Enhanced semantic segmentation in 3D interior visualization using CBAM-IAU-net and YODETR-V8

Abstract

The research presents an essential development of AI-based intelligent design tools that enable interactive visualization of interiors. The research solves major issues in current 3D interior visualization approaches because existing methods produce less accurate object recognition, reduce feature selection efficiency, and cannot adapt to multiple spatial settings. The current methodologies fail to achieve their full effectiveness because of these challenges when applied to smart architecture, automated design planning, and virtual staging. To overcome such limitations, in this research study, Internet of Things (IoT) is integrated with Software-Defined Networking (SDN) as the underlying infrastructure for adaptive and real-time interior visualization (SDN-IoT). Along with this, three new AI-based approaches are proposed that are Semantic Segmentation using CBAM-Infused Attention U-Net (CBAM-IAU-Net), utilizing Convolutional Block Attention Module (CBAM) and Attention U-Net and U-Net for precise segmentation of interior objects, Feature Selection by the Hippocoati Optimization Algorithm (HCOA)—hybrid optimization method by combining Hippopotamus Optimization Algorithm (HO) and Coati Optimization Algorithm (COA) to optimize feature selection, and Deep Learning (DL)-Based Visualization with YOLO-DEtection TRansformer Version 8 (YODETR-V8)—a hybrid detection algorithm combining YOLOv8 (You Only Look Once - Version 8) and DETR (DEtection TRansformer) for spatial reasoning, robust object detection, and real-time 3D visualization. Experimental findings validate the effectiveness of the proposed 3DMeta-Interior method with superior performance with an accuracy of 98.98%, precision of 98.27%, sensitivity of 98.07%, and specificity of 98.68%. It achieves an F1-score of 98.44%, MCC of 98.78%, NPV of 98.58%, with a low false positive rate (1.48%) and false negative rate (0.9%), outperforming all compared models. The findings present the proposed method as an extremely effective and scalable method for AI-based 3D interior visualization, paving the way for enhanced architectural design, home automation, and virtual design applications of the future.