Research on Prediction Model of Soil Nitrogen Content Based on Encoder-CNN

Rong-hua JI; Ying-ying ZHAO; Min-zan LI; Li-hua ZHENG

doi:10.3964/j.issn.1000-0593(2022)05-1372-06

Spectroscopy and Spectral Analysis, Volume. 42, Issue 5, 1372(2022)

Research on Prediction Model of Soil Nitrogen Content Based on Encoder-CNN

Rong-hua JI1,*... Ying-ying ZHAO2,2;, Min-zan LI2,2; and Li-hua ZHENG2,2; *; |Show fewer author(s)

Author Affiliations

¹1. Yantai Research Institute of China Agricultural University, Yantai 264670, China

²2. Key Laboratory of Modern Precision Agriculture System Integration Research, Ministry of Education, China Agricultural University, Beijing 100083, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(10)

Fig. 1. Correlation between soil nitrogen content and self-collected spectrum and its differential spectrum
(a): The original spectrum; (b): The first order differential spectrum; (c): The second order differential spectrum

Download full size

View in Article

Fig. 2. The basic structure of auto-encoder

Download full size

View in Article

Fig. 3. The schematic diagram of CNN network structure

Download full size

View in Article

Fig. 4. Changes of evaluation indexes of CNN-3 model
(a): Coefficient of determination; (b): Root-mean-square error; (c): Relative percent deviation

Download full size

View in Article

Fig. 5. The model prediction performance on Heilongjiang data set (900 iterations)

Download full size

View in Article

Table 1. List of strongly correlated wavebands for spectral data

View table

View in Article

Table 1. List of strongly correlated wavebands for spectral data

光谱数据	强相关波段/nm
原始光谱	918.7~1 878.7, 2 047.9~2 378.5
一阶微分光谱	1 356.0~1 448.4, 1 624.4~1 628.5, 1 710.2~1 726.1,1 733~1 742.3, 1 768.5~1 830.9, 1 906.3~1 931.9, 2 149.8~2 274.2, 2 294.3~2 298.4, 2 314.8~2 327.3
二阶微分光谱	1 412.2~1 427.6, 1 726.1~1 740, 1 742.3~1 771, 1 920.4~1 931.9, 1 997.4~2 025.5, 2 128.6~2 374.2, 2 427.5~2 441.2, 2 450.5~2 459.8, 2 473.9~2 478.6

Table 2. Characteristic bands and model input wavelength selection
View table
View in Article
Table 2. Characteristic bands and model input wavelength selection
特征波段/nm 波长间隔/nm 波长数量
1 356.0~1 448.4 1.5 62
1 624.4~1 628.5 2.0 3
1 710.2~1 830.9 2.4 51
2 128.6~2 374.2 3.8 64

Table 3. Spectral reconstruction effect of different automatic encoder structures

View table

View in Article

Table 3. Spectral reconstruction effect of different automatic encoder structures

结构编号	网络名称	神经元个数				R²
结构编号	网络名称	输入层	隐含层1	隐含层2	隐含层3	R²
1	AutoEnc1 AutoEnc2	180	30 60	- -	- -	0.638 0.485
2	AutoEnc3 AutoEnc4 AutoEnc5	180	60 90 120	30 30 30	- - -	0.802 0.857 0.952
3	AutoEnc6 AutoEnc7 AutoEnc8	180	120 120 120	60 60 90	- 30 30	0.910 0.999 0.951

Table 4. The parameters setting of convolution layers

View table

View in Article

Table 4. The parameters setting of convolution layers

网络层	编号	网络结构1				网络结构2
		CNN-1		CNN-2		CNN-1		CNN-2
		数量	尺寸	数量	尺寸	数量	尺寸	数量	尺寸
卷积层 (卷积核)	1	512	7×7	512	7×7	512	7×7	512	7×7
	2	64	3×3	512	1×1	512	1×1	512	1×1
	3	64	3×3	128	3×3	128	3×3	128	3×3
	4	64	3×3	128	1×1	128	1×1	128	1×1
	5	64	1×1	256	1×1	256	1×1	256	1×1
	6	64	1×1	64	3×3	64	3×3	64	3×3
	7	64	1×1	64	1×1	64	1×1	64	1×1
全连接层 (神经元)	1	64	-	64	-	64	-	256	-
全连接层 (神经元)	2	64	-	64	-	64	-	128	-

Table 5. The prediction performance of four models on different datasets(unit of RMSE: g·kg^-1)

View table

View in Article

Table 5. The prediction performance of four models on different datasets(unit of RMSE: g·kg^-1)

网络结构	模型	LUCAS数据集						黑龙江数据集
		训练集			测试集			黑龙江数据集
		R²	RMSE	RPD	R²	RMSE	RPD	R²	RMSE	RPD
1	CNN-1	0.86	0.79	2.62	0.85	0.86	2.54	0.61	6.37	1.61
1	CNN-2	0.89	0.70	2.98	0.88	0.74	2.93	0.70	5.60	1.83
2	CNN-3	0.92	0.59	3.51	0.90	0.68	3.21	0.73	4.81	2.13
2	CNN-4	0.93	0.53	3.84	0.94	0.51	4.05	0.78	5.34	1.92

Tools

Get Citation

Copy Citation Text

Rong-hua JI, Ying-ying ZHAO, Min-zan LI, Li-hua ZHENG. Research on Prediction Model of Soil Nitrogen Content Based on Encoder-CNN[J]. Spectroscopy and Spectral Analysis, 2022, 42(5): 1372

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Research Articles

Received: Apr. 5, 2021

Accepted: --

Published Online: Nov. 10, 2022

The Author Email: JI Rong-hua (jessic1212@cau.edu.cn)

DOI:10.3964/j.issn.1000-0593(2022)05-1372-06

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. List of strongly correlated wavebands for spectral data

Table 1. List of strongly correlated wavebands for spectral data

Table 2. Characteristic bands and model input wavelength selection

Table 2. Characteristic bands and model input wavelength selection

Table 3. Spectral reconstruction effect of different automatic encoder structures

Table 3. Spectral reconstruction effect of different automatic encoder structures

Table 4. The parameters setting of convolution layers

Table 4. The parameters setting of convolution layers

Table 5. The prediction performance of four models on different datasets(unit of RMSE: g·kg-1)

Table 5. The prediction performance of four models on different datasets(unit of RMSE: g·kg-1)

微信扫一扫：分享

Table 5. The prediction performance of four models on different datasets(unit of RMSE: g·kg^-1)

Table 5. The prediction performance of four models on different datasets(unit of RMSE: g·kg^-1)