Integration of Multiple Spectral Data via a Logistic Regression Algorithm for Detection of Crop Residue Burned Areas:A Case Study of Songnen Plain,Northeast China

The burning of crop residues in fields is a significant global biomass burning activity which is a key element of the terrestrial carbon cycle,and an important source of atmospheric trace gasses and aerosols.Accurate estimation of cropland burned area is both crucial and challenging,especially for the small and fragmented burned scars in China.Here we developed an automated burned area mapping algorithm that was implemented using Sentinel-2 Multi Spectral Instrument(MSI)data and its effectiveness was tested taking Songnen Plain,Northeast China as a case using satellite image of 2020.We employed a logistic regression method for integrating multiple spectral data into a synthetic indicator,and compared the results with manually interpreted burned area reference maps and the Moderate-Resolution Imaging Spectroradiometer(MODIS)MCD64A1 burned area product.The overall accuracy of the single variable logistic regression was 77.38%to 86.90%and 73.47%to 97.14%for the 52TCQ and 51TYM cases,respectively.In comparison,the accuracy of the burned area map was improved to 87.14%and 98.33%for the 52TCQ and 51TYM cases,respectively by multiple variable logistic regression of Sentind-2 images.The balance of omission error and commission error was also improved.The integration of multiple spectral data combined with a logistic regression method proves to be effective for burned area detection,offering a highly automated process with an automatic threshold determination mechanism.This method exhibits excellent extensibility and flexibility taking the image tile as the operating unit.It is suitable for burned area detection at a regional scale and can also be implemented with other satellite data.

Iterative Reweighted <i>l</i>₁Penalty Regression Approach for Line Spectral Estimation

In this paper, we proposed an iterative reweighted l1?penalty regression approach to solve the line spectral estimation problem. In each iteration process, we first use the ideal of Bayesian lasso to update the sparse vectors;the derivative of the penalty function forms the regularization parameter. We choose the anti-trigonometric function as a penalty function to approximate the?l0? norm. Then we use the gradient descent method to update the dictionary parameters. The theoretical analysis and simulation results demonstrate the effectiveness of the method and show that the proposed algorithm outperforms other state-of-the-art methods for many practical cases.

Regression Relationship between WI and FMC at Different Growth Periods of Sawtooth Oaks Leaf

[Objective] The aim was to study the regression relationship between water index （WI） and fuel moisture content （FMC） of different growth periods of sawtooth oaks leaf.[Method] Taking sawtooth oaks in Huaguo Mountain,Lianyungang City as research object,the sensitivity of WI to leaf FMC was studied at leaf level,and statistical characteristics were analyzed.[Result] The WI of sawtooth oaks leaves was sensitive to the changes of FMC,and the line regression level between them was significant.A fitting curve between leaf FMC and WI was obtained.[Conclusion] The research provides reference for acquisition methods of vegetation water remote sensing within the range of study area.

Prediction of Soil Salinity Using Remote Sensing Tools and Linear Regression Model

Soil salinity is one of the most damaging environmental problems worldwide, especially in arid and semi-arid regions. Multispectral data Sentinel_2 are used to study saline soils in southern Tunisia. 34 soil samples were collected for ground truth data in the investigated region. A moderate correlation was found between electrical conductivity and the spectral indices from SWIR. Different spectral indices were used from original bands of Sentinel_2 data. Statistical correlation between ground measurements of Electrical Conductivity (EC), spectral indices and Sentinel_2 original bands showed that SWIR bands (b11 and b12) and the salinity index SI have the highest correlation with EC. Based on these results and combining these remotely sensed variables into a regression analysis model yielded a coefficient of determination R2 = 0.48 and an RMSE = 4.8 dS/m.

Combining Environmental Factors and Lab VNIR Spectral Data to Predict SOM by Geospatial Techniques

Soil organic matter(SOM) is an important parameter related to soil nutrient and miscellaneous ecosystem services. This paper attempts to improve the performance of traditional partial least square regression(PLSR) model by considering the spatial autocorrelation and soil forming factors. Surface soil samples(n = 180) were collected from Honghu City located in the middle of Jianghan Plain, China. The visible and near infrared(VNIR) spectra and six environmental factors(elevation, land use types, roughness, relief amplitude, enhanced vegetation index, and land surface water index) were used as the auxiliary variables to construct the multiple linear regression(MLR), PLSR and geographically weighted regression(GWR) models. Results showed that: 1) the VNIR spectra can increase about 39.62% prediction accuracy than the environmental factors in predicting SOM; 2) the comprehensive variables of VNIR spectra and the environmental factors can improve about 5.78% and 44.90% relative to soil spectral models and soil environmental models, respectively; 3) the spatial model(GWR) can improve about 3.28% accuracy than MLR and PLSR. Our results suggest that the combination of spectral reflectance and the environmental variables can be used as the suitable auxiliary variables in predicting SOM, and GWR is a promising model for predicting soil properties.

Detection of the spatiotemporal field of a single-shot terahertz pulse based on spectral holography

According to electro-optical sampling theory, we propose a new method to detect the spatiotemporal field of a single- shot terahertz pulse by spectral holography for the first time. The single-shot terahertz pulse is coupled into a broadened chirped femtosecond pulse according to electro-optical sampling theory in the detecting system. Then the reference wave and the signal wave are split by Dammann grating and spread into the interference band-pass filter. The filtered sub-waves are at different central-frequencies because of the different incident angles. These sub-waves at different central-frequencies interfere to form sub-holograms, which are recorded in a single frame of a charge coupled device （CCD）. The sub-holograms are numerically processed, and the spatiotemporal field distribution of the original terahertz pulse is reconstructed. The computer simulations verify the feasibility of the proposed method.

Quantitative retrieval of soil salt content based on measured spectral data

Choosing the Minqin Oasis, located downstream of the Shiyang River in Northwest China, as the study area, we used field-measured hyperspectral data and laboratory-measured soil salt content data to analyze the characteristics of saline soil spectral reflectance and its transformation in the area, and elucidated the relations between the soil spectral re-flectance, reflectance transformation, and soil salt content. In addition, we screened sensitive wavebands. Then, a multiple linear regression model was established to predict the soil salt content based on the measured spectral data, and the accuracy of the model was verified using field-measured salinity data. The results showed that the overall shapes of the spectral curves of soils with different degrees of salinity were consistent, and the reflectance in visible and near-infrared bands for salinized soil was higher than that for non-salinized soil. After differential transformation, the correlation coefficient between the spectral reflectance and soil salt content was obviously improved. The first-order differential transformation model based on the logarithm of the reciprocal of saline soil spectral reflectance produced the highest accuracy and stability in the bands at 462 and 636 nm; the determination coefficient was 0.603, and the root mean square error was 5.407. Thus, the proposed model provides a good reference for the quantitative extraction and monitoring of regional soil salinization.

Predicting Surface Roughness and Moisture of Bare Soils Using Multi- band Spectral Reflectance Under Field Conditions

Soil surface roughness, denoted by the root mean square height(RMSH), and soil moisture(SM) are critical factors that affect the accuracy of quantitative remote sensing research due to their combined influence on spectral reflectance(SR). In regards to this issue, three SM levels and four RMSH levels were artificially designed in this study; a total of 12 plots was used, each plot had a size of 3 m × 3 m. Eight spectral observations were conducted from 14 to 30 October 2017 to investigate the correlation between RMSH, SM, and SR. On this basis, 6 commonly used bands of optical satellite sensors were selected in this study, which are red(675 nm), green(555 nm), blue(485 nm), near infrared(845 nm), shortwave infrared 1(1600 nm), and shortwave infrared 2(2200 nm). A negative correlation was found between SR and RMSH, and between SR and SM. The bands with higher coefficient of determination R^2 values were selected for stepwise multiple nonlinear regression analysis. Four characterized bands(i.e., blue, green, near infrared, and shortwave infrared 2) were chosen as the independent variables to estimate SM with R^2 and root mean square error(RMSE) values equal to 0.62 and 2.6%, respectively. Similarly, the four bands(green, red, near infrared, and shortwave infrared 1) were used to estimate RMSH with R^2 and RMSE values equal to 0.48 and 0.69 cm, respectively. These results indicate that the method used is not only suitable for estimating SM but can also be extended to the prediction of RMSH. Finally, the evaluation approach presented in this paper highly restores the real situation of the natural farmland surface on the one hand, and obtains high precision values of SM and RMSH on the other. The method can be further applied to the prediction of farmland SM and RMSH based on satellite and unmanned aerial vehicle(UAV) optical imagery.

Retrieval of Winter Wheat Canopy Carotenoid Content with Ground-and Airborne-Based Hyperspectral Data

Accurate assessment of canopy carotenoid content(CC_(x+c)C)in crops is central to monitor physiological conditions in plants and vegetation stress,and consequently supporting agronomic decisions.However,due to the overlap of absorption peaks of carotenoid(C_(x+c))and chlorophyll(C_(a+b)),accurate estimation of carotenoid using reflectance where carotenoid absorb is challenging.The objective of present study was to assess CC_(x+c)C in winter wheat(Triticum aestivum L.)with ground-and aircraft-based hyperspectral measurements in the visible and near-infrared spectrum.In-situ hyperspectral reflectance were measured and airborne hyperspectral data were acquired during major growth stages of winter wheat in five consecutive field experiments.At the canopy level,a remarkable linear relationship(R^(2)=0.95,p<0.001)existed between C_(x+c) and Ca+b,and correlation between CC_(x+c)C and wavelengths within 400 to 1000 nm range indicated that CC_(x+c)C could be estimated using reflectance ranging from visible to near-infrared wavebands.Results of Cx+c assessment based on chlorophyll and carotenoid indices showed that red edge chlorophyll index(CI red edge)performed with the highest accuracy(R^(2)=0.77,RMSE=22.27μg/cm^(2),MAE=4.97μg/cm^(2)).Applying partial least square regression(PLSR)in CC_(x+c)C retrieval emphasized the significance of reflectance within 700 to 750 nm range in CC_(x+c)C assessment.Based on CI red edge index,use of airborne hyperspectral imagery achieved satisfactory results in mapping the spatial distribution of CC_(x+c)C.This study demonstrates that it is feasible to accurately assess CC_(x+c)C in winter wheat with red edge chlorophyll index provided that C_(x+c) correlated well with C_(a+b) at the canopy scale.it is therefore a promising method for CC_(x+c)C retrieval at regional scale from aerial hyperspectral imagery.

Spatial-Aware Supervised Learning for Hyper-Spectral Image Classification Comprehensive Assessment

A comprehensive assessment of the spatial.aware mpervised learning algorithms for hyper.spectral image （HSI） classification was presented. For this purpose, standard support vector machines （ SVMs ）, mudttnomial logistic regression （ MLR ） and sparse representation （SR） based supervised learning algorithm were compared both theoretically and experimentally. Performance of the discussed techniques was evaluated in terms of overall accuracy, average accuracy, kappa statistic coefficients, and sparsity of the solutions. Execution time, the computational burden, and the capability of the methods were investigated by using probabilistie analysis. For validating the accuracy a classical benchmark AVIRIS Indian pines data set was used. Experiments show that integrating spectral.spatial context can further improve the accuracy, reduce the misclassltication error although the cost of computational time will be increased.

优化光谱指数结合PLSR的多金属矿区土壤As含量高光谱反演

砷(As)是我国多金属矿区的主要污染物之一,对环境、农业和人类健康构成严重威胁。近地高光谱技术具有快速、动态、无损、光谱分辨率高等优势,对于多金属矿区土壤As污染监测与综合治理具有巨大应用潜力。然而,由于受污染区域、土壤背景以及高光谱质量、光谱输入量等因素影响,高光谱反演模型的适用性和精度差异较大。本研究针对湘南某多金属矿区,基于Pearson相关性分析并结合变量投影重要性(VIP)准则,提取18种变换光谱形式下的单变量特征波段及4种光谱指数算法下的优化光谱指数作为光谱输入量,建立偏最小二乘回归(PLSR)模型,实现了矿区土壤As含量反演。结果表明:倒数(RT)、对数(L)、平方根(Sqrt)、标准正态变量变换二阶导(SNV_SD)等变换后的光谱数据与As含量具有较高的相关性;优化光谱指数能从二维光谱空间揭示As的光谱响应,相较于单变量特征波段,以优化光谱指数为自变量构建的模型性能更优;比值指数(RI)模型的R_(c)^(2)、RMSE_(c)、R_(p)^(2)、RMSE_(p)、RPD分别为0.908、50.8 mg/kg、0.949、35.6 mg/kg、4.45,是研究区土壤As含量反演的最优模型。单变量特征波段结合优化光谱指数预测土壤As含量具有较好的可行性,可为多金属矿区土壤As污染高光谱快速监测提供科学依据。

大豆冠层叶片氮含量检测研究——基于无人机多光谱图像

为快速获取大豆冠层叶片氮素含量(Leaf Nitrogen Content,LNC)信息,采用无人机获取大豆冠层LNC多光谱影像光谱特征,通过分析光谱变量与LNC的相关性,选出对大豆冠层LNC敏感的光谱变量。利用逐步回归分析方法建立黑河43、龙垦310、龙垦3401在3个关键生育时期(R1、R3、R5)大豆LNC估测模型。研究结果表明:①在3个品种的3个生育期,除R5时期龙垦3401品种外,NDVI与LNC具有高度相关性,说明NDVI可以较好地进行大豆冠层LNC的反演。②在建模的过程中发现,在R1时期龙垦3401、黑河43、龙垦310所建模型的R2和RMSE依次为0.857、0.133,0.845、0.156,0.821、0.187;在R3时期龙垦3401、黑河43、龙垦310所建模型的R2和RMSE依次为0.835、0.204,0.881、0.113,0.849、0.162;在R5时期龙垦3401、黑河43、龙垦310所建模型的R2和RMSE依次为0.835、0.208,0.814、0.215,0.836、0.211。由此表明,利用无人机多光谱遥感图像数据可以很好地监测大豆LNC的空间分布情况。

基于加速度信号的轮胎滚动阻力估计算法

提出了一种基于智能轮胎的滚动阻力估计算法,利用轮胎加速度信号实现轮胎滚动阻力更高精度的估算。使用ABAQUS软件建立205/55/R16子午线轮胎的稳态滚动阻力模型,通过对不同负载、胎压和车速下滚动阻力的仿真结果分析,验证了有限元模型的一致性。通过提取轮胎内衬中心轴线处的加速度信号,使用Yule-Walker分析法计算加速度信号的功率谱密度。基于汽车行驶参数和轮胎接地离地瞬间的加速度功率谱密度数据,采用偏最小二乘回归法对轮胎的滚动阻力和滚阻系数进行回归预测。结果表明:结合轮胎加速度信号与行驶参数的回归方程拟合效果比单纯使用行驶参数拟合的效果更好,研究结果为开发节能汽车提供了一定的指导作用。

基于分数阶微分的葡萄叶片SPAD值高光谱遥感反演研究

【目的】探明高光谱遥感技术反演葡萄叶片叶绿素含量的可能性,构建葡萄叶片叶绿素含量反演模型,为快速且无损估测葡萄长势提供技术参考。【方法】以西南山区成熟期葡萄叶片为研究对象,同步获取冠层叶片高光谱数据和SPAD值,研究不同分数阶(0.0~1.4阶,步长0.2阶)微分光谱反演葡萄叶片SPAD值的能力,构建多个基于特征波段和光谱指数的单因素模型及基于连续投影算法的多因素模型。【结果】不同SPAD值葡萄叶片原始光谱曲线整体一致,在可见光区域反射率较低而在近红外区域反射率高;可见光、近红外区域反射率与SPAD值分别呈反比和正比;随着分数阶上升,特征波段由近红外向红边靠近,光谱指数由近红外与蓝光组合变更为近红外与绿光组合,单因素模型建模变量相关性呈先升后降趋势,在0.6阶达峰值;除0.6与0.8阶外,其余分数阶微分光谱单因素模型建模变量均为DSI;多因素模型优于单因素模型,机器学习算法可提升传统回归模型精度,所有模型以0.6阶下SPA⁃GA⁃XGBoost回归模型精度最优,其建模与验证R^(2)分别为0.79和0.75,相应均方根误差(nRMSE)分别为15.54%和14.45%。【结论】分数阶微分变换在葡萄叶片SPAD值反演方面具有较大潜力,特定分数阶下,光谱指数优于特征波段,GA⁃XGBoost算法能产生较好的建模效果。

基于高光谱的不同生育期玉米花青素含量估测

花青素(Anthocyanin)是玉米体内的重要色素,对花青素含量的便捷、无损估测对监测玉米长势具有重要意义。利用关中地区拔节期、大喇叭口期、抽雄期以及乳熟期玉米冠层叶片Anth值及高光谱数据建立多个单因素模型和多因素模型。结果表明,不同生育期玉米叶片原始光谱特征总体一致、局部不同。变换光谱的特征波段与Anth值相关性优于原始光谱,其中一阶导数光谱特征波段最优。连续投影算法(SPA)降维能力较好,筛选出的建模参数在2~27个。最优单因素模型与多元性线性回归模型精度均为抽雄期最优,拔节期和大喇叭口期次之,乳熟期最差。所有模型中,抽雄期基于一阶导数光谱的麻雀搜索算法-极限学习机回归(SSA-ELMR)模型精度最佳,该模型建模与验证R2分别为0.847、0.895,相应nRMSE为6.44%和7.21%。本研究结果表明抽雄期是玉米叶片花青素含量反演的最佳时期,极限学习机能进一步提升传统模型精度。

影响能谱CTU虚拟平扫尿路结石检出的因素分析

目的 基于Logistic回归分析评估影响能谱CT尿路造影(CTU)虚拟平扫图像检出尿路结石的因素。方法 回顾性纳入150例尿路结石并行能谱CTU的患者(记录所用碘对比剂)。将平扫图像重建为120 kVp-like图像,静脉期和排泄期的增强图像采用碘去除技术分别重建为静脉期和排泄期的虚拟平扫图像。2位医师独立评估以上3组图像,并记录3组图像的尿路结石数量、所在解剖位置及虚拟平扫图像是否有碘残留;结石大小和CT值仅在真实平扫图像上测量。结石大小、CT值、结石所在位置和所用碘对比剂纳入Logistic回归分析,用于评估影响虚拟平扫尿路结石检出的因素。受试者工作特征(ROC)曲线用于绘制各指标的曲线下面积(AUC)值、诊断灵敏度和特异度及最佳临界值。结果 真实平扫上检出304枚结石,而静脉期和排泄期虚拟平扫结石检出率分别为92.4%和71.4%。结石大小和CT值是影响静脉期和排泄期虚拟平扫结石检出的独立风险因素(P<0.01),结石大小和CT值在静脉期虚拟平扫结石检出的AUC值达0.96以上,诊断临界值分别是3.52 mm和469 HU,而在排泄期虚拟平扫结石检出的结石大小、CT值及解剖位置等指标的综合AUC值降为0.88。排泄期虚拟平扫结石检出率在碘对比剂组间无统计学差异(P=0.57)。另外,排泄期肾盂肾盏处虚拟平扫结石检出率明显降低(P<0.001)。结论 在增强CT的2个扫描期相中,静脉期虚拟平扫结石检出效果更佳。结石大小和CT值是影响虚拟平扫结石检出的重要因素。肾盂肾盏处排泄期虚拟平扫结石检出率低与碘去除效果欠佳相关。

基于无人机高光谱影像的冬小麦叶片氮浓度遥感估测

叶片氮浓度(LNC)是反应作物光合作用、营养状况和长势的重要指标,为精准高效地估测不同生育期冬小麦叶片氮浓度,以新冬22为研究对象,利用无人机搭载Pika L高光谱相机获取4个关键生育期冬小麦冠层反射率数据。基于波段优化算法和相关性分析筛选LNC敏感光谱指数,结合逐步回归、多元线性回归和偏最小二乘回归建立关键生育期冬小麦叶片氮浓度估测模型,并与单变量估测模型进行比较。结果表明:基于波段优化算法筛选的组合光谱指数与LNC的相关性优于传统植被指数,且达到极显著性相关;在单变量LNC估测模型中,组合光谱指数构建的模型精度优于传统植被指数,其中,扬花期差值光谱指数(DSI(R940、R968))建立的估测模型最好,R2为0.789;多变量估测模型精度均优于单变量估测模型,其中,基于偏最小二乘回归构建的LNC估算模型最好,孕穗期和扬花期拟合效果较优,模型决定系数均为0.923,均方根误差为0.082、0.084。本研究结果可以作为冬小麦LNC估测和长势监测的科学依据。

基于多光谱图像角度融合测定库尔勒香梨理化指标

基于多光谱图像角度融合,提出了一种库尔勒香梨快速无损检测的方法。以库尔勒香梨为研究对象,应用多光谱成像采集系统,以10°为间隔,获取了相对方位角为10°~90°内的多光谱图像。使用ENVI5.1软件进行多光谱图像角度融合并提取感兴趣区域(ROI),获得多光谱数据。对光谱反射率、波段和相对方位角进行了皮尔逊相关性分析,发现波段和相对方位角都对光谱反射率呈极显著相关性,且相对方位角相关系数为0.1大于波段相关系数0.053,有必要在建模过程中加入相对方位角因素。采用标准正态变量变换(SNV)、均值中心化变换(MC)、卷积平滑处理(S_G)和归一化处理(Nor)等预处理方法,选用偏最小二乘回归(PLSR)建立全波段检测模型,通过校正集相关系数(R_(c))、预测集相关系数(R_(p))、校正集均方根误差(RMSEC)和预测集均方根误差(RMSEP)对模型的效果进行评价,对比探究特征角度下和角度融合下库尔勒香梨关键指标的模型检测效果。结果表明:采用角度融合处理后,所建立的PLSR和SVM模型预测效果都有显著提升。预测库尔勒香梨含水率最优模型为采用角度融合的偏最小二乘回归模型(AF-PLSR),其R_(c)为0.936, RMSEC为0.298,R_(p)为0.901, RMSEP为0.285;预测库尔勒香梨硬度和糖度的最优模型为以角度融合的支持向量机模型(AF-SVM),R_(c)分别为0.894、 0.905, RMSEC为0.527、 0.376;R_(p)为0.830、 0.901, RMSEP为0.532、 0.379。角度融合将不同角度下的光谱数据结合在一起,获得了比单一角度更加丰富的信息,得到了更加完善的光谱。所建立的检测模型精度更高。结果证明:基于多光谱图像角度融合技术预测库尔勒香梨的含水率、硬度和糖度是可行的。为提高多光谱无损检测精度和高光谱无损检测精度提供了一种新的思路。

大气CO_(2)浓度升高背景下冬小麦冠层光谱特征和地上生物量估算

本研究旨在探究大气CO_(2)浓度升高对冬小麦全生育时期冠层光谱特征的影响,并基于筛选的敏感波段建立地上生物量(AGB)与光谱参数的定量关系。为此,在2021—2022年的冬小麦生长季,利用开放式CO_(2)富集系统(Mini-FACE),设定大气CO_(2)浓度(ACO_(2),(420±20)μL L^(–1))和高CO_(2)浓度(ECO_(2),(550±20)μL L^(–1))两个处理水平,分析了高CO_(2)浓度下光谱特征变化,基于连续投影算法(SPA)、逐步多元线性回归(SMLR)和偏最小二乘法回归(PLSR)筛选AGB敏感波段并构建估算模型。结果表明:CO_(2)浓度升高使冬小麦拔节期和开花期AGB显著增加。红边和近红边反射率及红边面积在拔节期增加,在开花期和灌浆期降低,蓝边、黄边和红边位置在不同生育时期均发生移动;AGB的敏感光谱波段主要分布在红边和近红边区域,CO_(2)浓度升高缩小了AGB敏感波段范围,但不影响AGB的估算;AGB的SMLR和PLSR模型均取得了较高的估算精度(R^(2)>0.8),其中SMLR模型中的R_(799′)、D_(y)、SD_(y)和PRI等特征参数与AGB显著相关,R^(2)为0.866。PLSR模型(R^(2)>0.9)在估算精度和稳定性上优于SMLR模型。本研究可为未来高CO_(2)浓度下冬小麦生长发育的遥感监测提供理论基础和技术方法。

基于多目标筛选堆叠回归的光谱反射率重建

物体的光谱反射率完全决定了其物体色,因此研究光谱反射率对于色彩信息要求较高的行业具有重大意义。直接获取光谱反射率需要精密且昂贵的设备,而通过建立模型,由低成本的数码相机等设备获取的RGB响应值去预测光谱反射率,可以大大降低成本。基于回归方法的光谱反射率重建算法受到广泛关注,其核心是建立RGB向量到光谱反射率向量间的映射关系。对于大多数物体而言,其表面的光谱反射率曲线具有平滑属性,因此,光谱反射率分量之间具有一定的相关性。而已有的算法都是对光谱反射率向量每一个维度独立地建立预测模型,没有将光谱反射率分量之间的相关性利用起来。与传统的单输出回归方法不同,多目标堆叠回归方法通过将首次预测输出值重新注入输入端来利用输出端之间的相关性。基于多目标堆叠回归的光谱反射率重建取得了重要的进展;然而,传统的多目标堆叠回归方法存在着易受首次预测输出值误差影响的问题。针对这一问题,提出一种新的多目标堆叠方法,对于首次预测输出值进行筛选,从中选出误差较小的部分作为输入,以此来保证下一步建立的模型精度。该筛选方法可以在不知道真实值的情况下,极大程度地保留误差较低的部分样本。实验数据集来源为ICVL高光谱图像数据库,评价指标为均方根误差与色度误差。实验结果表明,所提出的多目标筛选堆叠回归可以有效克服传统多目标堆叠回归所存在的问题,做到比无堆叠时的误差更小,说明提出的方法可以有效地利用光谱反射率分量之间的相关性。