生成式对抗网络的土壤有机质高光谱估测模型

doi:10.3964/j.issn.1000-0593(2021)06-1905-07

Abstract
References
Related Citation (15)

Download: PDF (1251 KB)
Export: BibTeX | EndNote (RIS)

Abstract In the previous study of the estimation model of soil organic matter content, most models were based on the feature bands, linear and non-linear empirical models rarely explored the ability promotion using an extended modeling dataset. To further improve the performance of the estimation model, it proposed a dynamic estimation model of soil organic matter content using generative adversarial networks (GAN) to generate the pseudo hyperspectral and organic matter content. Paddy soil samples and hyperspectral data (350~2 500 nm) were collected from Changsha and its surrounding areas of Hunan Province, and the organic matter content was monitored chemically. Based on these data, equivalent new samples were generated by GAN and combined with the modeling set to form anenhanced modeling set. After completing each epochformal training of GAN, the prediction models of soil organic matter content were dynamically constructed using cross-validation ridge regression (RCV), partial least squares regression(PLSR) and BP neural network (BPNN) on four observation points (corresponding 50, 100, 150 and 239 generated samples in enhanced modeling set) (the abbreviation of models were GAN-RCV, GAN-PLSR and GAN-BPNN). The experimental results showed that: (1) Among the estimation models fitted on modeling set of the origin data, RCV was the best-performing model, whose determination coefficient (R²) and root square error (RMSE) were 0.831 1 and 0.189 6; (2) In the 150 epochs formal training of GAN, the performance of GAN-RCV, GAN-PLSR and GAN-BPNN dynamically constructed on the enhanced modeling set were significantly improved, specific performances: R² of GAN-RCV obtained the maximum 0.890 9 (RMSE 0.153 7), minimum 0.850 5 (RMSE 0.18) and mean 0.868 7 (RMSE 0.168 6), the maximum R² increased by 7.2% (RMSE decreased by 18.9%) compared with RCV fitted on the modeling dataset, R² of GAN-PLSR had the maximum 0.855 4 (RMSE 0.176 9), minimum 0.727 0 (RMSE 0.243 2) and mean 0.780 1 (RMSE 0.217 7), the maximum R² increased by 20.6% (RMSE decreased by 29.5%) than PLSR constructed on the modeling dataset, GAN-BPNN performed best, whose R² had the maximum 0.905 2(RMSE 0.143 3), minimum 0.801 7(RMSE 0.207 3) and mean 0.868 1(RMSE 0.168 6), the maximum R² increased by 30.8%(RMSE decreased by 44.5%) comparing BPNN fitted on the modeling set; (3) With the increase of the number of generated samples in the enhanced modeling dataset, the improvement effect of model accuracy showed a trend of increasing first and then decreasing, and among the four observation points, the model constructed on the third had the most significant performance improvement. Sufficient experiments showed that the dynamic estimation model based on GAN improved the performance effectively. According to the evaluation results on the test set, the optimum model could be used to predict the soil organic matter content in the follow-up application.

Key words：Soil organic matter; Hyperspectral data; Generative adversarial networks; Ridge cross validation; BP neural network

Received: 2020-06-11 Accepted: 2020-10-03

ZTFLH:

O657.3

Corresponding Authors: XIE Hong-xia E-mail: xiehongxia136@sina.com

Cite this article:

URL:

https://www.gpxygpfx.com/EN/10.3964/j.issn.1000-0593(2021)06-1905-07 OR https://www.gpxygpfx.com/EN/Y2021/V41/I06/1905

[1] XU Xi-bo, LÜ Jian-shu, WU Quan-yuan, et al(徐夕博，吕建树，吴泉源，等). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2018，38(8): 2556.
[2] LI Yang, LIU Xin-lu, PENG Jie, et al(李阳，刘新路，彭杰，等). Chinese Journal of Soil Science(土壤通报)，2018，49(4): 767.
[3] BAO Qing-ling, DING Jian-li, WANG Jing-zhe, et al(包青岭，丁建丽，王敬哲，等). Arid Land Geography(干旱区地理), 2019, 42(6): 1404.
[4] ZHANG Zhi-tao, LAO Cong-cong, WANG Hai-feng, et al (张智韬, 劳聪聪, 王海峰, 等). Transactions of the Chinese Society for Agricultural Machinery(农业机械学报)，2020，51(1): 156.
[5] GUO Jia-xin, ZHAO Xiao-min, GUO Xi, et al(国佳欣，赵小敏，郭熙，等). Acta Pedologica Sinica(土壤学报)，2020, 57(3): 636.
[6] SONG Ke-hui, ZHANG Ying, ZHANG Jiang-wei, et al(宋珂慧，张莹，张江伟，等). Journal of Computer Research and Development(计算机研究与发展)，2019，56(9): 1832.
[7] Goodfellow I, Pouget-abadie J, Mirza M, et al. Generative Adversarial Nets. In: Advances in Neural Information Processing Systems, Springer, Berlin, 2014: 2672.
[8] Radford A, Metz L, Chintala S. Unsupervised Representation Learning With Deep Convolutional Generative Adversarial Networks, 2016, CORR abs/1511.06434.
[9] WANG Wan-liang, LI Zhuo-rong(王万良，李卓蓉). Journal on Communications(通信学报)，2018，39(2): 135.
[10] YE Chen, GUAN Wei(叶晨，关玮). Journal of Tongji University·Natural Science(同济大学学报·自然科学版)，2020，48(4): 591.
[11] WANG Kun-feng, GOU Chao, DUAN Yan-jie, et al(王坤峰, 苟超, 段艳杰, 等). Acta Automatica Sinica(自动化学报), 2017, 43(3): 321.
[12] WU Yi-quan, ZHOU Yang, SHENG Dong-hui, et al(吴一全，周杨，盛东慧，等). J. Infrared Millim. Waves(红外与毫米波学报), 2018, 37(1): 119.
[13] ZHU Ya-xing, YU Lei, HONG Yong-sheng, et al(朱亚星, 于雷, 洪永胜, 等). Scientia Agricultura Sinica(中国农业科学), 2017, 50(22): 4325.
[14] Salimans T, Goodfellow I, Zaremba W, et al. Improved Techniques for Training GANs, International Conference on Neural Information Processing Systems, NIPS’16, 2016, 2234.
[15] GE Xiang-yu, DING Jian-li, WANG Jing-zhe, et al(葛翔宇, 丁建丽, 王敬哲, 等). Acta Optica Sinica(光学学报)，2018，38(10): 1030001.