全文总字数：18123字

基于无人机的林木光谱检测模型分析

摘要

作物产量事关农业经济。对农作物产量的提前预测，在农业生产成本投入、决策及实现精细化、准确化农业经营管理起到至关重要的作用，同时对国家粮食安全和农业政策的制定与执行具有重要的研究意义。现阶段随着精准农业的发展，人们对作物生产的估算精度和成本提出了更高的要求，而提高作物估产准确度的一个重要途径就是有效改进产量估算方法。在当前“互联网＋”农业的背景下，无人机以及无人机图像的发展和应用为作物的生长监测提供了更加信息化、智能化、高效化的新方法，这将是无损、快速、实时监测作物生长状况和产量的未来发展趋势。无人机平台较地面和卫星遥感平台具有数据获取效率高、成本低、飞行时间和飞行高度自由度较高等优点，为中小型区域遥感监测的发展提供了新的技术手段。同时机器学习、深度学习等算法对于预测模型的良好应用，也为估产模型的准确性研究提供了保证。

本研究旨在传统遥感监测平台对作物长势监测和产量预测的基础上，将新兴的无人机低空遥感平台应用于中小型区域作物的监测和预测，明确低空无人机平台获取多光谱影像的预处理流程，建立一套较为完整且普适性强的无人机平台多光谱影像预处理方法；探讨无人机平台监测梨树长势和预测产量的可行性，通过植被指数的获取和筛选，构建基于无人机平台的梨树关键生育时期主要生长指标的动态监测模型，并综合利用多个生育时期的光谱数据，构建梨产量预测模型。预期结果将为果园尺度的农田信息监测提供有效技术途径，促进我国精确农业的快速发展与应用实践。

关键词：无人机，多光谱，估产模型，植被指数，机器学习

Analysis of forest spectrum detection model based on UAV

Abstract

Crop yield is related to agricultural economy. The prediction of crop yield in advance plays an important role in the decision-making of agricultural production cost input and the realization of refined and accurate agricultural management. At the same time, it has important research significance for the formulation and implementation of national food security and agricultural policy. At present, with the development of precision agriculture, people put forward higher requirements for the estimation accuracy and cost of crop production, and an important way to improve the accuracy of crop yield estimation is to effectively improve the yield estimation method. Under the current background of 'Internet plus' agriculture, the development and utilization of UAV and UAV image provide a new method of more information, intelligence and efficiency for crop growth monitoring, which will be the future development trend of non-destructive, rapid and real-time monitoring of crop growth and yield. Compared with the ground and satellite remote sensing platforms, UAV platform has the advantages of high data acquisition efficiency, low cost, high degree of freedom in flight time and altitude, which provides a new technical means for the development of small and medium-sized regional remote sensing monitoring. At the same time, the good application of machine learning, deep learning and other algorithms for the prediction model also provides a guarantee for the accuracy of the yield estimation model.