基于BP神經(jīng)網(wǎng)絡的不同人為干擾強度下鹽漬土定量分析*

田安紅, 付承彪, 熊黑鋼, 趙俊三

田安紅1,2, 付承彪1**, 熊黑鋼3,4, 趙俊三2

(1. 曲靖師范學院信息工程學院 曲靖 655011; 2. 昆明理工大學國土資源工程學院 昆明 650093; 3. 北京聯(lián)合大學應用文理學院 北京 100083; 4. 新疆大學資源與環(huán)境科學學院 烏魯木齊 830046)

傳統(tǒng)對土壤元素的反演模型經(jīng)常采用線性的偏最小二乘模型(PLSR), 例如, 夏芳等[14]以浙江省36個縣市的農(nóng)田土壤為研究對象, 分析有機質(zhì)與8種重金屬的相關性, 并采用PLSR預測8種重金屬的含量, 仿真結果表明PLSR對重金屬Ni和Cr的預測效果較好, 其相對預測性能(RPD)值為1.8~2.0, 而剩余6種重金屬預測模型的RPD值均為1.0~1.4。王文俊等[15]以山西的褐土為研究對象, 利用PLSR對20種高光譜變換后的預處理方法進行建模估算總氮含量, 仿真結果表明一階導數(shù)預處理后建模能得到更好的預測結果, 且最佳的預處理方法為平均光譜曲線與標準差曲線的乘積, 其次為平均光譜曲線與平均光譜曲線的一階導數(shù)、與標準差曲線的乘積, PLSR模型能對總氮進行有效的預測。然而, 土壤高光譜與土壤某元素間的關系表現(xiàn)為非線性, 傳統(tǒng)線性PLSR對土壤元素的反演精度有限, 因此需要探索非線性的預測方法。

1 材料與方法

1.1 土壤樣本采集

研究區(qū)位于新疆維吾爾自治區(qū)昌吉回族自治州境內(nèi), 87°44￠~88°46￠E, 43°29￠~45°45￠N, 距烏魯木齊約70 km。該區(qū)域土壤鹽漬化嚴重, 土壤表層的鹽分含量為5.34~44.45 g×kg-1 [1], 夏季非常炎熱, 降水稀少, 蒸發(fā)強烈, 年蒸發(fā)量高達2 000 mm。

圖1 無人為干擾區(qū)(A)和人為干擾區(qū)(B)鹽漬土采樣點示意圖

藍色方框為水渠位置, 紅色圓圈為農(nóng)場位置, 黃色方框為無人為干擾區(qū)(A區(qū)), 綠色方框為人為干擾區(qū)(B區(qū))。The blue box is the location of the canal, the red circle is the location of the farm, the yellow box is the undisturbed area (area A), and the green box is the human disturbing area (area B).

1.2 野外高光譜測量

55個樣本點的野外高光譜采用FieldSpec?3 Hi- Res高精度地物光譜儀測量, 該儀器的波段范圍300~2 500 nm。350~1 000 nm波段的采樣間隔為1.4 nm, 1 000~2 500 nm波段的采樣間隔為1 nm。野外測量時選擇當?shù)貢r間13:00—15:00, 且晴朗無風的天氣進行。每次測量之前用白板進行光譜校正處理, 每個土壤樣本點采用梅花樁采樣法于5個方向重復采集10次高光譜, 測定高度為距離土壤表面15 cm。計算平均值為該樣點的原始高光譜數(shù)據(jù)。同時, 因邊緣波段(350~390 nm和2 401~2 500 nm)信噪比低及存在水分吸收帶(1 355~1 410 nm和1 820~1 942 nm)的干擾, 刪除這些波段范圍的高光譜數(shù)據(jù)。

1.3 建立BP神經(jīng)網(wǎng)絡模型

1.4 模型精度評價方法

2 結果與分析

2.1 研究區(qū)鹽漬土4種主要陰離子含量統(tǒng)計

表1 無人為干擾區(qū)和人為干擾區(qū)鹽漬土4種陰離子含量描述性統(tǒng)計

2.2 不同含量鹽漬土的高光譜曲線特征

圖2 無人為干擾區(qū)(A區(qū))和人為干擾區(qū)(B區(qū))不同含量鹽漬土壤樣本的高光譜曲線圖

2.3 不同人為干擾區(qū)鹽漬土含量反演特征波段的選擇

因此, 本研究將兩種光譜變換在0階、一階和二階微分中通過0.05檢驗的波段選擇為特征波段, 研究區(qū)通過0.05顯著性檢驗的波段數(shù)量個數(shù)如表2所示, 特征波段對應的高光譜值作為后續(xù)BP神經(jīng)網(wǎng)絡模型的輸入變量。

圖3 無人為干擾區(qū)(A區(qū))和人為干擾區(qū)(B區(qū))鹽漬土高光譜與含量的相關系數(shù)

<|0.05|表示顯著相關。<|0.05| indicates significant correlation.

表2 無人為干擾區(qū)和人為干擾區(qū)通過0.05顯著性檢驗的鹽漬土高光譜波段數(shù)量個數(shù)

R表示原始高光譜, LogR表示對數(shù)變換后的光譜。R is the original hyperspectral, LogR is logarithmic transformation of R.

2.4 不同人為干擾區(qū)鹽漬土含量高光譜反演模型

表3 無人為干擾區(qū)和人為干擾區(qū)鹽漬土含量高光譜反演模型的精度

RPD: 相對預測性能。RPD: relative prediction performance.

圖4 無人為干擾區(qū)(a)和人為干擾區(qū)(b)鹽漬土含量實測值和BP模型預測值的散點圖

圖中預測數(shù)據(jù)為高光譜對數(shù)二階微分(LogR)的BP模型預測值。The predicted values are prediction results of BP model with spectral logarithmic transformation.

圖5 無人為干擾區(qū)(a)和人為干擾區(qū)(b)鹽漬土含量實測值與BP模型預測值的擬合效果

圖中預測數(shù)據(jù)為高光譜對數(shù)二階微分(LogR)的BP模型預測值。The predicted values are prediction results of BP model with spectral Logarithmic transformation.

3 結論

圖6 無人為干擾區(qū)(a)和人為干擾區(qū)(b)鹽漬土含量BP模型的訓練過程

3)統(tǒng)計相關系數(shù)在0階、一階和二階微分中通過0.05檢驗的波段數(shù)量, R變換在無人為干擾區(qū)分別為0個、38個和77個, 在人為干擾區(qū)分別為0個、39個和74個; LogR變換在無人為干擾區(qū)分別為1 822個、264個和121個, 在人為干擾區(qū)分別為1 659個、121個和86個。

4)無人為干擾區(qū)的最佳反演模型為二階微分的LogR光譜變換對應的BP模型, 其RPD為3.309, 表明該模型的預測能力非常強。人為干擾區(qū)的最佳反演模型為一階微分的LogR光譜變換對應的BP模型, 其RPD為2.234, 表明該模型的預測能力很好。

[1] 段鵬程, 熊黑鋼, 李榮榮, 等. 不同干擾程度的鹽漬土與其光譜反射特征定量分析[J]. 光譜學與光譜分析, 2017, 37(2): 571–576 DUAN P C, XIONG H G, LI R R, et al. A quantitative analysis of the reflectance of the saline soil under different disturbance extent[J]. Spectroscopy and Spectral Analysis, 2017, 37(2): 571–576

[2] 高會, 陳紅艷, 劉慧濤, 等. 基于高光譜的魯西北平原土壤有效磷含量快速檢測研究[J]. 中國生態(tài)農(nóng)業(yè)學報, 2013, 21(6): 752–757 GAO H, CHEN H Y, LIU H T, et al. Spontaneous determination of soil available phosphorus using high spectrum in the northwest plain of Shandong Province[J]. Chinese Journal of Eco-Agriculture, 2013, 21(6): 752–757

[3] 張娟娟, 余華, 喬紅波, 等. 基于高光譜特征的土壤有機質(zhì)含量估測研究[J]. 中國生態(tài)農(nóng)業(yè)學報, 2012, 20(5): 566–572 ZHANG J J, YU H, QIAO H B, et al. Soil organic matter content estimation based on hyperspectral properties[J]. Chinese Journal of Eco-Agriculture, 2012, 20(5): 566–572

[4] 蔡亮紅, 丁建麗. 基于高光譜多尺度分解的土壤含水量反演[J]. 激光與光電子學進展, 2018, 55(1): 400–409 CAI L H, DING J L. Inversion of soil moisture content based on hyperspectral multi-scale decomposition[J]. Laser & Optoelectronics Progress, 2018, 55(1): 400–409

[5] MASOUD A A, KOIKE K, ATWIA M G, et al. Mapping soil salinity using spectral mixture analysis of landsat 8 OLI images to identify factors influencing salinization in an arid region[J]. International Journal of Applied Earth Observation and Geoinformation, 2019, 83: 101944

[6] LITALIEN A, ZEEB B. Curing the earth: A review of anthropogenic soil salinization and plant-based strategies for sustainable mitigation[J]. Science of the Total Environment, 2020, 698: 134235

[7] EL HASINI S, IBEN HALIMA O, EL AZZOUZI M, et al. Organic and inorganic remediation of soils affected by salinity in the Sebkha of Sed El Mesjoune — Marrakech (Morocco)[J]. Soil and Tillage Research, 2019, 193: 153–160

[8] LI N, KANG Y H, LI X B, et al. Response of tall fescue to the reclamation of severely saline coastal soil using treated effluent in Bohai Bay[J]. Agricultural Water Management, 2019, 218: 203–210

[9] SCUDIERO E, SKAGGS T H, CORWIN D L. Comparative regional-scale soil salinity assessment with near-ground apparent electrical conductivity and remote sensing canopy reflectance[J]. Ecological Indicators, 2016, 70: 276–284

[10] LI J G, PU L J, HAN M F, et al. Soil salinization research in China: Advances and prospects[J]. Journal of Geographical Sciences, 2014, 24(5): 943–960

[11] 梁靜, 丁建麗, 王敬哲, 等. 基于反射光譜與Landsat 8 OLI多光譜數(shù)據(jù)的艾比湖濕地土壤鹽分估算[J]. 土壤學報, 2019, 56(2): 320–330 LIANG J, DING J L, WANG J Z, et al. Quantitative estimation and mapping of soil salinity in the Ebinur lake wetland based on vis-NIR reflectance and landsat 8 OLI data[J]. Acta Pedologica Sinica, 2019, 56(2): 320–330

[12] WANG R S, WAN S Q, SUN J X, et al. Soil salinity, sodicity and cotton yield parameters under different drip irrigation regimes during saline wasteland reclamation[J]. Agricultural Water Management, 2018, 209: 20–31

[13] PENG J, BISWAS A, JIANG Q S, et al. Estimating soil salinity from remote sensing and terrain data in southern Xinjiang Province, China[J]. Geoderma, 2019, 337: 1309–1319

[14] 夏芳, 彭杰, 王乾龍, 等. 基于省域尺度的農(nóng)田土壤重金屬高光譜預測[J]. 紅外與毫米波學報, 2015, 34(5): 593–598 XIA F, PENG J, WANG Q L, et al. Prediction of heavy metal content in soil of cultivated land: Hyperspectral technology at provincial scale[J]. Journal of Infrared and Millimeter Waves, 2015, 34(5): 593–598

[15] 王文俊, 王璨, 李志偉, 等. 基于高光譜技術的褐土土壤總氮含量的預測[J]. 山西農(nóng)業(yè)大學學報: 自然科學版, 2018, 38(9): 71–76 WANG W J, WANG C, LI Z W, et al. Prediction of total nitrogen content in brown soil based on hyperspectral technology[J]. Journal of Shanxi Agricultural University: Natural Science Edition, 2018, 38(9): 71–76

[16] LIANG Y J, REN C, WANG H Y, et al. Research on soil moisture inversion method based on GA-BP neural network model[J]. International Journal of Remote Sensing, 2019, 40(5/6): 2087–2103

[17] CUI K, QIN X T. Virtual reality research of the dynamic characteristics of soft soil under metro vibration loads based on BP neural networks[J]. Neural Computing and Applications, 2018, 29(5): 1233–1242

[18] ZHAO H J, SHI S G, JIANG H Z, et al. Calibration of AOTF-based 3D measurement system using multiplane model based on phase fringe and BP neural network[J]. Optics Express, 2017, 25(9): 10413-10433

[19] AO H, LIAO X Y, ZHAO D, et al. Delineation of soil contaminant plumes at a co-contaminated site using BP neural networks and geostatistics[J]. Geoderma, 2019, 354: 113878

[20] GOLHANI K, BALASUNDRAM S K, VADAMALAI G, et al. A review of neural networks in plant disease detection using hyperspectral data[J]. Information Processing in Agriculture, 2018, 5(3): 354–371

[21] 劉煥軍, 潘越, 竇欣, 等. 黑土區(qū)田塊尺度土壤有機質(zhì)含量遙感反演模型[J]. 農(nóng)業(yè)工程學報, 2018, 34(1): 127–133 LIU H J, PAN Y, DOU X, et al. Soil organic matter content inversion model with remote sensing image in field scale of blacksoil area[J]. Transactions of the CSAE, 2018, 34(1): 127–133

[22] 喬娟峰, 熊黑鋼, 王小平, 等. 基于最優(yōu)模型的荒地土壤有機質(zhì)含量空間反演[J]. 江蘇農(nóng)業(yè)學報, 2018, 34(1): 68–75 QIAO J F, XIONG H G, WANG X P, et al. Spatial inversion of soil organic matter content in wasteland based on optimal model[J]. Jiangsu Journal of Agricultural Sciences, 2018, 34(1): 68–75

[23] BALDOCK J A, HAWKE B, SANDERMAN J, et al. Predicting contents of carbon and its component fractions in Australian soils from diffuse reflectance mid-infrared spectra[J]. Soil Research, 2013, 51(8): 577

[24] 王海峰, 張智韜, Arnon Karnieli, 等. 基于灰度關聯(lián)-嶺回歸的荒漠土壤有機質(zhì)含量高光譜估算[J]. 農(nóng)業(yè)工程學報, 2018, 34(14): 124–131 WANG H F, ZHANG Z T, KARNIELI A, et al. Hyperspectral estimation of desert soil organic matter content based on gray correlation-ridge regression model[J]. Transactions of the CSAE, 2018, 34(14): 124–131

[25] 張秋霞, 張合兵, 張會娟, 等. 糧食主產(chǎn)區(qū)耕地土壤重金屬高光譜綜合反演模型[J]. 農(nóng)業(yè)機械學報, 2017, 48(3): 148–155 ZHANG Q X, ZHANG H B, ZHANG H J, et al. Hybrid inversion model of heavy metals with hyperspectral reflectance in cultivated soils of main grain producing areas[J]. Transactions of the Chinese Society of Agricultural Machinery, 2017, 48(3): 148–155

[26] 田安紅, 熊黑鋼, 趙俊三, 等. 分數(shù)階微分對鹽漬土野外光譜預處理精度提升的機理分析[J]. 光譜學與光譜分析, 2019, 39(8): 2495-2500 TIAN A H, XIONG H G, ZHAO J S, et al. Mechanism improvement for pretreatment accuracy of field spectra of saline soil using fractional differential algorithm[J]. Spectroscopy and Spectral Analysis, 2019, 39(8): 2495-2500

Tian Anhong1,2, FU Chengbiao1**, XIONG Heigang3,4, ZHAO Junsan2

(1.College of Information Engineering, Qujing Normal University, Qujing 655011, China; 2. Faculty of Land Resource Engineering, Kunming University of Science and Technology, Kunming 650093, China; 3. College of Applied Arts and Science, Beijing Union University, Beijing 100083, China; 4. College of Resource and Environment Sciences, Xinjiang University, Urumqi 830046, China)

S151.9

10.13930/j.cnki.cjea.190700

* 國家自然科學基金項目(41901065, 41671198, 41761081)資助

付承彪, 主要從事遙感與地理信息系統(tǒng)的研究。E-mail: fucb305@163.com

田安紅, 主要從事干旱區(qū)鹽漬土的高光譜研究。E-mail: tianfucb@163.com

2019-09-26

2019-12-10

* This study was supported by the National Natural Science Foundation of China (41901065, 41671198, 41761081).

, E-mail: fucb305@163.com

Dec. 10, 2019

Sep. 26, 2019;