基于图像感兴趣区域融合特征的PM_(2.5)浓度预测方法  

PM_(2.5) Concentration Prediction Based on Fusion Features of Image Regions of Interest

作　　者：张雨琪 焦瑞莉[1] 薄宇 陶益凡 王立志[2] ZHANG Yuqi;JIAO Ruili;BO Yu;TAO Yifan;WANG Lizhi(School of Communication Engineering,Beijing Information Science and Technology University,Beijing 100101,China;Key Laboratory of Regional Climate and Environment for Temperate East Asia,Institute of Atmospheric Physics,Chinese Academy of Sciences,Beijing 100029,China;Public Technology Service Center,Institute of Atmospheric Physics,Chinese Academy of Sciences,Beijing 100029,China)

机构地区：[1]北京信息科技大学信息与通信工程学院,北京100101 [2]中国科学院大气物理研究所中国科学院东亚区域气候-环境重点实验室,北京100029 [3]中国科学院大气物理研究所公共技术服务中心,北京100029

出　　处：《中国环境监测》2022年第4期207-216,共10页Environmental Monitoring in China

基　　金：中国科学院战略性先导科技专项(A类-XDA19040202);北京信息科技大学其他纵向项目(20190193)。

摘　　要：通过图像预测PM_(2.5)浓度的准确性,在很大程度上取决于模型所选用的特征参数。为丰富特征参数的表达,设计了一种基于图像传统特征与深度特征充分融合的PM_(2.5)浓度预测方法。首先,根据不同PM_(2.5)浓度下的成像差异,选定图像感兴趣区域,解决图像尺寸过大导致的模型运算效率较低问题。然后,针对所选取的局部图像,利用传统图像处理方法手动设计并提取图像浅表视觉特征,同时利用卷积神经网络自动提取图像深层语义特征。最后,将两种特征融合,交由卷积神经网络的全连接层实现对PM_(2.5)浓度的回归预测。预测误差比对结果显示,相比使用单种特征,使用融合特征能够有效提高模型的预测性能。The accuracy of PM_(2.5) concentration prediction using image largely depends on the features extracted by the model.In order to enrich the expression of feature parameters,a PM_(2.5) concentration prediction method based on the fusion of traditional image features and depth features is designed.Firstly,based on the imaging differences under different PM_(2.5) concentrations,the region of interest in the image is selected to solve the problem of low model efficiency caused by the excessively large image size.Secondly,the traditional image processing method is used to manually design and extract the superficial visual features of the selected partial image,and the convolutional neural network is used to automatically extract the deep semantic features of the image.Finally,the two features are merged,and the regression prediction of PM_(2.5) concentration is achieved through the fully connected layer of the convolutional neural network.Comparison on the prediction error of using a single feature and a fusion feature shows that the prediction performance of the model can be effectively improved by the fusion feature.

关 键 词：大气污染监测 PM_(2.5)浓度 感兴趣区域 传统图像处理 深度学习 融合特征 

分 类 号：X830.2[环境科学与工程—环境工程]