Ground‐based cloud recognition method based on an improved DeepLabV3+ model

Author:

Liang Yue1ORCID,Ge Quanbo1

Affiliation:

1. School of Automation Nanjing University of Information Science and Technology Nanjing China

Abstract

AbstractAn improved method for recognising clouds on the ground map is proposed incorporating the DeepLabV3+ model to solve three issues. Firstly, an image preprocessing module is developed to enrich image quality, particularly at night, as clouds can appear to be too blurry to distinguish. Secondly, the CBAM attention mechanism is applied to safeguard texture and boundary information, ensuring the depiction of cirrus edges is not lost. Finally, the feature extraction network can be optimised to enhance the model and decrease its computational complexity. In comparison to the original model, the accuracy of the proposed method increased by 10.89%, resulting in an overall accuracy of 94.18%. Furthermore, the MIoU has improved from 66.02% to 79.31%. The number of parameters was reduced by 51.45% to 13.4 M. Of the various cloud types, the improvement in cirrus is particularly striking, with the MIoU increasing from 1.78% to 56.01%.

Publisher

Institution of Engineering and Technology (IET)

Subject

Artificial Intelligence,Cognitive Neuroscience,Computer Science Applications,Computer Vision and Pattern Recognition,Experimental and Cognitive Psychology

Reference34 articles.

1. CloudNet: Ground‐Based Cloud Classification With Deep Convolutional Neural Network

2. Cai D.:Design of cortex‐m lightweight neural network based ground‐based cloud map recognition. Master’s thesis.Liaocheng University(2022)

3. Zhu T.:Research on inter‐hour forecast of direct normalirradiance based on ground‐based cloud images. PhD thesis.Southeast University(2019)

4. A method for observing and counting day‐time cloud amounts in an astronomical site survey;Shi Y.;Astronomical Res. Technol.,2008

同舟云学术

1.学者识别学者识别

2.学术分析学术分析

3.人才评估人才评估

"同舟云学术"是以全球学者为主线,采集、加工和组织学术论文而形成的新型学术文献查询和分析系统,可以对全球学者进行文献检索和人才价值评估。用户可以通过关注某些学科领域的顶尖人物而持续追踪该领域的学科进展和研究前沿。经过近期的数据扩容,当前同舟云学术共收录了国内外主流学术期刊6万余种,收集的期刊论文及会议论文总量共计约1.5亿篇,并以每天添加12000余篇中外论文的速度递增。我们也可以为用户提供个性化、定制化的学者数据。欢迎来电咨询!咨询电话:010-8811{复制后删除}0370

www.globalauthorid.com

TOP

Copyright © 2019-2024 北京同舟云网络信息技术有限公司
京公网安备11010802033243号  京ICP备18003416号-3