聂小军, 洪雯雯, GILLAmmara,等.基于CWT-CARS-CNN综合方法的矿区土壤煤源碳质量分数高光谱估测[J].河南理工大学学报（自然科学版）,2024,43(3):91-100.

NIE X J , HONG W W, GILL Ammara,et al.Hyperspectral estimation of coal-derived carbon mass fraction in mine soil based on the CWT-CARS-CNN integrated method[J].Journal of Henan Polytechnic University(Natural Science) ,2024,43(3):91-100.

基于CWT-CARS-CNN综合方法的矿区土壤煤源碳质量分数高光谱估测

聂小军, 洪雯雯, GILLAmmara, 于海洋, 陈晓东

河南理工大学 测绘与国土信息工程学院，河南 焦作 454000

摘要:  目的  目前，尚无可靠方法定量识别土壤中煤源碳，为此，  方法  采集焦作矿区耕地土壤，配制249个不同质量分数的煤源碳土壤样品，利用ASD FieldSpec4获取样品光谱数据，采用连续小波变换（continuous wavelet transform，CWT）-竞争性自适应重加权采样（competitive adaptive reweighted sampling，CARS）-卷积神经网络（convolutional neural networks，CNN）方法估测土壤中煤源碳质量分数，并对比CWT-CARS-CNN方法与传统光谱指数法构建的煤源碳估测模型估测效果，检验CWT-CARS-CNN估测模型的适用性。  结果  结果表明：350~2 500 nm波段内，煤与土壤光谱特征截然不同，含煤土壤样品光谱反射率随煤源碳质量分数增加而降低；CWT方法提高了光谱对土壤煤源碳的敏感性，CARS方法提取的煤源碳敏感波段数量明显增加，基于CWT-CARS-CNN方法的煤源碳质量分数估测模型精度明显高于传统光谱指数法的；以L8分解尺度构建的CWT-CARS-CNN模型精度最高，验证集R₂=0.999 3， RPD =40.308 1。  结论  基于CWT-CARS-CNN的高光谱估测方法能较准确地估测矿区不同土地利用类型下土壤中的煤源碳质量分数，可为“双碳”背景下矿区土壤碳固存与肥力评估提供参考。

关键词:煤源碳;碳固存;高光谱估测;深度学习;矿区土壤

doi:10.16186/j.cnki.1673-9787.2023030004

基金项目:国家自然科学基金资助项目（41977284）；河南省科技攻关项目（222102320032）

收稿日期:2023/03/02

修回日期:2023/05/05

出版日期:2024/05/15

Hyperspectral estimation of coal-derived carbon mass fraction in mine soil based on the CWT-CARS-CNN integrated method

NIE Xiaojun, HONG Wenwen, GILL Ammara, YU Haiyang, CHEN Xiaodong

School of Surveying and Land Information Engineering，Henan Polytechnic University，Jiaozuo 454000，Henan，China

Abstract:  Objectives  There is a shortage of reliable methods to quantitatively identify the coal-derived in soil.    Methods  In this study，soil samples from cultivated lands in Jiaozuo mining area were collected，249 coal-derived source carbon soil samples with different mass fraction were prepared，and spectral data of the samples were obtained by ASD FieldSpec4，the continuous wavelet transform（CWT）-competitive adaptive reweighted sampling（CARS）-convolutional neural network（CNN） method was used to estimate c oal-derived  mass fraction in soil，the estimation effect of coal-derived carbon mass fraction between the CWT-CARS-CNN and traditional spectral index modelswas compared，and the applicability of the CWT-CARS-CNN model was also tested.    Results  The results showed that in the range of 350~2 500 nm，the hyperspectral characteristics between coal and soil were completely different.The spectral reflectance of coal-contained soil samples decreased with increasing coal-derived carbon mass fraction.The CWT method improved the sensitivity of the spectrum to the coal-derived carbon mass fraction in soil，the number of feature waveband of coal-derived carbon mass fraction extracted by the CARS was obviously increased.In general，accuracies of coal-derived carbon mass fraction estimation models based on the CWT-CARS-CNN integrated method were significantly higher than those based on traditional spectral index method.Especially，the CWT-CARS-CNN model constructed with L8 decomposition scale exhibited the highest accuracy，showing R₂=0.999 3 and  RPD =40.308 1 for itsvalidation set.    Conclusions  The study suggests that hyperspectral estimation based on the CWT-CARS-CNN integrated method can accurately estimate the coal-derived carbon mass fraction in soil under different land use types in mining areas，providing reference for accurate assessment of carbon sequestration and fertility in mine soil under the “Double C” background.

Key words:coal-derived carbon;carbon sequestration;hyperspectral estimation;deep learning;mine soil