王贝贝 姚艳丽 张争 王苗苗 郭琰煊

摘要 研究植被生态自动化观测方法，进而确定仪器的组成结构，根据传感器的接口、协议类型和神经网络处理器的应用方式设计了基于ARM架构控制处理器的外设接口，进一步结合非RKNN模型的转换应用流程，设计了控制处理器数据采集、处理、存储、传输和控制交互的具体功能和流程。针对3种不同的植被类型，将仪器分别部署在内蒙古、广西和安徽3个国家级农业气象试验基地，采用仪器测量数据与人工平行观测的方法进行外场试验。结果表明，这种植被生态自动化观测方法是可行的、有效的，仪器性能指标也符合实际业务观测要求。

关键词 植被生态自动化观测;深度学习;植被关键特征参数测量

中图分类号 TP391 文献标识码 A 文章编号 0517-6611（2021）17-0201-05

doi：10.3969/j.issn.0517-6611.2021.17.052

Abstract This paper studied the observation method of vegetation ecological automation， and then determined the composition and structure of the instrument. According to the sensor interface， protocol type and application mode of neural network processor， the peripheral interface of control processor based on ARM architecture was designed. Furthermore， combined with the application process of non RKNN model conversion，the specific function and process of the data acquisition， processing， storage and transmission， control interaction of control processor were designed. According to three different vegetation types， the instruments were deployed in three national agrometeorological experimental bases in Inner Mongolia， Guangxi and Anhui， and the field test was carried out by using the method of instrument measurement data and manual parallel observation. The results showed that the automatic observation method of vegetation ecology was feasible and effective， and the performance index of the instrument was also in line with the actual business observation requirements.

Key words Automatic observation of vegetation ecology;Deep learning;Measurement of key characteristic parameters of vegetation

植被生態观测作为生态气象自动观测中的重要项目，对于提高我国生态气象的综合观测能力，建立生态气象监测网络，开展气候与气候变化对生态环境质量的影响评估、气象灾害对生态安全的预警评估具有重要意义。目前的测量方法有小尺度的人工观测法和大尺度的遥感监测法，其中人工观测法因为其对植物冠层破坏大，耗费人力和物力，数据处理烦琐，正在逐渐淘汰。遥感监测法主要利用卫星遥感资料对地表植被进行监测估算[1-6]，是当前植被生态监测的主要手段，但只适合大尺度测量，监测频次低、局部测量精度低并且缺少星地校验[7]。

针对中尺度观……