Remote Sensing Sciencehttp://www.ivypub.org/journal/RSS.aspx?J=RSS&lang=cnen-USHyperspectral Estimation of Total Nitrogen Content in Orchard Soil Based on Successive Projections Algorithm2021-0<p class="abstract">Hyperspectral Estimation of Total Nitrogen Content in Orchard Soil Based on Successive Projections Algorithm</p><ul><li>Pages 1-11</li><li>Author Zizhen LiXicun ZhuZhongyu TianMeixuan LiYufeng PengXinyang YuXueyuan BaiLing WangYuanmao Jian</li><li>Abstract Soil total nitrogen content is an important factor to affect the growth of fruit trees. The objective of the paper is to propose a successive projections algorithm for hyperspectral estimation of soil total nitrogen content in orchard. A total of 60 soil samples were collected from apple orchard, Guanli Town, Qixia City, Shandong Province. The spectral reflectivity of the samples was measured by a ASD Field-Spce4.0 spectrometer in the laboratory. The total nitrogen content of the samples was measured by a Kjeldahl apparatus. The spectral data were preprocessed by Savitzky-Golay smoothing and first-order derivative transformation. The sensitive bands were screened by successive projections algorithm. Partial least squares regression (PLSR) model, random forest (RF) model and support vector machine regression (SVR) model were established respectively, and the accuracy of the models was tested and compared. The sensitive wavelengths selected by successive projections algorithm were 849, 2100, 1305 and 2427nm, the determination coefficients (R2) of PLSR, SVR and RF models were 0.822, 0.724 and 0.683, and the relative percent deviation (RPD) were 2.329, 1.940 and 1.702, respectively. PLSR model has the highest accuracy and can play a good role in the estimation of total nitrogen content in orchard soil; the accuracy of SVR and RF models is lower than that of PLSR models, but the estimated results reach the usable level.</li></ul>/RSS/paperinfo/56588.shtmlRemote Sensing Science/RSS/paperinfo/56588.shtml