浏览全部资源
扫码关注微信
1.桂林理工大学 计算机科学与工程学院
2.广西嵌入式技术与智能系统重点实验室,广西 桂林 541006
邓昀(1980-),男,桂林理工大学计算机科学与工程学院教授、硕士生导师,研究方向为机器学习与嵌入式、数据分析和深度学习
刘畅(1999-),男,桂林理工大学计算机科学与工程学院硕士研究生,研究方向为深度学习与数据分析。
收稿:2025-03-13,
纸质出版:2026-04-15
移动端阅览
邓昀,刘畅.基于SA-CNN-LSTM的广西土壤速效氮含量预测方法研究[J].软件导刊,2026,25(04):1-11.
DENG Yun,LIU Chang.Research on Prediction Method of Soil Available Nitrogen Content in Guangxi Based on SA-CNN-LSTM[J].Software Guide,2026,25(04):1-11.
邓昀,刘畅.基于SA-CNN-LSTM的广西土壤速效氮含量预测方法研究[J].软件导刊,2026,25(04):1-11. DOI: 10.11907/rjdk.251096.
DENG Yun,LIU Chang.Research on Prediction Method of Soil Available Nitrogen Content in Guangxi Based on SA-CNN-LSTM[J].Software Guide,2026,25(04):1-11. DOI: 10.11907/rjdk.251096.
速效氮是衡量土壤肥力的重要指标,使用高光谱技术准确预测其含量对于林业生长至关重要,但现有光谱一阶微分处理方法缺乏敏感性,且传统的长短期记忆网络(LSTM)预测精度不高。鉴于此,提出一种SA-CNN-LSTM混合模型,并将其与CNN、LSTM和BiLSTM等模型进行对比分析。在广西黄冕和雅长国有林场采集了196个土壤样本,对原始光谱数据进行了28种对数阶微分(LOG)结合小波变换(WT)的预处理,并使用卷积神经网络(CNN)和自注意力机制(SA)以加强LSTM的长距离依赖挖掘精度问题。结果表明,对数阶微分处理优于一阶微分,在对数阶微分的幂为9时建模精度最高,SA-CNN-LSTM模型在验证集上的R²分别比CNN、LSTM和BiLSTM提高了6.79%、7.34%和10.37%,均方根误差(RMSE)降低了18.60%、19.58%和24.44%。使用对数阶微分对光谱进行预处理和SA-CNN-LSTM建模效果最优,验证集R²为0.889,RMSE为16.572 2,PRD为2.998 7,实现了对广西林地土壤速效氮含量的精确预测。
Available nitrogen is an important indicator of soil fertility, and the accurate prediction of its content using hyperspectral technology is crucial to forestry growth. The existing spectral first-order differential processing method lacks sensitivity and the traditional long short-term memory network (LSTM) has low prediction accuracy. Therefore, a SA-CNN-LSTM hybrid model was proposed, and compared with CNN, LSTM and BiLSTM models. In this study, 196 soil samples were collected from Huangmian and Yachang State Forest Farms in Guangxi, and the original spectral data were preprocessed with 28 logarithmic order differentials (LOG) combined with wavelet transform (WT). Convolutional neural network(CNN) and self-attention mechanism (SA) were used to enhance the long-distance dependency mining accuracy of LSTM. The results show that logarithmic order differential processing is better than first order differential, and the modeling accuracy is highest when the power of logarithmic order differential is 9. The R² of SA-CNN-LSTM model on the validation set is 6.79%, 7.34% and 10.37% higher than that of CNN, LSTM and BiLSTM, respectively, and the root mean square error (RMSE) is reduced by 18.60%, 19.58% and 24.44%. The effect of using logarithmic order differential to preprocess the spectrum and SA-CNN-LSTM modeling is the best, with a validation set R² of 0.889, RMSE of 16.572 2 and PRD of 2.998 7, which achieves accurate prediction of available nitrogen content in forest soil in Guangxi.
GUO Y J , LI H P , ZHANG Q B , et al . Response of soil available nitrogen mineralization capacity to temperature and organic fertilizer addition in alpine grasslands of the Qinghai-Tibet Plateau [J]. Grassland Science , 2024 , 41 ( 5 ): 1048 - 1056 .
郭雅婧 , 李辉鹏 , 张其斌 , 等 . 青藏高原高寒草地土壤速效氮矿化能力对温度和有机肥添加的响应 [J]. 草业科学 , 2024 , 41 ( 5 ): 1048 - 1056 .
ZHANG Y , JIAN J . Hyperspectral estimation modeling of soil As content in the eastern part of Tianfu New District, Chengdu based on SPA-BPNN [J]. Journal of Guilin University of Technology , 2024 , 44 ( 1 ): 58 - 66 . doi: 10.3969/j.issn.1674-9057.2024.01.007 http://dx.doi.org/10.3969/j.issn.1674-9057.2024.01.007
张宇 , 简季 . 基于SPA-BPNN的成都市天府新区东部土壤As含量高光谱估测建模 [J]. 桂林理工大学学报 , 2024 , 44 ( 1 ): 58 - 66 . doi: 10.3969/j.issn.1674-9057.2024.01.007 http://dx.doi.org/10.3969/j.issn.1674-9057.2024.01.007
TAN Y S , WEI Z X , XIAO Y , et al . Inversion of organic carbon content in lime soil in Karst area based on hyperspectral and multispectral fusion [J]. Chinese Journal of Applied Ecology , 2025 , 36 ( 1 ): 197 - 207 . doi: 10.13287/j.1001-9332.202501.025 http://dx.doi.org/10.13287/j.1001-9332.202501.025
谭咏诗 , 韦真茜 , 肖雁 , 等 . 基于高光谱和多光谱融合的喀斯特地区石灰土有机碳含量反演 [J]. 应用生态学报 , 2025 , 36 ( 1 ): 197 - 207 . doi: 10.13287/j.1001-9332.202501.025 http://dx.doi.org/10.13287/j.1001-9332.202501.025
LIANG Z Y , CHEN S H , LI N , et al . Spectral estimation of available phosphorus in soils with the same particle size [J]. Chinese Journal of Soil and Fertilizer , 2024 ( 12 ): 230 - 239 .
梁智永 , 陈署晃 , 李娜 , 等 . 不同粒径土壤有效磷光谱估测 [J]. 中国土壤与肥料 , 2024 ( 12 ): 230 - 239 .
BREMNER J , MULVANEY C . Nitrogen total[DB/OL] . https://onlinelibrary.wiley.com https://onlinelibrary.wiley.com . doi: 10.2134/agronmonogr9.2.2ed.c31 http://dx.doi.org/10.2134/agronmonogr9.2.2ed.c31
NOCITA M , STEVENS A , NOON C , et al . Prediction of soil organic carbon for different levels of soil moisture using Vis-NIR spectroscopy [J]. Geoderma , 2013 , 199 : 37 - 42 . doi: 10.1016/j.geoderma.2012.07.020 http://dx.doi.org/10.1016/j.geoderma.2012.07.020
GHOLIZADEH A , BORŮVKA L , SABERIOON M , et al . Visible,near-infrared, and mid-infrared spectroscopy applications for soil assessment with emphasis on soil organic matter content and quality: state-of-the-art and key issues [J]. Applied Spectroscopy , 2013 , 67 ( 12 ): 1349 - 1362 . doi: 10.1366/13-07288 http://dx.doi.org/10.1366/13-07288
ZHONG L , GUO X , GUO J X , et al . Hyperspectral estimation of organic matter in red soil based on different convolutional neural network models [J]. Transactions of the Chinese Society of Agricultural Engineering , 2021 , 37 ( 1 ): 203 - 212 . doi: 10.11975/j.issn.1002-6819.2021.01.025 http://dx.doi.org/10.11975/j.issn.1002-6819.2021.01.025
钟亮 , 郭熙 , 国佳欣 , 等 . 基于不同卷积神经网络模型的红壤有机质高光谱估算 [J]. 农业工程学报 , 2021 , 37 ( 1 ): 203 - 212 . doi: 10.11975/j.issn.1002-6819.2021.01.025 http://dx.doi.org/10.11975/j.issn.1002-6819.2021.01.025
HUANG Z D , CHEN Z L , WANG C , et al . Comparison of different multivariate calibration methods for measuring soil properties using visible/shortwave near infrared spectroscopy combined with machine learning algorithms [J]. Spectroscopy and Spectral Analysis , 2023 , 43 ( 11 ): 3535 - 3540 .
黄招娣 , 陈再良 , 王琛 , 等 . 比较用可见/短波近红外光谱结合机器学习算法测量土壤性质的不同多元校准方法分析 [J]. 光谱学与光谱分析 , 2023 , 43 ( 11 ): 3535 - 3540 .
LI Z X , GAO S , HU D H , et al . Analysis on the influence of space-based payload characteristics, spectral resolution and signal-to-noise ratio on the quantitative inversion of organic matter in black soil [J]. Remote Sensing Technology and Application , 2024 , 39 ( 4 ): 917 - 926 . doi: 10.11873/j.issn.1004-0323.2024.4.0917 http://dx.doi.org/10.11873/j.issn.1004-0323.2024.4.0917
李泽鑫 , 高爽 , 胡登辉 , 等 . 天基载荷特性光谱分辨率与信噪比对黑土地有机质定量反演影响分析 [J]. 遥感技术与应用 , 2024 , 39 ( 4 ): 917 - 926 . doi: 10.11873/j.issn.1004-0323.2024.4.0917 http://dx.doi.org/10.11873/j.issn.1004-0323.2024.4.0917
LIU Z M , LI X Q , CHEN G L , et al . Inversion of total nitrogen content in rubber plantation soil in Yunnan based on hyperspectral [J]. Tropical Agricultural Science & Technology , 2023 , 46 ( 1 ): 35 - 41 . doi: 10.16005/j.cnki.tast.2023.01.009 http://dx.doi.org/10.16005/j.cnki.tast.2023.01.009
刘忠妹 , 黎小清 , 陈桂良 , 等 . 基于高光谱的云南胶园土壤全氮含量反演 [J]. 热带农业科技 , 2023 , 46 ( 1 ): 35 - 41 . doi: 10.16005/j.cnki.tast.2023.01.009 http://dx.doi.org/10.16005/j.cnki.tast.2023.01.009
YIN C Y , BAI Z J , LUO D F , et al . Comparative study on soil total nitrogen content estimation models based on hyperspectral data [J]. Soil and Fertilizer Sciences in China , 2022 ( 1 ): 9 - 15 . doi: 10.11838/sfsc.1673-6257.20573 http://dx.doi.org/10.11838/sfsc.1673-6257.20573
殷彩云 , 白子金 , 罗德芳 , 等 . 基于高光谱数据的土壤全氮含量估测模型对比研究 [J]. 中国土壤与肥料 , 2022 ( 1 ): 9 - 15 . doi: 10.11838/sfsc.1673-6257.20573 http://dx.doi.org/10.11838/sfsc.1673-6257.20573
ZENG J H , DUAN S B , YAO Y M , et al . Research on hyperspectral inversion of soil organic matter by combining fractional differential and outlier identification [J]. China Agricultural Information , 2023 , 35 ( 4 ): 11 - 25 .
曾佳辉 , 段四波 , 姚艳敏 , 等 . 结合分数阶微分和异常值识别的土壤有机质高光谱反演研究 [J]. 中国农业信息 , 2023 , 35 ( 4 ): 11 - 25 .
PAN Z Y , WU R J , LIN Y F , et al . Hyperspectral remote sensing image classification based on improved residual 3D-CNN and neighbor attention [J]. Remote Sensing of Natural Resources , 2025 ( 2 ): 1 - 12 .
潘增滢 , 吴瑞姣 , 林易丰 , 等 . 改进的残差式3D-CNN和近邻注意力的高光谱遥感图像分类 [J]. 自然资源遥感 , 2025 ( 2 ): 1 - 12 .
ZHANG D Y , WU T J , LUO J C , et al . Remote sensing fine mapping of agricultural planting structure based on spatiotemporal coordination [J]. National Remote Sensing Bulletin , 2024 , 28 ( 8 ): 2014 - 2029 . doi: 10.11834/jrs.20241349 http://dx.doi.org/10.11834/jrs.20241349
张冬韵 , 吴田军 , 骆剑承 , 等 . 时空协同的农业种植结构遥感精细制图 [J]. 遥感学报 , 2024 , 28 ( 8 ): 2014 - 2029 . doi: 10.11834/jrs.20241349 http://dx.doi.org/10.11834/jrs.20241349
VASWANI A , SHAZEER N , PARMAR N , et al . Attention is all you need [C]// NIPS'17: Proceedings of the 31st International Conference on Neural Information Processing Systems , 2017 : 6000 - 6010 .
ZHAO Y T , LI Z , LIU X L , et al . Load identification method integrating convolutional neural network and attention mechanism [J]. Power Engineering Technology , 2025 , 44 ( 1 ): 227 - 235 . doi: 10.12158/j.2096-3203.2025.01.023 http://dx.doi.org/10.12158/j.2096-3203.2025.01.023
赵毅涛 , 李钊 , 刘兴龙 , 等 . 融合卷积神经网络和注意力机制的负荷识别方法 [J]. 电力工程技术 , 2025 , 44 ( 1 ): 227 - 235 . doi: 10.12158/j.2096-3203.2025.01.023 http://dx.doi.org/10.12158/j.2096-3203.2025.01.023
REN D Z , ZHANG Q H . Rolling bearing fault diagnosis method based on improved one-dimensional convolutional neural network [J]. Henan Science and Technology , 2024 , 51 ( 10 ): 20 - 26 .
任德珍 , 张清华 . 基于改进一维卷积神经网络的滚动轴承故障诊断方法 [J]. 河南科技 , 2024 , 51 ( 10 ): 20 - 26 .
LIANG Y Q , BAI S C , ZHANG Z Y . Research progress of neural networks combined with Hamiltonian mechanics in deep learning [J]. Computer Engineering and Applications , 2025 , 61 ( 14 ): 20 - 36 . doi: 10.3778/j.issn.1002-8331.2409-0436 http://dx.doi.org/10.3778/j.issn.1002-8331.2409-0436
梁永琦 , 白双成 , 张志一 . 深度学习中结合哈密顿力学的神经网络研究进展 [J]. 计算机工程与应用 , 2025 , 61 ( 14 ): 20 - 36 . doi: 10.3778/j.issn.1002-8331.2409-0436 http://dx.doi.org/10.3778/j.issn.1002-8331.2409-0436
CUI S Q , GAO Y M , HUO J H , et al . Wind speed prediction method based on EMD-ARIMA-LSTM [J]. Science and Technology Innovation and Application , 2025 , 15 ( 5 ): 50 - 53 .
崔胜秋 , 郜伊明 , 霍家豪 , 等 . 基于EMD-ARIMA-LSTM的风速预测方法 [J]. 科技创新与应用 , 2025 , 15 ( 5 ): 50 - 53 .
DENG Y , NIU Z W , FENG Q Y , et al . Improved temporal convolutional network based hyperspectral prediction model for red soil organic matter [J]. Spectroscopy and Spectral Analysis , 2023 , 43 ( 9 ): 2942 - 2951 .
邓昀 , 牛照文 , 冯琦尧 , 等 . 改进时间卷积网络的红壤有机质高光谱预测模型 [J]. 光谱学与光谱分析 , 2023 , 43 ( 9 ): 2942 - 2951 .
WANG T G , LI H Q , KONG S , et al . Analysis of soil salinity characteristics in cotton-growing areas of Tumushuke City, southern Xinjiang [J]. Special Economic Plants and Animals , 2023 , 26 ( 12 ): 19 - 22 .
王潭刚 , 李慧琴 , 孔松 , 等 . 南疆图木舒克市植棉区土壤盐分特征分析 [J]. 特种经济动植物 , 2023 , 26 ( 12 ): 19 - 22 .
SHEN J L , WU L H , LI L S , et al . Effects of land use types on soil organic carbon components and carbon sequestration in a typical Karst mountain small watershed [J]. Journal of Ecology and Environment , 2025 , 34 ( 3 ): 358 - 367 . doi: 10.16258/j.cnki.1674-5906.2025.03.003 http://dx.doi.org/10.16258/j.cnki.1674-5906.2025.03.003
申佳龙 , 吴栎宏 , 李林霜 , 等 . 典型喀斯特山地小流域土地利用类型对土壤有机碳组分及其固碳效应的影响 [J]. 生态环境学报 , 2025 , 34 ( 3 ): 358 - 367 . doi: 10.16258/j.cnki.1674-5906.2025.03.003 http://dx.doi.org/10.16258/j.cnki.1674-5906.2025.03.003
WU Q Y , ZHU Z Y , WU J M , et al . Dataset representativeness measurement and block sampling strategy for the pan-Kennard-Stone algorithm [J]. Chemical Journal of Chinese Universities , 2022 , 43 ( 10 ): 150 - 157 . doi: 10.7503/cjcu20220397 http://dx.doi.org/10.7503/cjcu20220397
武晴滢 , 祝震予 , 吴剑鸣 , 等 . 泛Kennard-Stone算法的数据集代表性度量与分块采样策略 [J]. 高等学校化学学报 , 2022 , 43 ( 10 ): 150 - 157 . doi: 10.7503/cjcu20220397 http://dx.doi.org/10.7503/cjcu20220397
ZHANG Y L , WANG H , XIAO J J , et al . Estimation of soil organic matter content in mountainous cultivated land based on hyperspectral [J]. Jiangsu Journal of Agricultural Sciences , 2024 , 40 ( 1 ): 112 - 120 .
张永亮 , 汪泓 , 肖玖军 , 等 . 基于高光谱的山区耕地土壤有机质含量估测 [J]. 江苏农业学报 , 2024 , 40 ( 1 ): 112 - 120 .
HU W , HUANG Y , WEI L , et al . Deep convolutional neural networks for hyperspectral image classification [J]. Journal of Sensors , 2015 ( 2 ): 1 - 12 . doi: 10.1155/2015/258619 http://dx.doi.org/10.1155/2015/258619
DONOHO D L , JOHNSTONE I M . Ideal spatial adaptation by wavelet shrinkage [J]. Biometrika , 1994 , 81 ( 3 ): 425 - 455 . doi: 10.1093/biomet/81.3.425 http://dx.doi.org/10.1093/biomet/81.3.425
0
浏览量
17
下载量
0
CSCD
关联资源
相关文章
相关作者
相关机构
鄂公网安备42010602004949号