DTW-kNN and Sentinel-2 NDVI Time Series for Field-Scale Maize Identification and Multi-Crop Classification in an African Smallholder Landscape
Accurate crop mapping provides spatially explicit information for agricultural planning, monitoring and resource allocation, while supporting agribusiness decisions related to supply chains and risk management. However, mapping crops in smallholder landscapes remains difficult because fields are often fragmented, small, and irregularly shaped, with varying planting dates. This study evaluated Sentinel-2 Normalised Difference Vegetation Index (NDVI) time series and Dynamic Time Warping (DTW) for maize identification and multi-crop field-boundary classification in a smallholder farming area. A maize reference trajectory was developed from known maize fields and tested using an independent set of maize fields. The maize reference fields showed low DTW distances (0.106 - 0.281) to the reference trajectory, while independent validation fields produced comparable distances (0.074 - 0.325), indicating that maize followed a recognisable NDVI trajectory despite variable planting dates. DTW distances (0.151 to 0.565) of candidate fields suggested that some fields had maize-like phenological behaviour while others were less similar to the maize reference. Days After Planting (DAP) alignment produced a clear maize curve, with NDVI increasing after planting, peaking between 90 and 120 DAP, and later declining. The approach was then extended to multi-crop classification using maize, soybean, potato and tea reference trajectories. The multi-reference DTW classifier achieved an overall validation accuracy of 80.0%, with maize and tea classified most reliably. When applied to 128 unlabelled fields, most were predicted as maize, although DTW confidence varied. The findings show that Sentinel-2 NDVI time series and DTW provide an interpretable and data-efficient approach for maize detection and field-level crop classification in smallholder systems.
Fine-grained sugarcane varietal classification using high-resolution PRSS-1 satellite imagery: a field-validated dataset and a spatial context-aware deep learning framework
Introduction Precise mapping of sugarcane varieties is crucial for precision agriculture, yield evaluation, protection of breeders’ rights, efficient supply chain management and policy formulation. Previous studies highlighted that remote sensing techniques suffered issues such as low spatial resolution and inadequate field-validated ground truth data. Literature review reported usage of statistical analysis methods and machine learning approaches in classification of sugarcane varieties, while constraining the use of deep learning techniques, primarily due to low quality and insufficient dataset size. Furthermore, the literature highlights the scarcity of high spatial resolution datasets that could be attributed to employment of open-source hyper-spectral and low spatial resolution remote sensing data. To the best of our knowledge, neither any industrial solution exists nor any research study is conducted on sugarcane varietal classification in Pakistan. Methods In this study, we made an endeavor to prepare a high-resolution dataset of Pakistani sugarcane varieties (CPF-253 & SL-284) based on field-validated ground truth dataset using Pakistan Remote Sensing Satellite-1 (PRSS-1) single date imagery. We also included non crop classes like urban, water and road as well. Pan-sharpening is employed to enhance the spatial resolution of multispectral bands and retaining the original spectral characteristics, by aligning the 3 m bands to the 1 m panchromatic grid and then injecting the panchromatic residual. The dataset was prepared based on field-validated ground truth collection duly verified by sugarcane scientist, a farmer and research team to make the approach efficient and reliable. The study performed a random pixel split strategy as well as polygon-based experiments; the polygon split is the primary protocol so that a field is not shared across train and test. We proposed a Spatial Context Aware Transformer that includes the spatial context from neighboring pixels, resulting in an enhanced feature space. Specifically, our proposed model is characterized by a unique pixel feature augmentation approach, thus significantly increasing the size of the spectral feature space and making it possible to account for local neighborhood spectral context. We experimented with benchmark machine learning approaches such as support vector machine and random forest on our dataset to evaluate their potential as well. Moreover, we also evaluated deep learning methods suggested by previous studies like CNN and transformer architectures. Results The experimental results demonstrate that our proposed spatial context-aware transformer performed better than other architectures on the polygon-based test set (accuracy 95.2%, macro F1 0.92), compared with SVM 90.5%, Random Forest 92.5%, 1-D CNN 93.0% and the Tabular Transformer 93.4% (gains of 4.7% and 1.8% in accuracy versus SVM and the Tabular-Transformer). On the same 3x3 pixels, a center-only Transformer scored 94.2%; the 3x3 Transformer 95.2%; and 3x3 MLP, CNN and RF scored 93.48%, 93.75% and 92.96%, showing that the neighborhood pixels spectral context along with attention mechanism account for the classification results. We also conducted a binary classification experiment employing CNN, Tabular Transformer, Centre-pixel only Transformer (1x1) and proposed Transformer (3x3). The results demonstrated that Center-pixel only Transformer achieved highest overall accuracy of 88.59%, followed by Tabular Transformer (87.63%), 1D-CNN (87.10%) and proposed Transformer (3x3) (86.99%). Discussion The feature significance analysis identified the Green Normalized Difference Vegetation Index and the red spectral band as the most discriminative features for sugarcane varieties classification. The study recommends that future research should incorporate time-series, multi-seasonal and multi-region imagery analysis including non-image information such as climate and production environment for highly precise sugarcane varieties classification in real-world conditions. precise sugarcane varieties classification in real-world conditions.
关注点该研究用PRSS-1单期影像与田间验证数据区分两个巴基斯坦甘蔗品种,3×3邻域Transformer在按地块划分测试集上达95.2%精度,但仅限两品种、单区域单时相,外推需谨慎。
Integrating spectral and texture indices from Jilin-1 imagery with soil and fertilization data to predict plot-level sugarcane yield, millable stalk population and Brix
Abstract Purpose The primary objective of this study was to evaluate the potential of high spatial resolution Jilin-1 satellite imagery for plot-level prediction of sugarcane yield, millable stalk population, and Brix, and to assess whether integrating spectral, texture, soil, and fertilization information could improve prediction performance for precision sugarcane management. Methods Jilin-1 satellite imagery acquired at four growth stages, from seedling to maturity, was used to derive vegetation indices (VIs) and texture indices (TIs), including the normalized difference texture index (NDTI), enhanced vegetation texture index (EVTI), and double-difference ratio texture index (DDRTI). Soil chemical properties (SCPs) and fertilization information (FI) were further incorporated with the remotely sensed variables. Machine learning models were developed for plot-level prediction of sugarcane traits across plant cane and first ratoon cane, and texture window size was optimized to improve TI extraction and model performance. Results For yield, the combination of VIs and TIs outperformed VIs alone at the tillering stage (R 2 CV = 0.65, RMSECV = 15.06 t/ha, RPDCV = 1.68). Adding SCPs and FI further improved yield prediction across plant cane and first ratoon cane (R 2 CV = 0.70, RMSECV = 13.84 t/ha, RPDCV = 1.83). Millable stalk population was best predicted at the maturation stage by VIs and Tis, achieving the best performance (R 2 CV = 0.63, RMSECV = 6602 stalks/ha, RPDCV = 1.66). The best Brix model integrated VIs, TIs, SCPs, and FI at the maturation stage (R 2 CV = 0.44, RMSECV = 0.53 °Bx, RPDCV = 1.33). SHAP analysis identified VIs as the dominant features for sugarcane traits prediction. And, DDRTI contributed more than NDTI and EVTI in yield and Brix prediction. Conclusion It is concluded that integrating spectral, texture, soil, and fertilization information from high spatial resolution Jilin-1 imagery is a promising approach for improving plot-level prediction of key sugarcane traits.
Mapping sugarcane sucrose quality dynamics: A systematic review of multi-temporal sensing, data fusion, and Geo-AI frameworks
Mapping sugarcane sucrose quality dynamics: A systematic review of multi-temporal sensing, data fusion, and Geo-AI frameworks。Computers and Electronics in Agriculture
关注点该综述整合多时相遥感与Geo-AI框架用于甘蔗蔗糖品质动态监测,可为作物品质遥感建模选型提供方法参考,但结论依赖既有文献,缺乏原始验证数据。
PHYIELD: Towards multi-scale maize yield estimation with limited ground reference data via integrated crop models and phenology-guided deep learning
提出PHYIELD混合框架,在产量数据稀缺下实现县与田块尺度玉米估产。
关注点该框架仅用2023年伊利诺伊州县级产量数据建模,即向过去20年、周边州和地块尺度迁移,R为0.53–0.85、RMSE约1.06–1.76 t/ha,适用于地面产量样本稀缺地区的估产,但结论基于美国玉米带,跨气候区适用性待验证。
The hidden dimension: Resolving the vertical gradients of solar-induced chlorophyll fluorescence emissions and gross primary production in modern maize canopy
The hidden dimension: Resolving the vertical gradients of solar-induced chlorophyll fluorescence emissions and gross primary production in modern maize canopy。Remote Sensing of Environment
关注点该研究聚焦现代玉米冠层内太阳诱导叶绿素荧光与总初级生产力的垂直梯度解析,为改进冠层尺度遥感估算模型提供方法参考,适用边界限于玉米冠层遥感监测场景。
Interpretable minimal-input reference evapotranspiration modeling for smart agriculture in data-scarce regions: Integrating reconstructed land surface temperature and air temperature
Interpretable minimal-input reference evapotranspiration modeling for smart agriculture in data-scarce regions: Integrating reconstructed land surface temperature and air temperature。European Journal of Agronomy
关注点该研究用重建地表温度替代传统多气象输入估算参考蒸散量,适用于气象站点稀疏地区的灌溉决策,但精度与可迁移性需在本地数据上验证。
Integrating tower-based camera observations with time-series Sentinel-1/2 images for parcel-level crop yield estimation over large areas
Integrating tower-based camera observations with time-series Sentinel-1/2 images for parcel-level crop yield estimation over large areas。Smart Agricultural Technology
关注点该研究把地面塔基相机时序观测与Sentinel-1/2卫星影像结合,用于大范围地块级产量估算,其适用性依赖塔基站点覆盖密度与地块破碎程度。
Interpretable multi-branch fusion GRU framework for wheat yield prediction using multi-source phenological remote sensing time series
融合多源物候遥感时序数据,构建可解释多分支GRU模型预测小麦产量,兼顾预测精度与模型可解释性,为区域产量早测与田间管理提供依据。
Machine learning-driven mapping of salt-affected soils: integrating laboratory analysis and environmental covariates to establish extent and spatial distribution
该研究利用机器学习结合遥感与土壤化验数据,绘制三个灌区盐渍土范围与空间分布图。
关注点该研究在三个灌溉区以250米网格采样,结合GEE遥感指数与RF等模型预测盐渍化指标,RF的R²达0.83–0.91,可用于识别需耐盐品种与土壤改良的优先地块。
Deep learning–based object detection for automated classification of ripe and unripe tomatoes in field conditions
对比Faster R-CNN与YOLO26s在温室番茄成熟度检测中的性能。
Hyperspectral, Multimodal Fusion, and TinyML for Rice Crop Monitoring: A Focused Survey and Synthesis Toward the TinyRiceNet Framework
Hyperspectral, Multimodal Fusion, and TinyML for Rice Crop Monitoring: A Focused Survey and Synthesis Toward the TinyRiceNet Framework。Smart Agricultural Technology
关注点该综述系统梳理高光谱与多模态融合结合TinyML用于水稻监测的技术路线,并指向TinyRiceNet框架,可为轻量化边缘部署研究提供选型参考,但尚无实测性能数据支撑。
A tool for quantifying Grapevines Vigour: Accurate canopy cover and crop volume assessment via drone-based point clouds from conventional and oblique RGB cameras
This study aimed to develop a tool capable of capturing the spatial variability of grapevine vigour from a geometric approach. The methodology was evaluated during the 2023, 2024, and 2025 growing seasons in a 1.27 ha trellised vineyard in Tarazona de la Mancha (Albacete, Spain). Green canopy cover (GCC) estimates were compared with reference data obtained using GEO-LAIC software across 33 uncrewed aerial vehicle (UAV) flights. Biomass volume (BV) estimates were compared with values derived from the alphashape algorithm using 10540 observations. The relationship between BV and pruning weight was also assessed to identify the optimal period for biomass estimation. GCC estimates showed strong agreement with GEO-LAIC, achieving coefficients of determination (R 2 ) of up to 0.97 and root mean square error (RMSE) values below 0.04. BV estimates showed moderate agreement with the alphashape-derived values (R 2 = 0.65), providing a basis for further calibration. The optimal period for estimating pruning weight was identified between BBCH growth stages 75 and 81. BV showed a stronger relationship with pruning weight than GCC. The proposed geometric approach effectively captures grapevine vigour variability while prioritising geometric accuracy over radiometric quality, potentially enabling the use of low-cost RGB sensors. This methodology offers a promising tool for precision viticulture and spatially targeted vineyard management.
A novel probabilistic generative framework of denoising point clouds for individual crops and large-scale natural vegetation
A novel probabilistic generative framework of denoising point clouds for individual crops and large-scale natural vegetation。Computers and Electronics in Agriculture
关注点该框架尝试用概率生成模型统一处理单株作物与大规模自然植被的点云去噪,适用于三维表型与遥感点云预处理场景,但摘要未给出数据规模与精度指标,结论可靠性待正文验证。
Hierarchical radiative transfer model with actual leaf angles for vertical chlorophyll profiling in rice
Hierarchical radiative transfer model with actual leaf angles for vertical chlorophyll profiling in rice。Computers and Electronics in Agriculture
关注点该模型以实测叶倾角分层刻画冠层辐射传输,用于水稻叶绿素垂直分布反演,适用边界在于特定品种与冠层结构条件下的验证精度。
Sowing the seeds for practical multimodal crop yield prediction: A systematic review with recommendations for future work
Sowing the seeds for practical multimodal crop yield prediction: A systematic review with recommendations for future work。Computers and Electronics in Agriculture
关注点该综述系统梳理多模态数据融合用于作物产量预测的方法体系,并提出面向实际应用的后续研究建议,可作为方法选型与研究方向参考,但不提供新的实证结果。
面向作物表型的多光谱影像超分辨率方法性能评测 [预印本]
评测8种单图超分辨率方法对作物多光谱影像分辨率与光谱保真的提升效果。
关注点在4×与16×重建下,Transformer类方法(HAT、HiT-SR、SRFormer)在保持植被指数可靠性上优于GAN与扩散模型,可作为作物表型多光谱影像超分选型的初步依据,但结论来自预印本、尚未经同行评审。
Neural network-based UAV thermal calibration and crop water stress assessment in an apple orchard
Neural network-based UAV thermal calibration and crop water stress assessment in an apple orchard。Smart Agricultural Technology
关注点该研究将神经网络用于无人机热红外影像校准,以提升苹果园作物水分胁迫评估精度;其适用性依赖特定果园与传感器条件,推广前需外部数据验证。
Satellite-derived vegetation indices and machine learning for food crop yield prediction: A systematic review of methods, performance and research gaps
系统综述32项研究,评估卫星植被指数与机器学习预测粮食作物产量的方法与性能。
Application of UAV Remote Sensing in Crop Detection
系统综述无人机遥感在作物病虫害、倒伏、杂草和土壤墒情检测中的应用进展。