计算机技术与发展2025，Vol.35Issue(8)：53-60,8.DOI:10.20165/j.cnki.ISSN1673-629X.2025.0098
基于MSC-Net的地标性景点检测网络
Landmark-based Tourist Site Detection Network Based on MSC-Net
摘要
Abstract
Landmark recognition and retrieval based on natural attractions&man-made landmarks is currently receiving extensive attention from academia,while there is a research gap in open source datasets and models for landmark location detection in image data.In view of this,we firstly construct the first LLD dataset for landmark detection.Secondly,we propose the MSC-Net landmark detection method based on the multi-head spatial-channel hybrid attention mechanism.The method uses the improved ResNeXt as the backbone network,reduces the computational volume and the number of parameters by introducing the multi-head attention mechanism and one-di-mensional accelerated convolution,and combines the detection advantages of the feature pyramid network at different scales to improve the landmark detection performance of the MSC-Net model.Finally,PCA-AblationCAM interpretable method is proposed to improve computational complexity and provide understanding and interpretation of landmark detection.The proposed method is validated on the self-constructed LLD landmark detection dataset and the public Pascal VOC2012 dataset.Interpretability analysis and experimental results show that the proposed method has fewer parameters and outperforms multiple comparison networks on the open-source dataset as a whole,and in particular,MSC-Net achieves optimal performance on 9 out of 11 sub-datasets of LLD.关键词
地标检测/地标检测数据集/MSC-Net/多头空间通道混合注意力机制/PCA-AblationCAM/可解释性Key words
landmark detection/landmark detection dataset/MSC-Net/multi-head spatial-channel hybrid attention mechanism/PCA-AblationCAM/interpretability分类
信息技术与安全科学引用本文复制引用
姚欣睿,殷文斐,卢小芬,田霖轩,江梦琳,王学林,刘莹,郄良宇,卢碧瑶,刘建..基于MSC-Net的地标性景点检测网络[J].计算机技术与发展,2025,35(8):53-60,8.基金项目
国家自然科学基金(61906057) （61906057）
