Abstract: Multi-temporal remote sensing data in large-scale crop phenology identification and classification have become increasingly utilized, particularly for precision management in arid oasis agricultural regions with complex cropping systems. In this study, we developed a deep learning framework integrating Sentinel-2 multi-temporal imagery and normalized difference vegetation index (NDVI) time series for mapping cotton, winter jujube, and tiger nut crops in Tumushuke City, Xinjiang Uygur Autonomous Region, China. We employed the minimum redundancy maximum relevance (mRMR) algorithm for spectral and vegetation index feature selection, followed by Savitzky-Golay (S-G) filtering and double logistic function fitting, to automatically extract the key phenological parameters (start of season (SOS), peak of season (POS), and end of season (EOS)), significantly improving phenological feature extraction accuracy. By incorporating multi-temporal Sentinel-2 data and a multi-scale feature fusion approach, we could systematically compare five classification models (multi-layer perceptron (MLP), residual network-18 (ResNet-18), convolutional long short-term memory (ConvLSTM), Transformer, and random forest classifier (RFC)), demonstrating that high-resolution spatial details substantially enhance crop boundary delineation and classification consistency in complex environments. Further optimization of Transformer’s spatial representation through multi-scale window analysis revealed that the use of 1×1+3×3+5×5 convolutional windows achieves an optimal balance between accuracy and computational efficiency. Independent validation on unseen areas confirmed robust model transferability, with F1 scores of 94.37%, 87.75%, and 86.35% for the three crops (winter jujube, cotton, and tiger nut), respectively. This study validates the high-precision identification potential of Sentinel-2 temporal data and deep neural networks for multi-crop environments, enabling the precise spatial mapping of crop distributions and providing methodological support for smart agricultural decision-making in arid oasis regions.

基于深度学习的干旱绿洲地区Sentinel-2时间序列物候提取与作物分类

[1]CaoJ, MaSP, YuanWH, et al., 2022. Characteristics of diurnal variations of warm-season precipitation over Xinjiang Province in China. Atmos Oceanic Sci Lett, 15(2):100113.

[2]ChaudharyA, KolheS, KamalR, 2016. An improved random forest classifier for multi-class classification. Inf Process Agric, 3(4):215-222.

[3]ChenXM, DingYL, ZhengXM, et al., 2024. Improved estimation of non-photosynthetic vegetation cover using a novel multispectral slope difference index with soil information, Sentinel-1 data, and machine learning. Ecol Inform, 84:102930.

[4]ChengJR, LiH, LiDB, et al., 2022. A survey on image semantic segmentation using deep learning techniques. Comput Mater Contin, 74(1):1941-1957.

[5]de la GuardiaL, de MirandaJH, dos Santos LucianoAC, 2024. Assessment of irrigation water use for dry beans in center pivots using ERA5 Land climate variables and Sentinel 2 NDVI time series in the Brazilian Cerrado. Agric Water Manag, 305:109128.

[6]DuRQ, XiangYZ, ChenJY, et al., 2024. The daily soil water content monitoring of cropland in irrigation area using Sentinel-2/3 spatio-temporal fusion and machine learning. Int J Appl Earth Obs Geoinf, 132:104081.

[7]FaqeerzadaMA, KimH, KimMS, et al., 2025. Hyperspectral imaging VIS-NIR and SWIR fusion for improved drought-stress identification of strawberry plants. Comput Electron Agric, 237:110702.

[8]FarboA, SarviaF, de PetrisS, et al., 2024. Forecasting corn NDVI through AI-based approaches using Sentinel 2 image time series. ISPRS J Photogramm Remote Sens, 211:244-261.

[9]FernRR, FoxleyEA, BrunoA, et al., 2018. Suitability of NDVI and OSAVI as estimators of green biomass and coverage in a semi-arid rangeland. Ecol Indic, 94(Pt 1):16-21.

[10]GámezAL, SegarraJ, VatterT, et al., 2025. Alfalfa yield estimation using the combination of Sentinel-2 and meteorological data. Field Crops Res, 326:109857.

[11]GaoXY, ChiH, HuangJL, et al., 2024. Comparison of cloud-mask algorithms and machine-learning methods using Sentinel-2 imagery for mapping paddy rice in Jianghan Plain. Remote Sens, 16(7):1305.

[12]GaoY, SunZZ, HuD, et al., 2025. GMOPNet: a GAN-MLP two-stage network for optical properties measurement of kiwifruit and peaches with spatial frequency domain imaging. Food Chem, 465(Pt 1):141944.

[13]GhilardiF, de PetrisS, TortiV, et al., 2025. A possible role of NDVI time series from Landsat Mission to characterize lemurs habitats degradation in Madagascar. Sci Total Environ, 974:179243.

[14]GuerriMF, DistanteC, SpagnoloP, et al., 2025. Boosting hyperspectral image classification with Gate-Shift-Fuse mechanisms in a novel CNN-Transformer approach. Comput Electron Agric, 237:110489.

[15]HanW, ChenSH, XiaoSL, et al., 2025. Large-scale tobacco identification via a very-high-resolution unmanned aerial vehicle benchmark and a ConvFlow Transformer. Int J Appl Earth Obs Geoinf, 139:104549.

[16]HaseebM, TahirZ, MahmoodSA, et al., 2025. Winter wheat yield prediction using linear and nonlinear machine learning algorithms based on climatological and remote sensing data. Inf Process Agric, 12(4):431-444.

[17]HeJQ, JiaYL, LiY, et al., 2025. Regional-scale precision mapping of cotton suitability using UAV and satellite data in arid environments. Agric Water Manag, 307:109215.

[18]HuangXZ, WangH, LiXB, 2024. A multi-scale semantic feature fusion method for remote sensing crop classification. Comput Electron Agric, 224:109185.

[19]IsmailiM, KrimissaS, NamousM, et al., 2024. Mapping soil suitability using phenological information derived from MODIS time series data in a semi-arid region: a case study of Khouribga, Morocco. Heliyon, 10(2):e24101.

[20]JiYS, LiuZH, LiuR, et al., 2024. High-throughput phenotypic traits estimation of faba bean based on machine learning and drone-based multimodal data. Comput Electron Agric, 227(Pt 2):109584.

[21]JiaTY, ShamseldinAY, LiuTX, et al., 2025. Soil moisture inversion method for semi-arid regions using multi-temporal Sentinel-1 and Sentinel-2 data. J Hydrol Reg, 661:133603.

[22]JiangDG, ShiXY, LiangY, et al., 2024. Feature extraction technique based on Shapley value method and improved mRMR algorithm. Measurement, 237:115190.

[23]KhankeshizadehE, TahermaneshS, MohsenifarA, et al., 2024. FBA-DPAttResU-Net: forest burned area detection using a novel end-to-end dual-path attention residual-based U-Net from post-fire Sentinel-1 and Sentinel-2 images. Ecol Indic, 167:112589.

[24]KordiF, YousefiH, 2022. Crop classification based on phenology information by using time series of optical and synthetic-aperture radar images. Remote Sens Appl Soc Environ, 27:100812.

[25]LiCL, GuoSL, CuiZ, et al., 2025. Flow simulation based on MISO and LSTM models in the Yangtze River-Dongting Lake System. J Hydrol Reg Stud, 60:102469.

[26]LiGG, ZhangH, LyuT, et al., 2024. Regional significant wave height forecast in the East China Sea based on the Self-Attention ConvLSTM with SWAN model. Ocean Eng, 312(Pt 1):119064.

[27]LiWC, ChenR, MaD, et al., 2023. Tracking autumn photosynthetic phenology on Tibetan plateau grassland with the green–red vegetation index. Agric For Meteorol, 339:109573.

[28]LiuY, YangFQ, YueJB, et al., 2024. Crop canopy volume weighted by color parameters from UAV-based RGB imagery to estimate above-ground biomass of potatoes. Comput Electron Agric, 227(Pt 2):109678.

[29]LuDN, XuLL, ZhouJ, et al., 2025. 3DLST: 3D Learnable Supertoken Transformer for LiDAR point cloud scene segmentation. Int J Appl Earth Obs Geoinf, 140:104572.

[30]MarconeA, ImpolloniaG, CrociM, et al., 2024. Estimation of above ground biomass, biophysical and quality parameters of spinach (Spinacia oleracea L.) using Sentinel-2 to support the supply chain. Sci Hortic, 325:112641.

[31]MarinDB, FerrazGAES, SantanaLS, et al., 2021. Detecting coffee leaf rust with UAV-based vegetation indices and decision tree machine learning models. Comput Electron Agric, 190:106476.

[32]MarinoS, 2023. Understanding the spatio-temporal behaviour of the sunflower crop for subfield areas delineation using Sentinel-2 NDVI time-series images in an organic farming system. Heliyon, 9(9):e19507.

[33]Martínez-MovillaA, Rodríguez-SomozaJL, RománM, et al., 2024. Rapid diagnosis of the geospatial distribution of intertidal macroalgae using large-scale UAVs. Ecol Inform, 83:102845.

[34]MedelytėS, RzhanovY, ŠiaulysA, et al., 2025. Evaluating textural descriptors for automated image classification of stony reefs in turbid temperate waters. Ecol Inform, 90:103236.

[35]MendesJ, LimaJ, CostaL, et al., 2025. Impact of hyper-parameter tuning on CNN accuracy in agricultural image classification. Smart Agric Technol, 11:101016.

[36]MishraK, TiwariHL, PooniaV, 2025. An integrated approach of machine learning methods coupled with cellular automaton for monitoring and forecasting of land use and land cover. J Arid Environ, 226:105293.

[37]NaqviSMZA, AwaisM, KhanFS, et al., 2021. Unmanned air vehicle based high resolution imagery for chlorophyll estimation using spectrally modified vegetation indices in vertical hierarchy of citrus grove. Remote Sens Appl Soc Environ, 23:100596.

[38]NeweteSW, AbutalebK, ChirimaGJ, et al., 2024. Phenology-based winter wheat classification for crop growth monitoring using multi-temporal Sentinel-2 satellite data. Egypt J Remote Sens Space Sci, 27(4):695-704.

[39]NiveditaV, BegumSS, AldehimG, et al., 2024. Plastic debris detection along coastal waters using Sentinel-2 satellite data and machine learning techniques. Mar Pollut Bull, 209:117106.

[40]PandeyGK, MittalK, BansalA, et al., 2025. Fire detection with ResNet 18: comparative analysis across different hyperparameters. Procedia Comput Sci, 260:708-716.

[41]PengKY, LiuYM, ZhangKY, et al., 2025. Regional NDVI reconstruction based on tree-ring width of Pinus massoniana Lamb. in the north-south transition zone of China. Dendrochronologia, 92:126373.

[42]QiuZQ, LiuD, YanNX, et al., 2024. Improving the observations of suspended sediment concentrations in rivers from Landsat to Sentinel-2 imagery. Int J Appl Earth Obs Geoinf, 134:104209.

[43]RenK, ZhangDW, WanMJ, et al., 2021. An infrared and visible image fusion method based on improved DenseNet and mRMR-ZCA. Infrared Phys Technol, 115:103707.

[44]RenZQ, TianFM, WangSQ, et al., 2025. Research on maize leaves surface action potential recognition method based on ResNet-18SE. Smart Agric Technol, 10:100819.

[45]SarkarA, MaityPP, RayM, et al., 2023. Inclusion of fractal dimension in four machine learning algorithms improves the prediction accuracy of mean weight diameter of soil. Ecol Inform, 74:101959.

[46]ShojaeezadehSA, ElnasharA, David WeberTK, 2025. A novel fusion of Sentinel-1 and Sentinel-2 with climate data for crop phenology estimation using machine learning. Sci Remote Sens, 11:100227.

[47]TesfayeAA, OsgoodDE, AwekeBG, 2025. Application of a novel vegetation condition index using MODIS EVI for structuring crop index insurance under a smallholder system. Environ Sustain Indic, 26:100696.

[48]TestaS, SoudaniK, BoschettiL, et al., 2018. MODIS-derived EVI, NDVI and WDRVI time series to estimate phenological metrics in French deciduous forests. Int J Appl Earth Obs Geoinf, 64:132-144.

[49]TranKH, ZhangXY, ZhangHK, et al., 2025. A Transformer-based model for detecting land surface phenology from the irregular harmonized Landsat and Sentinel-2 time series across the United States. Remote Sens Environ, 320:114656.

[50]TufailR, TassinariP, TorreggianiD, 2025. Assessing feature extraction, selection, and classification combinations for crop mapping using Sentinel-2 time series: a case study in northern Italy. Remote Sens Appl Soc Environ, 38:101525.

[51]WangC, ZhangXY, WangWJ, et al., 2024. Understanding the potentials of early-season crop type mapping by using Landsat-8, Sentinel-1/2, and GF-1/6 data. Comput Electron Agric, 224:109239.

[52]WangLY, DaiLM, SunL, 2025. ConvLSTM-based spatiotemporal and temporal processing models for chaotic vibration prediction of a microbeam. Commun Nonlinear Sci Numer Simul, 140(Pt 2):108411.

[53]XuLC, SuXY, WangKT, et al., 2025. Enhancing canopy nitrogen estimation in Torreya grandis based on advanced SLIC-EVI and HMT-seCNN methods using hyperspectral UAV data. Comput Electron Agric, 231:109977.

[54]YuDF, RenLR, ChenC, et al., 2025. An AttSDNet model for multi-scale feature perception enhanced remote sensing classification of coastal salt-marsh wetlands. Mar Environ Res, 204:106899.

[55]ZhaiLL, ZanM, YeM, et al., 2025. Time-series forest age estimation in Xinjiang based on forest disturbance and recovery detection. Ecol Indic, 170:113043.

[56]ZhengB, LiuX, WuY, 2022. Evaluation of urban human settlement environment in corps based on the entropy method: a case study of Tumushuke City. Proceedings of the 3rd International Conference on Humanities, Arts, and Social Sciences (HASS 2022), New York, USA, p.108-118.

[57]ZhouJP, GuXH, LiuCL, et al., 2024. A new approach to extract the upright maize straw from Sentinel-2 satellite imagery using new straw indices. Comput Electron Agric, 216:108506.

[58]ZhouZJ, PlauborgF, ThomsenAG, et al., 2017. A RVI/LAI-reference curve to detect N stress and guide N fertigation using combined information from spectral reflectance and leaf area measurements in potato. Eur J Agron, 87:1-7.

Similar articles