基于深度学习的中东铁路沿线历史建筑颜色特征提取方法  

Method for Extracting Color Features of Historic Buildings Along the Chinese Eastern Railroad Based on Deep Learning

作　　者：刘雯雯 叶芷姗 毛靓[1] 

机构地区：[1]东北林业大学园林学院

出　　处：《建筑与文化》2025年第2期267-269,共3页Architecture & Culture

基　　金：黑龙江省哲学社会科学研究规划项目(项目编号:23YSB221);黑龙江省经济社会发展重点研究课题(课题编号:24564)。

摘　　要：建筑色彩特征的提取对于建筑遗产的保护至关重要,因为它提供了科学的决策支持。文章采用YOLOv8s-seg实例分割模型,加入了RFAHead检测头和EMAttention注意力机制,显著提升了模型的表现性能。实验数据显示,经过改进的YOLOv8s-seg模型在建筑实例分割的平均精度上达到了95.1%,相比原YOLOv8s-seg模型提高了2.2个百分点。这一提升,不仅验证了模型在建筑个体检测精度上的有效性,还更好地增强了建筑实例的分割效果。此外,文章进一步改进了K-means聚类算法,通过轮廓系数确定聚类数目,并通过遗传算法确定初始聚类中心,减少了聚类的随机性。对分析结果进行精细的聚类处理,旨在高效提取建筑物的色彩特征。这些技术的综合应用,为建筑色彩特征的准确提取和有效利用提供了新的方法。The extraction of architectural color characteristics is crucial for the protection of architectural heritage,as it provides scientific decisionmaking support.The paper employs the YOLOv8s-seg instance segmentation model,incorporating the RFAHead detection head and EMAttention attention mechanism,which significantly improves the performance of the model.Experimental data show that the improved YOLOv8s-seg model achieves an average precision of 95.1%in architectural instance segmentation,an increase of 2.2 percentage points compared to the original YOLOv8s-seg model.This enhancement not only verifies the model's effectiveness in the precision of architectural individual detection but also better improves the segmentation effect of architectural instances.Furthermore,the article refines the K-means clustering algorithm by determining the number of clusters through silhouette coefficients and the initial cluster centers through genetic algorithms,reducing the randomness of clustering.Fine clustering processing of the analysis results aims to efficiently extract the color characteristics of buildings.The comprehensive application of these technologies provides a new method for the accurate extraction and effective utilization of architectural color characteristics.

关 键 词：Yolov8s-em-rfa-seg Kmeans-S聚类 实例分割 深度学习 中东铁路建筑颜色 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]