| 深度学习在裸地扬尘源监测中的应用研究 |
| |
| 引用本文： | 邱昀,姜磊,李金香,李令军,刘保献,鹿海峰,沈秀娥,孙爽.深度学习在裸地扬尘源监测中的应用研究[J].中国环境监测,2023,39(S1):72-79. |
| |
| 作者姓名： | 邱昀 姜磊 李金香 李令军 刘保献 鹿海峰 沈秀娥 孙爽 |
| |
| 作者单位： | 1. 北京市生态环境监测中心;2. 大气颗粒物监测技术北京市重点实验室 |
| |
| 基金项目： | 国家重点研发项目(2021YFC1809000)； |
| |
| 摘 要： | 裸地是扬尘的重要来源，施工建设过程中形成的裸地极易在大风天气作用下造成扬尘污染。因此，快速、有效地定位裸地位置，并确认其管控措施落实情况，对于开展裸地扬尘源监管具有重要意义。基于高分辨率遥感监测数据，结合人工解译裸地扬尘源数据集，以北京市大兴区为例，利用深度学习方法对裸地和防尘网覆盖裸地进行分类识别。同时，利用颜色匹配法对大兴区防尘网覆盖裸地进行识别，横向评估深度学习方法的识别精度。结果显示：深度学习方法对防尘网覆盖裸地的识别精度达97%,对裸地的识别精度达61%;颜色匹配法对防尘网覆盖裸地的识别精度达85%。防尘网覆盖裸地的颜色特征鲜明，深度学习方法和颜色匹配法对防尘网覆盖裸地的识别精度都在85%以上。深度学习方法对于面积大于2 000 m2的图斑有着较好的识别精度。深度学习方法可以提高裸地遥感解译的效率，实现规范化图像识别，可以作为人工判读的辅助手段。在实际应用中，可通过进一步积累样本来增强模型性能。深度学习方法适用于裸地扬尘源线索快速发现、工地防尘网措施落实情况快速检测等场景。
|
| 关 键 词： | 深度学习 施工裸地 防尘网覆盖裸地 扬尘源 遥感监测 分类识别 |
| 收稿时间： | 2023/5/16 0:00:00 |
| 修稿时间： | 2023/9/18 0:00:00 |
Research on the Application of Deep Learning in Monitoring Dust Sources in Bare Land |
| |
| QIU Yun,JIANG Lei,LI Jinxiang,LI Lingjun,LIU Baoxian,LU Haifeng,SHEN Xiue,SUN Shuang.Research on the Application of Deep Learning in Monitoring Dust Sources in Bare Land[J].Environmental Monitoring in China,2023,39(S1):72-79. |
| |
| Authors: | QIU Yun JIANG Lei LI Jinxiang LI Lingjun LIU Baoxian LU Haifeng SHEN Xiue SUN Shuang |
| |
| Affiliation: | Beijing Municipal Ecological and Environmental Monitoring Center, Beijing 100048, China;Beijing Municipal Ecological and Environmental Monitoring Center, Beijing 100048, China;Beijing Key Laboratory of Atmospheric Particulate Monitoring Technology, Beijing 100048, China |
| |
| Abstract: | Bare land is an important source of fugitive dust.The bare land formed in the process of construction is easy to cause dust pollution under the action of strong wind.Therefore,it is of great significance to quickly and effectively locate bare land and confirm the implementation of its control measures for the supervision of fugitive dust.Based on high-resolution remote sensing monitoring data,deep learning method was used to classify bare land and anti-dust mesh covered bare land with Daxing artificial interpreted bare land fugitive dust source data set.The color matching method was used to identify anti-dust mesh covered bare land in Daxing District.The accuracy of these two methods were compared.The results showed that the accuracy of anti-dust mesh covered bare land identification was 97% and the accuracy of bare land identification was 61% by using deep learning method,the accuracy of color matching method for anti-dust mesh covered bare land identification was 85%.The anti-dust mesh covered bare land has distinct color characteristics,resulting in the recognition accuracy was above 85% by using both of deep learning method and color matching method.The accuracy of deep learning method has a better identification on the patterns with the area of which are more than 2000 square meters.Deep learning method can improve the efficiency of remote sensing interpretation of bare land,and realize standardized image recognition,which can be used as a method to assist manual interpretation.In practical application,the model performance can be enhanced by further accumulating samples.It is applicable to the rapid detection of bare land clues,and the implementation of anti-dust mesh in construction site based on deep learning method. |
| |
| Keywords: | deep learning construction bare land anti-dust mesh covered bare land dust sources remote sensing classification identification |
|
| 点击此处可从《中国环境监测》浏览原始摘要信息 |
|
点击此处可从《中国环境监测》下载全文 |
|
