温室作物三维重建与生长信息获取方法研究

China is the country of facility agriculture with vegetable cultivation as its mainstay. Its area of agricultural facilities and yield of vegetables firmly in the first place. Monitoring plants information can report the plants growth stage efficiently, which is important to lead to scientific agriculture production, Thus, design a monitor platform for vegetables with the practical requirement of agricultural production via Kinect V2 is very necessary..This project aims at getting high-precision plant information through Kinect V2 for relative researches, like agriculture management and water & fertilizer proportioning optimization..Considering the actual production process, this project will design a movable monitoring platform, which can get real time information, and research a new and effective 3D-reconstruction method for group plants in greenhouse. combined the special plants’ shape feature, this project will optimize this 3D-reconstruction method to make it suitable the same class vegetables. Combined with deep-learning technology, we design an efficient, precise, high-resolution auto-segmentation algorithm to get part labels of plants. Then get shape information of the plants automatically. Last at all, series of experiments will be done to make correction and compensation of the morphology features automatically. The relations between morphology features and non-morphology features of plants such as rate of water content and biomass, will be explored as well. This project aims to build a automatically monitoring system for plant information to provide accurate and stable plant growth parameters for water & fertilizer management and agricultural operation in greenhouse.

我国是以蔬菜栽培为主体的设施农业生产大国，设施面积与蔬菜产量稳居世界第一。及时掌握作物生长状况对科学指导农业生产具有重要意义。项目以第二代Kinect作为温室作物信息获取设备，探究水肥配比和光环境等农艺优化研究中的单株作物生长信息高精度获取方法；针对温室实际生产中实时、动态作物生长信息的实际生产需求，设计移动监测平台，研究温室环境下群体作物三维重建方法；针对作物形态特征，优化其三维建模方法，使之对同类作物的三维建模具有一定普适性；结合深度学习的方法，探究一套高效、准确、高分辨率的作物标注和目标部位自动分割方法，并在分割基础上实现作物形态参数的自动获取；在大量实验的基础上，对自动获取形态参数的模型进行补偿和修正，探究作物含水率、生物量等非形态参数和作物三维模型可提取参数的相关性；开发作物信息监测系统，实现作物参数的自动获取并探究其可扩展性，为温室内水肥管理、作物的科学生产等提供准确的作物信息

基于三维视觉系统的作物重建与生长参数获取，实时监测作物全生长周期内的生长发育状况，借助自动化和智能化技术手段搭建生长监测与生长信息管理系统，对科学指导设施温室内的农业生产具有重要意义。本研究以消费级深度相机作为主要的信息感知装备，探究其在实验室场景、玻璃温室场景以及植物工厂场景下对单株作物和群体作物的三维重建方法。根据使用场景的差异优化建模方法，搭建作物三维重建平台。将平台搭载在自动巡检机器人上，实现设施温室、植物工厂等场景下，作物三维模型的自动化获取。分析、对比传统计算机视觉系统、三维几何学以及深度学习方法在作物模型分割问题上的表现情况，通过实验验证基于二维图像的作物分割方法在分割准确性、稳定性以及时效性上均优于基于三维模型的分割方法。基于作物的形态特征，修补因视角有限导致的三维模型缺陷。通过实验验证从三维模型上提取的生长参数与实际测量值线性拟合高度相关，并且总叶面积、体积等生长参数与鲜重、干重的测量值相关性优异，可用于作物生物量的估计。搭建作物生长信息管理系统集成了数据采集、处理、管理和查看等功能。整套系统包含了作物三维重建、生长参数获取以及生长信息管理，可以为科学生产等提供准确的作物信息，为温室内水肥管理、作物的科学生产等提供准确的作物信息。