关键词: 水稻, SPAD值, 直播, 无人机, 径向基神经网络

Abstract: Accuately obtaining chlorophyll content in direct‑seeded rice is of great significance for innovating unmanned rice direct seedling cultivation management technologies.Field experiments were conducted to identify the optimal monitoring model for the relative chlorophyll content（SPAD value）in leaves of direct‑seeded rice.This study systematically examined the correlations between 13 commonly used multispectral feature indices and SPAD values，followed by a comparative analysis of SPAD estimation results derived from four modeling approaches：Particle swarm optimization‑support vector machine（PSO‑SVM），random forest（RF），radial basis function（RBF）neural network，and least squares support vector machine（LSSVM）.The results showed that the SPAD value of direct‑seeded rice leaves exhibited significant variation with growth progression under different treatments，and the SPAD values at the same growth stage generally followed the trend：N4（N 240 kg/ha）>N3（N 195 kg/ha）>N2（N 150 kg/ha）>N1（N 75 kg/ha）>N0（N 0 kg/ha）.During the three critical growth periods（the tillering，jointing，heading periods）of direct‑seeded rice，the vegetation indices NDVI，RVI，SAVI，CI_green，and GNDVI all showed strong correlations with SPAD values，with the absolute values of correlation coefficients reaching 0.838，0.783，0.838，0.671，and 0.690，respectively. Independent validation using PSO‑SVM，RF，RBF，and LSSVM models yielded determination coefficients（R_cv²）of 0.770，0.771，0.857，and 0.773，respectively.This indicates that the RBF neural network‑based model provides the best predictive performance for monitoring SPAD values in the leaves of direct‑seeded rice.

Key words: Rice, SPAD values, Direct?seeding, Unmanned aerial vehicle, Radial basis function neural network