欢迎访问《棉花学报》! 今天是

棉花学报 ›› 2021, Vol. 33 ›› Issue (3): 224-234.doi: 10.11963/1002-7807.yxzlf.20210428

• 研究与进展 • 上一篇    下一篇

基于无人机高光谱融合连续投影算法估算棉花地上部生物量

易翔1(),张立福1,3,*(),吕新1,张泽1,田敏2,印彩霞1,马怡茹1,范向龙1   

  1. 1.石河子大学农学院/新疆生产建设兵团绿洲生态农业重点实验室,新疆 石河子 832003
    2.石河子大学机械电气工程学院,新疆 石河子 832003
    3.中国科学院空天信息创新研究院/遥感科学国家重点实验室,北京100094
  • 收稿日期:2020-12-28 出版日期:2021-05-15 发布日期:2021-06-02
  • 通讯作者: *张立福 E-mail:20192012003@stu.shzu.edu.cn;zhanglf@radi.ac.cn
  • 作者简介:易翔(1996―),男,硕士研究生, 20192012003@stu.shzu.edu.cn
  • 基金资助:
    国家自然科学基金(61962053);新疆生产建设兵团棉花生产大数据关键技术及农业大数据平台研发应用(2018AA004)

Estimation of cotton above-ground biomass based on unmanned aerial vehicle hyperspectral and successive projections algorithm

Yi Xiang1(),Zhang Lifu1,3,*(),Lü Xin1,Zhang Ze1,Tian Min2,Yin Caixia1,Ma Yiru1,Fan Xianglong1   

  1. 1. Agriculture College of Shihezi University/The Key Laboratory of Oasis Eco-Agriculture, Xinjiang Production and Construction Group, Shihezi, Xinjiang 832003, China
    2. College of Mechanical and Electrical Engineering, Shihezi University, Shihezi, Xinjiang 832003, China
    3. Aerospace Information Research Institute, Chinese Academy of Sciences/State Key Laboratory of Remote Sensing Science, Beijing 100094, China
  • Received:2020-12-28 Online:2021-05-15 Published:2021-06-02
  • Contact: *Zhang Lifu E-mail:20192012003@stu.shzu.edu.cn;zhanglf@radi.ac.cn

摘要:

【目的】地上部生物量是表征植物生命活动的重要参数。探索不同的光谱预处理方法和建模方法,实现对棉花地上部生物量快速、无损、准确的估算,对棉花长势监测和大田精准管理具有重要意义。【方法】以新陆早53号、新陆早45号为研究对象,设置不同施氮处理,于出苗后不同阶段获取棉花地上部生物量和无人机高光谱数据,通过连续投影算法(Successive projections algorithm,SPA)筛选不同预处理[一阶导数、二阶导数、Savitzky-Golay(SG平滑)、多元散射校正]后的特征波长,基于筛选出的不同波长组合使用偏最小二乘法回归(Partial least square regression,PLSR)和随机森林回归(Random forest regression,RFR)分别构建棉花地上部生物量估算模型,比较不同预处理后建立模型的精度,确定最优估算模型。【结果】(1)利用SPA算法对不同预处理后的光谱信息筛选出特征波长9~26个,可实现光谱信息降维。(2)基于SG平滑-SPA处理及PLSR方法建立的模型最佳,R2达到了0.63,均方根误差(Root mean square error,RMSE)为0.42,验证集的R2为0.67,RMSE为0.44。(3)一阶导数-SPA处理后,采用RFR构建的模型最佳,R2达到0.87,RMSE为0.45,验证集R2为0.81,RMSE为0.37。【结论】采用一阶导数预处理结合SPA筛选特征波长,经RFR构建的估算模型结果和验证效果均最佳,可用于棉花地上部生物量定量估算。

关键词: 棉花; 地上部生物量; 光谱预处理; 特征波长; 偏最小二乘法回归; 随机森林回归

Abstract:

[Objective] Above-ground biomass is an important parameter of plant life activity. Exploring different spectral pretreatment methods and modeling methods to achieve rapid, nondestructive and accurate estimation of cotton above-ground biomass is of great significance for cotton growth monitoring and field precision management. [Method] Xinluzao 53 and Xinluzao 45 were selected as the research objects, and different nitrogen application treatments were set up to obtain cotton above-ground biomass and UAV hyperspectral data at different stages after emergence. The successive projections algorithm (SPA) was used to select the characteristic wavelengths after different pretreatments [first derivative, second derivative, Savitzky-Golay (SG smoothing), multiple scatter correction]. Based on the selected wavelengths, partial least squares regression (PLSR) and random forest regression (RFR) were used to construct cotton above-ground biomass (dry matter) estimation models, respectively, and the optimal estimation model was determined by comparison of the model estimation accuracy. [Result] (1) The SPA algorithm can effectively remove the redundant bands from the spectra, and the number of above-ground biomass-sensitive characteristic wavelengths after different pretreatments ranges from 9 to 26, which could be used to reduce the dimension of spectral information. (2) Based on SG smoothing-SPA processing, the model established by PLSR is the best, with a coefficient of determination (R2) of 0.63, root mean square error (RMSE) of 0.42, and validation set R2, RMSE of 0.67, 0.44. (3) After the first derivative-SPA processing, the model constructed by RFR is the best, with R2 of 0.87, RMSE of 0.45, validation set R2 and RMSE of 0.81 and 0.37. [Conclusion] Using the first derivative pretreatment combined with SPA to select biomass-sensitive wavelengths, the RFR model has the best results and validation performance, which can be used for quantitative estimation of cotton above-ground biomass.

Key words: cotton; above-ground biomass; spectral pretreatment; sensitive bands; partial least squares; random forest regression