Architectural design methods based on image recognition technology and virtual VR

To enhance the conceptual expression of architectural design and enhance the interactive experience, and to strengthen the application of computer technology in architectural design drawings, this study explores architectural design methods on the ground of image deep learning image recognition technology and augmented reality technology. The experiment demonstrated that when the accuracy of the improved You Only Look Once version 4 (YOLOv4) model was 0.9, the recall rate was 0.98, and the curve area was 0.93. The model loss function curve converged to the minimum value of 0.04, with the fastest convergence speed and the highest model recognition efficiency. Its time consumption was decreased by as much as 70.06%, indicating better overall performance. Meanwhile, the clustering strategy design of the model was relatively optimal, with the highest values of purity, standard mutual information, and Rand coefficient reaching 0.944, 0.931, and 0.942, respectively. In practical analysis of architectural design, the average accuracy and intersection over union of the improved YOLOv4 model confirmed the good detection performance of this method. The application of virtual reality technology in building information models has significantly improved the visualization delay rate, and the subjective evaluation of users was relatively high. The combination of visible image recognition and augmented reality can achieve intelligent processing and application of drawing information, improve design efficiency and quality, and optimize design experience.