基于卷积神经网络与特征融合的天气识别方法

侯晓明;邱亚峰

应用光学 ›› 2023, Vol. 44 ›› Issue (2) : 323-329.

应用光学 ›› 2023, Vol. 44 ›› Issue (2) : 323-329. DOI: 10.5768/JAO202344.0202004

基于卷积神经网络与特征融合的天气识别方法

  • 侯晓明1, 邱亚峰1
作者信息 +

Weather recognition method based on convolutional neural network and feature fusion

  • HOU Xiaoming1, QIU Yafeng1
Author information +
文章历史 +

摘要

在太阳能热水器及太阳能电池等太阳能发电领域,下雨、下雪、阴天等气候因素将严重影响发电效果,而太阳能随动系统工作也必须消耗能量,所以迅速判断当前的天气状况,并设计自适应的开关随动系统极其重要。当天气状况为阴雨或者雪天时,系统应当关闭从而减少能耗。鉴于传统的天气识别方法效率低、准确度差、计算量大的问题,在公开的天气图像基础上创建了一个具有多种类别的天气分类集,并提供了一种基于卷积神经网络与特征融合的天气图像识别技术。通过采用传统方式获取图像的颜色、纹理、形状3种特征作为整个模型的底层特征,在原本的VGG16(visual geometry group-16)模型基础上进行了改进,从而提取图像的深层特征,最后将底层特征与深层特征融合起来在Softmax上进行输出,总识别率达到94%。

Abstract

In the field of solar power generation such as solar water heaters and solar cells, the climate factors such as rain, snow and cloudy days will seriously affect the power generation effect, and the work of solar servo system must also consume the energy. Therefore, it is extremely important to quickly judge the current weather conditions and design an adaptive on-off servo system. When the weather is rainy or snowy, the system should be shut down to reduce the energy consumption. In view of the problems of low efficiency, poor accuracy and large amount of calculation of traditional weather recognition methods, a weather classification set with multiple categories on the basis of public weather images was created, and a weather image recognition technology based on convolutional neural network and feature fusion was provided. By using the traditional way to obtain the color, texture and shape of the image as the bottom features of the whole model, it was improved on the basis of the original visual geometry group-16 (VGG16) model, so as to extract the deep features of the image. Finally, the bottom features and deep features were fused and output on Softmax, and the total recognition rate is 94%.

关键词

卷积神经网络 / 特征融合 / 天气识别 / 随动系统

Key words

servo system / feature fusion / convolutional neural network / weather recognition

引用本文

导出引用
侯晓明, 邱亚峰. 基于卷积神经网络与特征融合的天气识别方法. 应用光学. 2023, 44(2): 323-329 https://doi.org/10.5768/JAO202344.0202004
HOU Xiaoming, QIU Yafeng. Weather recognition method based on convolutional neural network and feature fusion. Journal of Applied Optics. 2023, 44(2): 323-329 https://doi.org/10.5768/JAO202344.0202004

基金

国防预研基金(1171011485)

参考文献

路绍琰, 吴丹, 马来波, 等. 中国太阳能利用技术发展概况及趋势[J]. 科技导报,2021,39(19):66-73.
韩晨, 李明杰. 热力型向日葵太阳能随动系统设计[J]. 科技与创新,2021(18):63-64.
蔡荣山, 杨勇, 张虹, 等. 太阳能电池自动实时逐日系统设计[J]. 可再生能源,2016,34(6):797-802.
罗毅欣. 智能逐日控制系统设计与实现[D]. 沈阳: 东北大学, 2011.
杨文佳, 朱海龙, 刘靖宇. 基于卷积神经网络的天气现象识别方法研究[J]. 智能计算机与应用,2019,9(6):214-216.
王辉. 基于特征融合的视频单样本人脸识别技术研究[D]. 杭州: 浙江工业大学, 2017.
蔡伟, 徐佩伟, 杨志勇, 等. 复杂背景下红外图像弱小目标检测[J]. 应用光学,2021,42(4):643-650.
左杰格, 柳晓鸣, 蔡兵. 基于图像分块与特征融合的户外图像天气识别[J]. 计算机科学,2022,49(3):197-203.
周志华. 机器学习[J]. 中国民商,2016,25(3):93-106.
徐嘉杰,
LU Shaoyan, WU Dan, MA Laibo, et al. Development situation and trend of solar energy utilization technology in China[J]. Science & Technology Review,2021,39(19):66-73.
HAN Chen, LI Mingjie. Design of thermal sunflower solar servo system[J]. Science and technology innovation,2021(18):63-64.
CAI Rongshan, YANG Yong, ZHANG Hong, et al. Design of solar cell automatic real-time day-to-day system[J]. Renewable Energy Resources,2016,34(6):797-802.
LUO Yixin. Design and implementation of intelligent day-to-day control system[D]. Shenyang: Northeastern University, 2011.
VILLARREAL GUERRA J C, KHANAM Z, EHSAN S , et al. Weather classification: a new multi-class dataset, data augmentation approach and comprehensive evaluations of convolutional neural networks [C]// 2018 NASA/ESA Conference on Adaptive Hardware and Systems (AHS). USA: IEEE, 2018.
YANG Wenjia, ZHU Hailong, LIU Jingyu. Research on weather phenomenon recognition method based on convolutional neural network[J]. Intelligent Computer an

21

Accesses

0

Citation

Detail

段落导航
相关文章

/