Soybean Yield Estimation Method Based on Multi-Source Data Fusion

doi:10.12133/j.smartag.SA202512004

Abstract

Abstract:

[Objective] The aim is to develop a high-precision, large-scale soybean yield estimation framework by integrating multi-source remote sensing data, addressing the critical need for accurate and timely crop production monitoring, and to support precision agriculture and food security decision-making at both field and regional scales. [Methods] Based on multi-source data fusion theory, a soybean yield precision prediction method integrating Unmanned Aerial Vehicle (UAV) and Sentinel-2 multi-source data was proposed. Jianshan Farm was chosen as the study area, and a total of 127 Sentinel-2 images and 47 UAV images covering the whole soybean growth period (May-September) in 2022-2023 were collected. According to the image acquisition time, the Savitzky-Golay filter was adopted to construct half-monthly and monthly synthetic images for UAV and Sentinel-2 data, which guaranteed temporal consistency of multi-source data and reduced the influence of cloud contamination. Meanwhile, three multi-scale spatial feature fusion strategies for UAV and satellite imagery were designed, and three improved Convolutional Neural Network models were constructed by coupling the above fusion methods. These models were subsequently trained and evaluated on the constructed time-series multi-feature datasets. [Results and Discussions] The Long Short-Term Memory-based fusion method obtained the optimal accuracy with R² of 0.93 among three spatial feature fusion schemes. Furthermore, the Spatial Attention-Gated Recurrent Unit-Convolutional Neural Network model exhibited the best soybean yield estimation capability, with its R² reaching 0.94. [Conclusions] The UAV-Sentinel-2 multi-source data fusion framework can effectively improve the accuracy of regional soybean yield estimation, and the optimized fusion algorithm and improved model in this study can provide a reliable technical reference for large-area and high-precision crop yield monitoring.

Key words: UAV, Sentinel-2, multi-source data fusion, yield, soybean

CLC Number:

S127

YIN Qiwei, HE Yan, WANG Zongli, RAO Yuan. Soybean Yield Estimation Method Based on Multi-Source Data Fusion[J]. Smart Agriculture, doi: 10.12133/j.smartag.SA202512004.

Figures/Tables 23

Fig. 1

Table 1

Fig. 2

Table 2

Fig. 3

Fig. 4

Fig. 5

Fig. 6

Fig. 7

Fig. 8

Fig. 9

Fig. 10

Fig. 11

Fig. 12

Fig. 13

Fig. 14

Table 3

Table 4

Table 5

Table 6

Table 7

Fig. 15

Table 8

References 32

[1]	MA Y J, WU S R, YANG P, et al. Research progress and challenges in remote-sensing monitoring of oil-crop yield [J]. Smart Agriculture, 2023, 5(3): 1-16.
[2]	SONG M X, BAI B, YANG J T, et al. Robust UAV-based method for peanut plant height estimation using bare-soil invariant constraints[J]. Smart Agriculture, 2025, 7(6): 124-135.
[3]	SHEN F, SUI M M, ZHU Y F, et al. Using BDS MEO and IGSO satellite SNR observations to measure soil moisture fluctuations based on the satellite repeat period[J]. Remote Sensing, 2021, 13(19): 3967.
[4]	BAI X Q, FAN S L, LI R Q, et al. Estimating root zone soil moisture in farmland by integrating multi-source remote sensing data based on the water balance equation[J]. Agricultural Water Management, 2025, 314(C). DOI: 10.1016/j.agwat.2025.109544 .
[5]	LIAO C H, WANG J F, DONG T F, et al. Using spatio-temporal fusion of Landsat-8 and MODIS data to derive phenology, biomass and yield estimates for corn and soybean[J]. Science of the Total Environment, 2019, 650(Pt 2): 1707-1721.
[6]	WANG J S. Summer-maize yield estimation in the Yuncheng Basin based on multi-source remote-sensing data [D]. Taiyuan: Shanxi Agricultural University, 2023.
[7]	YANG S R, LI L, FEI S P, et al. Wheat yield prediction using machine learning method based on UAV remote sensing data[J]. Drones, 2024, 8(7): 284.
[8]	YANG S. Retrieval of potato growth parameters and yield prediction based on multi-source remote-sensing data and deep learning [D]. Lanzhou: Gansu Agricultural University, 2025.
[9]	VAN HOUT M C, KLANKWARTH U B, STÖVER H. Conjugal visitation rights, privileges and standards of provision inside European prisons: a socio-legal study of extant literature[J]. Social Science & Medicine, 2025, 374: 117879.
[10]	ZHOU L F, MENG R, YU X, et al. Improved yield prediction of ratoon rice using unmanned aerial vehicle-based multi-temporal feature method[J]. Rice Science, 2023, 30(3): 247-256.
[11]	HAN W T, PENG X S, ZHANG L Y, et al. Summer maize yield estimation using multi-temporal UAV remote-sensing vegetation indices [J]. Transactions of the Chinese Society of Agricultural Machinery, 2020, 51(1): 148-155.
[12]	LU B H, LI W G, TIAN M, et al. Winter-wheat yield estimation based on multi-temporal GF-1 imagery and machine learning [J]. Transactions of the Chinese Society of Agricultural Machinery, 2025, 56(10): 448-457.
[13]	Zhao Y, Li M L, XU W, et al. A winter-wheat yield prediction model based on UAV multi-temporal and multi-feature data [J]. Journal of Triticeae Crops, 2025, 45(08): 1089-1100.
[14]	BLESSING S, GIERING R. Error covariance structure caused by uncertainties in atmospheric correction for optical sensors[J]. Surveys in Geophysics, 2026. DOI: 10.1007/s10712-025-09921-8 .
[15]	KIM D, JI A Q, JACKSON A L, et al. Inhibition of cellulase activity by liquid hydrolysates from hydrothermally pretreated soybean straw[J]. Frontiers in Chemical Engineering, 2022, 4: 1004240.
[16]	PORTZ G, MOLIN J P, CANATA T F, et al. Integration of ultrasonic and optical sensing systems to assess sugarcane biomass and N-uptake[J]. Precision Agriculture, 2024, 25(1): 83-99.
[17]	TENGOMO S N, AZEFACK MBOUNOU R L, KAMGANG TCHUIFONG A B, et al. Mapping lithological units and alteration minerals in volcanic rocks of Nkondjock, Littoral Region, Cameron using Sentinel-2 remote sensing satellite imagery and comprehensive fieldwork[J]. Journal of African Earth Sciences, 2026, 236: 106003.
[18]	DEBELLA-GILO M. Relative performance of super-resolved Sentinel-2 and Copernicus VHR images in mapping built-up areas and building footprints using deep learning[J]. European Journal of Remote Sensing, 2025, 58(1): 2517381.
[19]	DEVASTHALE A, KARLSSON K G, ANDERSSON S, et al. Difference between WMO climate normal and climatology: insights from a satellite-based global cloud and radiation climate data record[J]. Remote Sensing, 2023, 15(23): 5598.
[20]	HUANG Y X, ZHU T Q, LIANG Z H, et al. Research on intelligent wood species identification method based on multimodal texture-dominated features and deep learning fusion[J]. Plants, 2025, 15(1): 108.
[21]	KONG S L, TAN Q Y, ZHOU K, et al. A machine learning-based approach for broadband sound recognition: a case study on investigating potential fire hazards of grinding machines[J]. Journal of Loss Prevention in the Process Industries, 2026, 100: 105918.
[22]	DUHAIR H T M, IBRAHIM MBIN MAT, ALSAYAYDEH J A J, et al. Progressive transfer learning Unifies Multi-Corpus EEG for robust and scalable Imagined-Speech decoding[J]. Biomedical Signal Processing and Control, 2026, 116: 109475.
[23]	MEI L F, YAN W J, YUEN K V. Semi-supervised structural damage classification with constrained Bayesian nonparametric mixture models and transmissibility-guided knowledge transfer[J]. Mechanical Systems and Signal Processing, 2026, 245: 113806.
[24]	CANELLAS E, NERIN C, VERA P. Beyond direct contact: chemical migration from external printed components of tea packaging into infusions assessed by UPLC-IM-QTOF analysis[J]. Food Packaging and Shelf Life, 2025, 52: 101658.
[25]	YI Z H, WU P, WANG Y R, et al. A multi-feature fusion approach for intelligent facial paralysis evaluation[J]. Biomedical Signal Processing and Control, 2026, 116: 109533.
[26]	LI Z X, LI T J, SUN J, et al. A physically based neural network for flood routing: the Muskingum-Recurrent neural network[J]. Journal of Hydrology, 2026, 673: 135403.
[27]	SHAKIB M S, ALI M E, BHUIYAN N H, et al. MonoGRU: a theoretical and empirical evaluation of a streamlined gated recurrent unit for rainfall forecasting[J]. Science of the Total Environment, 2026, 1012: 181085.
[28]	AO L C, HUANG Z X, LIU Z D, et al. BTFNet: bridging time and frequency with multi-view pyramid for arrhythmia detection[J]. Biomedical Signal Processing and Control, 2026, 116: 109457.
[29]	DARAMOLA O, ADEYEMI O, ADIGUN J, et al. The effect of the period of weed interference on the growth and yield of soybean (Glycine max L. Merrill)[J]. Journal of Agricultural Sciences, Belgrade, 2020, 65(3): 225-237.
[30]	WANG L Q, DAI S X, KANG Z J, et al. A CNN-SA-GRU model with focal loss for fault diagnosis of wind turbine gearboxes[J]. Energies, 2025, 18(14): 3696.
[31]	ISLAM M, BHATTACHARJEE K, BRÄUNING A, et al. Integrating dendrochronology and satellite NDVI to assess climate sensitivity and canopy resilience along a disturbance gradient in tropical moist forests[J]. Science of the Total Environment, 2026, 1029: 181739.
[32]	CANDAN Ç. Polynomial order selection for Savitzky-Golay smoothers via N-fold cross-validation[J]. Signal Processing, 2026, 242: 110443.

Band number	Band	Central wavelength/μm	Bandwidth/nm	Spatial resolution/m
1	Coastal aerosol	0.443	20	60
2	Blue	0.490	65	10
3	Green	0.560	35	10
4	Red	0.665	30	10
5	Vegetation red edge	0.705	15	20
6	Vegetation red edge	0.740	15	20
7	Vegetation red edge	0.783	20	20
8	Near Infrared	0.842	115	10
8A	Vegetation red edge	0.865	20	20
9	Water vapour	0.945	20	60
10	Short-Wave Infrared -Cirrus	1.375	20	60
11	SWIR	1.610	90	20

Texture feature parameters	Abbreviation	Central band/μm
Mean texture	Mean	Describe the brightness and darkness of an image
Homogeneity texture	Hom	Evaluate the local grayscale uniformity of the image
Dissimilarity texture	Dis	Characterize the differences in texture features between pixels
Entropy texture	Ent	Indicate the size of the amount of information
Second Moment texture	Sm	Describe the uniformity and thickness of texture features
Correlation texture	Corr	Predict the main trend of texture

Model	MAE/（kg/hm²）	RMSE/（kg/hm²）	MAPE	R ²
CNN-UAV-15	22.294 1	30.850 5	0.106 6	0.39
CNN-Sentinel-15	26.461 6	32.930 7	0.111 6	0.32
CNN-US-15	21.827 5	28.974 6	0.095 7	0.44
CNN-UAV-30	24.895 8	31.450 9	0.105 8	0.37
CNN-Sentinel-30	27.005 5	34.359 5	0.120 5	0.21
CNN-US-30	22.177 5	29.737 7	0.096 8	0.40

Model	MAE/（kg/hm²）	RMSE/（kg/hm²）	MAPE	R ²
CNN-US-15	21.827 5	28.974 6	0.095 7	0.44
CNN-TCA-15	15.172 2	18.633 1	0.064 3	0.88
CNN-GRU-15	12.526 8	17.254 5	0.055 5	0.90
CNN-LSTM-15	10.093 1	14.137 4	0.045 0	0.93
CNN-US-30	22.177 5	29.737 7	0.096 8	0.40
CNN-TCA-30	21.568 8	27.953 3	0.092 6	0.71
CNN-GRU-30	19.322 2	26.188 7	0.083 2	0.79
CNN-LSTM-30	18.030 7	25.131 2	0.080 7	0.80

Model	MAE/（kg/hm²）	RMSE/（kg/hm²）	MAPE	R ²
SVM-US-CNN	42.304 6	55.776 7	0.191 5	0.38
SVM-TCA-CNN	35.184 3	45.623 3	0.168 3	0.66
SVM-GRU-CNN	31.577 1	38.886 2	0.153 2	0.73
SVM-LSTM-CNN	26.336 9	32.266 6	0.132 6	0.80
SA-US-CNN	10.730 5	14.740 8	0.050 4	0.85
SA-TCA-CNN	9.488 0	13.941 9	0.048 9	0.87
SA-GRU-CNN	8.403 8	12.283 8	0.037 8	0.94
SA-LSTM-CNN	10.219 8	14.569 3	0.0454	0.91
SE-US-CNN	14.983 6	12.230 6	0.104 2	0.70
SE-TCA-CNN	11.693 5	13.983 2	0.0963	0.71
SE-GRU-CNN	11.707 1	12.191 7	0.053 0	0.90
SE-LSTM-CNN	8.520 9	12.609 2	0.036 8	0.92