基于Faster R-CNN的气象设备观测环境影响图像研究  

Research on the impact of environmental observation images on meteorological equipment based on the Faster R-CNN

作　　者：王超然 周若[2] 李中华[2] 邬昀[2] 白子诚 WANG Chaoran;ZHOU Ruo;LI Zhonghua;WU Yun;BAI Zicheng(Huangshi Meteorological Bureau,Huangshi 435000,China;Hubei Meteorological Information and Technological Support Centre,Wuhan 430074,China)

机构地区：[1]湖北省黄石市气象局,湖北黄石435000 [2]湖北省气象信息与技术保障中心,湖北武汉430074

出　　处：《电子设计工程》2025年第4期128-132,共5页Electronic Design Engineering

基　　金：黄石市气象局科研基金项目(202311)。

摘　　要：为确保气象站点实况观测数据的准确性,观测场地需要有良好的环境。该研究利用深度学习技术中的Faster R-CNN模型,自动检测气象观测站场景中可能干扰仪器读数的视觉障碍物。通过收集并详细标注实际观测场景的图像数据,建立一个包含环境对观测设备产生影响情况的数据集,涵盖正常与异常环境情况的百叶箱和雨量筒图像及其标注信息。对Faster R-CNN模型进行微调和超参数优化,以适应该特定识别任务。实验结果验证了模型在识别和定位障碍物方面的高效性,准确率为97.1%,展现出了较好的鲁棒性。该项研究将图像识别处理用于自动站探测环境,不仅证明了深度学习在改善气象观测条件中的有效性,也为相似领域的应用提供了方法论上的指导。In order to ensure the accuracy of meteorological observation data,the observation site needs a good environment.The study uses the Faster R-CNN model in deep learning technology to automatically detect visual obstacles that may interfere with instrument readings in weather observation station scenes.By collecting and annotating the image data of the actual observation scene,a data set containing the impact of the environment on the observation equipment is established,including the shelter and rain gauge images of normal and abnormal environment conditions and their annotation information.The Faster R-CNN model is fine-tuned and hyperparameter optimized to adapt to the specific recognition task.The experimental results verify the high efficiency of the model in identifying and locating obstacles,and the accuracy rate is 97.1%,showing a good robustness.This study applies image recognition processing to automatic station detection environment,which not only proves the effectiveness of deep learning in improving meteorological observation conditions,but also provides methodological guidance for applications in similar fields.

关 键 词：深度学习 Faster R-CNN 气象观测场 图像处理 

分 类 号：TN919.8[电子电信—通信与信息系统]