基于RF和SPA的无人机高光谱估算棉花叶片全氮含量

为分析棉花叶片全氮含量（LNC）与冠层光谱反射特征的关系，实现作物生长过程中氮素水平的快速、准确和无损监测，以石河子大学教学试验场2019年棉花小区试验为基础，选用多元散射校正、SG平滑算法、变量标准化校正和一阶导数4种方法分别对棉花冠层原始光谱进行预处理，使用随机蛙跳（random frog,RF）和连续投影算法（successive projections algorithm,SPA）筛选特征波长并结合偏最小二乘回归法建立棉花LNC光谱估算模型。RF和SPA算法从棉花冠层398～1000nm的光谱中优选5组LNC的敏感特征波段，波段数目下降了93.0%～96.3%，有效降低了光谱的冗余信息；基于SPA算法筛选的敏感波段构建的LNC偏最小二乘回归模型的决定系数和均方根误差分别为0.52和2.55，模型验证的决定系数和均方根误差分别为0.70和2.37，模型具有较好的精度和稳定性，可作为棉花LNC的无人机高光谱估算方法。

Unmanned Aerial Vehicle Hyperspectral Estimation of Nitrogen Content in Cotton Leaves Based on RF and SPA

In order to analysis the relationship between cotton leaf nitrogen content (LNC) and canopy spectral reflection characteristics, and achieve rapid, accurate and nondestructive monitoring of nitrogen level in the process of crop growth, based on the cotton plot experiment in Xinjiang Shihezi University Teaching and Experimental Ground in 2019, multiplicative scatter correction, smoothing algorithm, standard normal variate and first derivative were used to pretreat the original spectra of the cotton canopy. Random frog (RF) algorithm and successive projections algorithm (SPA) were used to select characteristic wavelengths and combine with partial least squares regression to establish a spectral prediction model of nitrogen content in cotton leaves. RF and SPA algorithms selected five sensitive characteristic bands of LNC from the cotton canopy spectra of 398-1000nm, and the number of bands decreased by 93.0%-96.3%, which effectively reduced the spectral redundancy information. The determination coefficient and root mean square error of partial least squares regression of LNC constructed based on SPA algorithm screening sensitive band were 0.52 and 2.55, respectively, and the determination coefficient and root mean square error of model verification were 0.70 and 2.37, respectively. The model had good accuracy and stability. It can be used as a method to estimate nitrogen content in cotton leaves by unmanned aerial vehicle.