 3D model is constructed. In this study, two deep neural networks are proposed for the automatic classification of construction drawings and object detection using alternative strategies for feature extraction. One deep neural network is conducted to classify seven types of construction drawings. The neural network is trained using two CNN architectures (VGG16, VGG19). The second neural network aims to localize the title boxes of the construction drawings. The object detection network is trained on two pre-trained models (YOLOv5m, YOLOv5s) to achieve the highest possible performance of the network. The different cases were trained, fine-tuned, and evaluated. The results showed image classification network with the VGG16 architecture achieved an accuracy of 94% and the object detection network, which used the parameters of the yolov5m model achieved an accuracy of 100% in localizing the bounding boxes and an accuracy of 93% in localizing the bounding boxes with classifying the type of the construction drawing.     «

Digital drawings, which provide a detailed graphical representation, play a crucial part by featuring instant feedback on a vast amount of information among users. Artificial intelligence (AI) adopts a routine to extract features from the digital drawings and classify these drawings and as a result, a 3D model is constructed. In this study, two deep neural networks are proposed for the automatic classification of construction drawings and object detection using alternative strategies for feature...     »