光电半导体在精准农业机器人感知方面的应用

精准农业(Precision agriculture)是在农作物生长过程中根据作物具体的时空信息对其作出定时、定量、定位的管理决策并指导机器作业, 达到最优使用资源和实现最大收益的新型智慧农业技术。搭载多传感器的农业机器人(Agricultural robot)在农业感知测量中的普及与使用，可以用来对植物进行数据采集、智能检测、自主作业等操作，对精准农业起到整体推动的作用。而现有的传感器存在只能测量一个小点、特征波段短、测量精度低、实时移动性差等不足，无法满足精准农业中的实时感知需求。为解决这些智能感知需求，需要在农业机器人上搭载高精度、高维度、高可靠性的新型传感器。 光电半导体材料以其在光学遥感、距离测量等农业应用中展现出的高精度、多光学谱段、高灵敏性的优点被大规模使用。光电半导体传感器(Optoelectronic semiconductor sensor)是把光和电这两种物理量联系起来,使光和电互相转化的新型光电半导体材料器件。其中，主要分为利用半导体光敏特性工作的光电导器件和利用半导体发光特性工作的半导体发光器件。 激光雷达(Lidar)作为开始普及应用于室外大尺度场景下的新型三维距离测量光电半导体激光(如砷化镓GaAs半导体材料)和感光(如硅Si半导体材料)器件，以其探测精度高、测量距离远、三维感知效果好被广泛关注。多光谱相机(Multispectral camera)作为另一热门研究的新型多特征波段测量的光电导器件(如硅Si,铟镓砷InAs半导体材料)，在光谱域中进一步扩展了对农业信息的感知与探测。 为了精准、可靠的探测更高维度的农业信息。本课题搭建了一个农业感知机器人,使用新型光电半导体传感器激光雷达和多光谱相机，并探究相机和激光雷达融合原理(Sensor Fusion)与机器人同时定位和建图原理(Simultaneous localization and mapping)，实时生成兼具真实尺度和多谱段信息的三维多光谱高精地图(3D multispectral high-precision map)。 在识别农业土地材质的应用中，选取多光谱波段中红光波段(650 nm)和近红外波段(840 nm)的光谱信息合成归一化差分植被指数(NDVI)图像，与激光雷达的深度信息融合生成附带植被指数信息的三维多光谱(NDVI)高精地图，解决对农业信息高精度、大尺度、多光谱波段的感知问题。此外，在农业测绘、辅助自动驾驶、等各个领域中，三维多光谱高精地图也能提供很多不可取代性的感知信息。

Precision agriculture is a new type of smart agricultural technology that makes timing, quantitative, and positioning management decisions based on the specific time and space information of crops during the growth of crops and guides machine operations to achieve optimal use of resources and achieve maximum benefits. The popularization and use of agricultural robots equipped with multi-sensors in agricultural perception measurement can be used to perform data collection, intelligent detection, and autonomous operations on plants, and play an overall role in promoting precision agriculture. However, the existing sensors can only measure a small point, the characteristic band is short, the measurement accuracy is low, and the real-time mobility is poor, and they cannot meet the real-time sensing requirements in precision agriculture. In order to solve these intelligent sensing requirements, it is necessary to carry new sensors with high precision, high dimensions and high reliability on agricultural robots. Optoelectronic semiconductor materials are used on a large scale due to their advantages of high precision, multiple optical bands, and high sensitivity in agricultural applications such as optical remote sensing and distance measurement. The optoelectronic semiconductor sensor is a new type of optoelectronic semiconductor material device that connects the two physical quantities of light and electricity, and converts light and electricity to each other. Among them, it is mainly divided into photoconductive devices that use semiconductor photosensitivity to work and semiconductor light-emitting devices that use semiconductor light-emitting characteristics. As a new type of three-dimensional distance measurement photoelectric semiconductor laser (such as gallium arsenide GaAs semiconductor material) and photosensitive (such as silicon Si semiconductor material) devices that have begun to be widely used in outdoor large-scale scenes, Lidar has high detection accuracy and measurement The long distance and good three-dimensional perception effect are widely concerned. Multispectral camera, as another hot research new type of multi-characteristic waveband measurement photoconductive device (such as silicon Si, indium gallium arsenide InAs semiconductor materials), further expands the perception and detection of agricultural information in the spectral domain. In order to accurately and reliably detect higher-dimensional agricultural information. This project builds an agricultural perception robot, using new photoelectric semiconductor sensor lidar and multispectral camera, and explores the principle of camera and lidar fusion (Sensor Fusion) and the principle of simultaneous positioning and mapping of the robot (Simultaneous localization and mapping), which is generated in real time A 3D multispectral high-precision map (3D multispectral high-precision map) with both true scale and multispectral information. In the application of identifying the material of agricultural land, select the spectral information of the red light band (650 nm) and the near-infrared band (840 nm) in the multi-spectral band to synthesize the normalized differential vegetation index (NDVI) image, and the depth information of the lidar Fusion generates a three-dimensional multi-spectral (NDVI) high-precision map with vegetation index information to solve the problem of high-precision, large-scale, and multi-spectral perception of agricultural information. In addition, three-dimensional multi-spectral high-precision maps can also provide a lot of irreplaceable perceptual information in various fields such as agricultural surveying and mapping, assisted autonomous driving, and so on.