基于混合蝙蝠智能算法优化PLSR的柑橘叶片氮含量预测方法  

Predicting Citrus Leaf Nitrogen Content Based on Hybrid Bat Algorithm Optimized PLSR

作　　者：刘霞 周凌霄 夏心杰 韩如婷 吕强 谢让金 易时来 LIU Xia;ZHOU Lingxiao;XIA Xinjie;HAN Ruting;LYU Qiang;XIE Rangjin;YI Shilai(Citrus Research Institute,Southwest University/National Citrus Engineering Research Center/National Digital Planting(Citrus)Innovation Sub-Center,Chongqing 400712,China)

机构地区：[1]西南大学柑桔研究所/国家柑桔工程技术研究中心/国家数字种植业(柑橘)创新分中心,重庆400712

出　　处：《西南大学学报（自然科学版）》2025年第2期160-170,共11页Journal of Southwest University(Natural Science Edition)

基　　金：国家重点研发计划项目(2018YED0700602);重庆市技术创新与应用发展专项面上项目(cstc2020jscx-msxmX0026)。

摘　　要：高光谱成像技术具有快速、无损检测的特点,尤其在农作物生长监测方面具有广阔的应用前景。叶片氮含量是评估作物生长状况的重要指标之一,对诊断作物生长状况和制定精准施肥策略至关重要。通过获取柑橘叶片高光谱和检测叶片氮含量,探索二者的数学关系与模型,旨在建立柑橘叶片氮含量高光谱监测技术。为提升高光谱数据建模的精度,采用了标准正态变量变换(SNV)、平滑滤波函数(SG)等方法去除光谱数据中的噪声。利用竞争自适应加权采样法(CARS)和连续投影算法(SPA)筛选出与叶片氮含量关联度高的特征波段,结合偏最小二乘回归(PLSR)、支持向量机回归(SVR)、基于遗传算法(GA)和蝙蝠算法(BA)等对PLSR进行智能优化(O-PLSR)后预测叶片氮含量。结果表明:与SVR模型相比,PLSR模型呈现更高的精度;结合GA和BA算法对PLSR模型进行优化处理,进一步提高了建模精度,相较原始PLSR模型,决定系数(R^(2))最高提高了14.8%,且经SG滤波、CRAS特征波段选取、O-PLSR优化后的模型(SG-CRAS-O-PLSR)表现出最优的估算性能,其R^(2)、均方根误差分别为0.94和0.55。由此可见,建立的SG-CARS-O-PLSR模型具有较高的精度,可为今后推进实施果园信息化与智能化精准高效施肥管理提供理论依据和技术支持。Hyperspectral imaging technology,characterized by its rapid and non-destructive detection capabilities,holds promising application prospects,particularly in agricultural fields such as crop growth monitoring.Leaf nitrogen content serves as an important indicator for assessing crop growth conditions,which is vital for understanding crop development and establishing precise fertilization strategies.This study focused on exploring the potential application of hyperspectral technology in assessing the nitrogen content of citrus leaves by collecting both hyperspectral data and corresponding nitrogen content data of the leaves.To enhance the preprocessing effect of the hyperspectral data,methods such as Standard Normal Variate Transformation(SNV)and Savitzky-Golay(SG)smoothing filters were utilized to eliminate noise from the spectral data.The Competitive Adaptive Reweighted Sampling(CARS)and Sequential Projection Algorithm(SPA)were adopted for screening the featured bands closely associated with leaf nitrogen content,and combined with various models including Partial Least Squares Regression(PLSR),Support Vector Regression(SVR)to optimize the PLSR(O-PLSR)based on Genetic Algorithm(GA)and Bat Algorithm(BA)for predicting the leaf nitrogen content.The results indicate that compared to the SVR model,the PLSR model exhibited higher precision.Moreover,optimizing the PLSR model via GA and BA algorithms facilitated further enhancement of modelling precision,with the R^(2) value increasing by up to 14.8%.Notably,the SG-filtered and CARS-optimized PLSR model(SG-CARS-O-PLSR)demonstrated optimal performance in estimating accuracy,with a determination coefficient(R^(2))of 0.94 and Root Mean Square Error of 0.55.The high precision of the SG-CARS-O-PLSR model underscores the practical value of hyperspectral imaging technology in agriculture,particularly in the monitoring the level of nitrogen in crop.This contributes to the advancement of intelligent orchard management and the implementation of precision agriculture.

关 键 词：柑橘 高光谱成像技术 叶片氮含量 偏最小二乘回归 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]