计算机与数字工程2025，Vol.53Issue(1)：108-114,7.DOI:10.3969/j.issn.1672-9722.2025.01.021
基于自适应感受野的多尺度显著性目标检测
Multi-scale Saliency Object Detection Based on Adaptive Receptive Field
缑新科 1李飞飞1
作者信息
- 1. 兰州理工大学电气工程与信息工程学院 兰州 730050
- 折叠
摘要
Abstract
The saliency detection technology based on CNN has gradually replaced the traditional saliency detection technolo-gy.Most of the existing saliency detection techniques do not distinguish the features of different layers,and process the features of different layers in the same way.In order to extract rich multi-scale information,the backbone network VGG-16 is divided into three layers:low,medium and high,and different attention mechanisms are used for the features of different layers.In order to ex-tract more information,SK rolls are used according to different input information.The product is used to adaptively increase the re-ceptive field.Finally,the obtained features are optimized and fused.The model is tested on 5 commonly used datasets and compared with other saliency detection algorithms.Through qualitative and quantitative evaluation,the model has good performance.关键词
显著性检测/自适应感受野/注意力机制/多尺度特征Key words
saliency detection/adaptive receptive field/attention mechanism/multi-scale features分类
信息技术与安全科学引用本文复制引用
缑新科,李飞飞..基于自适应感受野的多尺度显著性目标检测[J].计算机与数字工程,2025,53(1):108-114,7.
