Architectural Heritage Style Identification Using Avian Swarm Optimized K-Nearest Neighbours and Deep Learning

Abstract

The architectural heritage style comprises unique design elements and building techniques that are indicative of the historical, social, and cultural milieu of certain eras or locales. Chinese architectural heritage styles are a result of growing interest in employing Deep Learning (DL) algorithms and digital data for these kinds of analysis and identification. By preprocessing the data with grayscale image approaches, it becomes easier to extract features using Histogram of Oriented Gradient (HOG) descriptors to capture textural and structural attributes. Following that, the architectural styles are grouped using a new identification model that combines Avian Swarm Optimized K-Nearest Neighbors (ASO-KNN), an improvement over the standard KNN method that incorporates ASO behavior. The efficiency and accuracy of data categorization are increased by this hybrid approach, which maximizes KNN selections. The goal of the study is to properly categorize and describe traditional styles of architecture depending on their distinctive attributes and qualities. The proposed ASO-KNN method outperforms then the existing models AlexNet, DenseNet, and ResNet with the parameters of (89.25%) accuracy, (93.36%) F1-score, (91.59%) recall, and (98.79%) precision, the suggested methodology better than the state-of-the-art methods. These findings support the method's effectiveness and set the stage for further developments in automated architectural style recognition.