没有合适的资源?快使用搜索试试~ 我知道了~
温馨提示
内容概要:本文提出了一种新的深度学习架构——HCF-Net(Hierarchical Context Fusion Network),旨在提升红外图像中小目标物体的检测精度和鲁棒性。HCF-Net采用了多尺度特征提取、平行补丁注意力模块(PPA)、维度感知选择性集成模块(DASI)以及多扩张通道细化器(MDCR),有效解决了由于红外图像低对比度及背景复杂所导致的小目标检测困难的问题。该研究对SIRST数据集进行了广泛实验评估,表明HCF-Net性能优于现有主流传统和深网方法。HCF-Net通过优化下采样过程中的特征表示与细节捕捉,大幅提高了对微小目标位置识别及形状边界描写的准确性。此外,研究团队还在论文中探讨了相关领域的最新进展和其他基于卷积神经网络的技术。 适合人群:对于计算机视觉尤其是遥感成像与自动目标识别有浓厚兴趣的研究人员和技术爱好者。同时适用于从事国家安全、军事侦察、灾害监测等领域工作的专业人士。 使用场景及目标:应用于各种需要精确探测小型目标物的应用场合,比如海上搜索救援行动、消防监控、边境安防巡逻、天文观测系统等。其目的是提高这些应用场景中设备的工作效率和服务质量。 其他说明:文中还介绍了大量关于红外线成像特性的基础知识,并详细阐述了几种传统的滤波器和机器学习算法用于解决这一任务时面临的局限性;强调了深度学习相对于传统方法所具有的明显优势。
资源推荐
资源详情
资源评论
HCF-Net: Hierarchical Context Fusion Network for
Infrared Small Object Detection
Shibiao Xu
1
, ShuChen Zheng
1
, Wenhao Xu
1
, Rongtao Xu
3,4
, Changwei Wang
2,3,5, ∗
,
Jiguang Zhang
3,4
, Xiaoqiang Teng
1
Ao Li
6, ∗
, Li Guo
2
1
Artificial Intelligence, Beijing University of Posts and Telecommunications
2
Key Laboratory of Computing Power Network and Information Security, Ministry of Education,
Shandong Computer Science Center, Qilu University of Technology (Shandong Academy of Sciences)
3
State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences
4
School of Artificial Intelligence, University of Chinese Academy of Sciences
5
Shandong Provincial Key Laboratory of Computer Networks, Shandong Fundamental Research Center for Computer Science
6
School of Computer Science and Technology, Harbin University of Science and Technology
Abstract—Infrared small object detection is an important
computer vision task involving the recognition and localization of
tiny objects in infrared images, which usually contain only a few
pixels. However, it encounters difficulties due to the diminutive
size of the objects and the generally complex backgrounds in
infrared images. In this paper, we propose a deep learning
method, HCF-Net, that significantly improves infrared small
object detection performance through multiple practical modules.
Specifically, it includes the parallelized patch-aware attention
(PPA) module, dimension-aware selective integration (DASI)
module, and multi-dilated channel refiner (MDCR) module. The
PPA module uses a multi-branch feature extraction strategy to
capture feature information at different scales and levels. The
DASI module enables adaptive channel selection and fusion. The
MDCR module captures spatial features of different receptive
field ranges through multiple depth-separable convolutional lay-
ers. Extensive experimental results on the SIRST infrared single-
frame image dataset show that the proposed HCF-Net performs
well, surpassing other traditional and deep learning models. Code
is available at https://github.com/zhengshuchen/HCFNet.
Index Terms—Infrared small object detection, Deep learning,
Multi-scale features.
I. INTRODUCTION
Infrared small object detection is a crucial technology for
identifying and detecting minute objects in infrared images.
Due to the ability of infrared sensors to capture the infrared
radiation emitted by objects, this technology enables precise
detection and identification of small objects, even in dark
or low-light environments. As a result, it holds significant
application prospects and value in various fields, including
military, security, maritime rescue, and fire monitoring.
However, Infrared small object detection is still challenging
for the following reasons. First, deep learning currently serves
This work is supported by Beijing Natural Science Foundation No.
JQ23014, in part by the National Natural Science Foundation of China (Nos.
62271074, 62071157, 62302052, 62171321 and 62162044), and in part by the
Open Project Program of State Key Laboratory of Virtual Reality Technology
and Systems, Beihang University (No. VRLAB2023B01).
∗
Changwei Wang and Ao Li are the corresponding authors (Email:
wangchangwei2019@ia.ac.cn; ao.li@hrbust.edu.cn).
as the primary method for infrared small object detection.
However, almost all existing networks adopt classic downsam-
pling schemes. Infrared small objects, due to their small size,
often come with weak thermal signals and unclear contours.
There is a significant risk of information loss during multiple
downsampling processes. Second, compared to visible light
images, infrared images lack physical information and have
lower contrast, making small objects easily submerged in
complex backgrounds.
To tackle these challenges, We propose an infrared small
object detection model named HCF-Net. This model aims
for a more precise depiction of object shape and boundaries,
enhancing the accuracy of object localization and segmentation
by framing infrared small object detection as a semantic
segmentation problem. As illustrated in Fig. 1, it incorporates
three key modules: PPA, DASI, and MDCR, which address
the challenges mentioned above on multiple levels.
Specifically, as a primary component of the encoder-
decoder, PPA employs hierarchical feature fusion and at-
tention mechanisms to maintain and enhance representations
of small objects, ensuring crucial information is preserved
through multiple downsampling steps. DASI enhances the
skip connection in U-Net, focusing on the adaptive selection
and delicate fusion of high and low-dimensional features to
enhance the saliency of small objects. Positioned deep within
the network, MDCR reinforces multi-scale feature extraction
and channel information representation, capturing features
across various receptive field ranges. It more finely models
the differences between objects and backgrounds, enhancing
its ability to locate small objects. The organic combination
of these modules enables us to address the challenges of
small object detection more effectively, improving detection
performance and robustness.
In summary, our contributions in this paper can be summa-
rized as follows:
• We model infrared small object detection as a semantic
segmentation problem and propose HCF-Net, a layer-
arXiv:2403.10778v1 [cs.CV] 16 Mar 2024
资源评论
pk_xz123456
- 粉丝: 2970
- 资源: 4150
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
最新资源
- 2000-2020年31省份国内市场一体化与分割指数(excel版)-最新出炉.zip
- 1999-2022年中国对各国机电产品进出口额数据(出口额、进口额)-最新出炉.zip
- 2000-2020年300+地级市土地出让数据汇总-最新出炉.zip
- 单相并网逆变,heric电路设计,基于sogi-pll锁相功能,直接功率调节,动态性能好,稳定可靠
- 光储直流微电网simulink仿真模型 双向变器 ,独立光伏系统能量管理,最大功率点跟踪mppt 在传统的独立光伏发电系统中,蓄电池直接与直流母线相连接,其充放电电流不能得到有效的控制,当负载突变时
- 欧姆龙PLC解密软件,支持(CP1H、CP1E、CJ2M、CP2E)USB口直读,不破坏源程序,可读UM读取保护密码和任务读保护密码 因为有的plc设置了禁止上传,不保证所有都能读出来
- 在线电影推荐系统.7z
- springboot基于java的火车票订票系统的设计与实现.zip
- 1988-2020年各省犯罪率长期面板数据汇总-最新出炉.zip
- 1989-2022年省级人均受教育年限数据-最新出炉.zip
- 1988-2020年分省份犯罪率长期面板数据统计-最新出炉.zip
- 1990-2020年全国297个地级市R&D人员数据统计-最新出炉.zip
- 2000-2021年各省份环境污染责任保险收入及总保费收入(绿色保险数据)-最新出炉.zip
- 2000-2021年各省外商直接投资全数据-最新出炉.zip
- 2000-2021年全国范围人口分布栅格数据-最新出炉.zip
- 2000-2021年投资效率Richardson模型(非效率投资情况,OLS和GMM)数据-最新出炉.zip
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功