计算机科学
背景(考古学)
分割
特征提取
特征(语言学)
残余物
数据挖掘
人工智能
地形地貌
遥感
模式识别(心理学)
地图学
地理
语言学
哲学
考古
算法
作者
Xiaoling Ye,Haohua Zhang,Tao Sheng,Yingchao Zhang,Ruilin Zou,Xiong Xiong
标识
DOI:10.1080/01431161.2023.2295835
摘要
Foreign object debris (FOD) intrusion is a common problem along high-speed railway lines, usually caused by the entry of foreign objects into the railway infrastructure during high winds. Although the capability to address FOD incidents has improved significantly, it is still necessary to predict environmental conditions along the railway lines beforehand to prevent FOD incidents. In this paper, we propose a novel dual-branch feature extraction network that takes into account the frequently occurring landforms of lakes and forests along the railway lines to improve the recognition rate of these regions and provide theoretical support for FOD prevention. The network adopts a dual-branch structure, introducing multi-branch residual weighted module, context feature refinement module (CFR), and high-low feature fusion module (HLFF) to enhance feature extraction, refine contextual information, and boundary information, thus improving the model's segmentation performance. Experiments are conducted on our self-built forest-lake dataset, a railway dataset, and a publicly available Aerial Imagery Dataset. The results show that the proposed method achieves MIOU scores of 84.77%, 86.73% and 94.09% on these three datasets, respectively, indicating strong segmentation performance for forests and lakes and good generalization performance on other datasets.
科研通智能强力驱动
Strongly Powered by AbleSci AI