Accurate estimation of above-ground biomass (AGB) is critically important for monitoring crop growth and guiding precision agriculture practices. However, the quantitative relationship between vegetation indices (VIs) and AGB is significantly influenced by biomass allocation, particularly during the reproductive growth stage when dry matter allocation shifts toward storage organs, substantially reducing the proportion of leaf biomass to total AGB. Moreover, in high-density wheat populations, vegetation indices often suffer from saturation effects, limiting the accuracy of traditional spectral index-based AGB estimation. To overcome these challenges, we integrate spectral remote sensing with agronomic prior knowledge (biomass partitioning patterns and phenological variables) to develop a power function-based method for predicting winter wheat AGB. The Zadoks scale (ZS), days after sowing (DAS), and growing degree days (GDD) were used as phenological variables (PVs) to investigate their impact on model performance. Results demonstrated that models incorporating PV significantly outperformed both the traditional remote sensing method (R2 = 0.34, RMSE = 3.70 t/ha) and the model without PV (R2 = 0.01, RMSE = 4.94 t/ha), achieving higher accuracy (R2 = 0.78-0.82, RMSE = 2.09-2.69 t/ha). Among these, the ZS-based model performed best in the training set, while the GDD-based model showed greater stability in the testing set. This study suggests that integrating leaf biomass and PVs can overcome spectral saturation issues and significantly improve the accuracy of winter wheat AGB estimation across growth stages.

Accurate, real-time, non-destructive estimation of single-bud biochemistry is critical for managing green-tea quality, yet non-imaging hyperspectral measurements of arbitrarily stacked buds are confounded by structured interference from mixed adaxial/abaxial surfaces and their three-dimensional arrangement. We propose Bilinear Spectral Derivative Gaussian Process Regression (BSDGPR), which couples a bilinear spectral-derivative model with Gaussian process regression to minimise structured interference. First- and second-order spectral derivatives attenuate the weakly wavelength-dependent leaf-spatial-arrangement term (RLSA), while unit-sphere normalization combined with a chord-distance radial basis function kernel eliminate the plan-view leaf-front area fraction (f). This effectively standardizes tea-bud heap spectra to single-bud equivalents. (i) Structure suppression and invariance: independent-contribution analysis shows that moving to the derivative domain boosts the heap term's explanatory power to 80.7–99.4 % across bands while reducing RLSA to ≤9.3 % (TH:LSA ratio of between 4.8 and 40.9), and the sphere-normalized chord-distance kernel cancels f-driven gain variation across samples. (ii) Accuracy and efficiency: relative to reflectance-domain GPR, BSDGPR raises R2 by 0.13 and lowers NRMSE by 3–5 percentage points for tea polyphenols (0.68 → 0.81; 15.55 % → 12.64 %), amino acids (0.56 → 0.71; 19.78 % → 15.72 %), and soluble sugars (0.86 → 0.91; 10.85 % → 9.05 %), while maintaining fast inference (10–13 s per model) suitable for near real-time use. (iii) Transferability: models trained on tea-bud heaps generalize to canopy measurements with minimal loss (ΔR2 ≤ 0.05; canopy NRMSE 11–15 %), remain credible across cultivars (R2 = 0.37–0.74; NRMSE ≤23 % across 15 varieties), are stable at typical sensor resolutions (≤ 10–15 nm), and achieve R2 = 0.63–0.66 with NRMSE ≤22 % for wheat canopy chlorophyll. Overall, BSDGPR is a reliable, field-ready AI approach for non-destructive biochemical sensing across crops, sensors, and management scenarios, offering broad applicability and strong potential for precision agriculture.

Timely and accurate estimation of component biomass of winter wheat, including leaf dry biomass (LDB), stem dry biomass (SDB), and reproductive organ dry biomass (RDB), is critical for crop growth monitoring and yield assessment. Canopy spectra mainly reflect leaf information, allowing for effective LDB estimation, whereas estimating SDB and RDB requires consideration of growth stage effects. To address this, we developed a hybrid biomass estimation framework by combining deep learning with biomass allocation law. Specifically, (1) a component biomass hierarchical (CBH) model was proposed based on accumulated growing degree days (AGDD) and biomass allocation laws; (2) a deep learning model (LBNet), based on two-dimensional fractional-order differential (2DFOD) hyperspectral images, was pre-trained on PROSAIL-simulated data and fine-tuned with field data to improve LDB estimation; and (3) the LBNet and CBH models were integrated to estimate and map component biomass across multiple scales. The hybrid framework achieved robust performance across interannual, regional, and UAV-based validations. For LDB, the root mean square error (RMSE) was 0.28-0.38 t ha-1, with a normalized RMSE (nRMSE) of 9.79%-14.50%. The RMSEs for SDB and RDB were 0.88-1.63 t ha-1 (nRMSE = 11.05%-19.25%) and 0.76-2.22 t ha-1 (nRMSE = 9.29%-22.66%), respectively. Overall, the proposed method provides an effective approach for multi-stage biomass estimation of winter wheat and demonstrates highly promising potential for applications in smart agriculture and crop yield assessment. (c) 2025 Crop Science Society of China and Institute of Crop Science, CAAS. Production and hosting by Elsevier B.V. on behalf of KeAi Communications Co., Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Aiming at the challenges of strong background noise, complex feature extraction, and limited edge-side computing resources in power quality disturbance (PQD) monitoring for next-generation power systems, this paper proposes a novel lightweight recognition framework based on IMW-Denoising and M3-MobileViT. First, a three-stage preprocessing strategy integrating ICEEMDAN, multi-scale permutation entropy (MPE), and wavelet thresholding is developed to achieve adaptive denoising and refined signal reconstruction under low signal-to-noise ratio conditions. Subsequently, Gramian Angular Field (GAF), Continuous Wavelet Transform (CWT), and Recurrence Plot (RP) are employed to construct complementary multi-view feature representations from temporal, time-frequency, and nonlinear dynamic perspectives. Furthermore, a lightweight M3-MobileViT model is designed by embedding a MONA module with structural reparameterization (MBRConv) into the MobileViT backbone, enabling efficient heterogeneous feature fusion without increasing inference complexity. Experimental results demonstrate that the proposed method achieves an average recognition accuracy of 99.645% for 31 types of single and compound disturbances under a 20 dB noise environment, exhibiting excellent anti-noise robustness and recognition performance. The proposed framework provides an effective solution for real-time PQD monitoring in edge-side power systems.

In the red and yellow soil regions of southern China, long-term practices in soil acidification constraint management have revealed several challenges, including fragmented knowledge systems, high technical thresholds, and pronounced regional disparities in technical capacity, which collectively hinder the systematic integration and dissemination of management experience. To address these limitations, this study proposes a lightweight autonomous construction framework for a knowledge graph of soil acidification management based on large language model (LLM) agents. By introducing an "AI associative matrix" and a semantic expansion-retrieval mechanism, the framework enables the system to automatically infer related concepts and relationships from a small set of domain-specific keywords, thereby generating an initial ontology structure and knowledge graph schema for management tasks. In addition, a network protocol-based control strategy is incorporated to semantically parse publicly available governmental reports and monitoring data, facilitating automated processes such as entity recognition and relation extraction. At the system implementation level, a cloud-edge collaborative lightweight architecture is adopted to reduce computational requirements and improve usability. Through an empirical in-situ deployment and lightweight field-level testing conducted directly at the farm sites of Chengmai County, Hainan Province, this framework demonstrated a robust capacity to adapt to users with different technical backgrounds and regions with varying levels of resource input, while providing extensibility for the continuous integration of emerging technologies and updated policy regulations. Overall, this study offers a lightweight, user-friendly, and scalable intelligent approach for cultivated land quality improvement and agricultural ecological management.

[Objective]As a major grain crop,maize plays a critical role in global food security.The ears of maize serves as a key pheno-typic trait,providing essential information on the plant's physiological and agronomic status.Its morphological characteristics,size,and color effectively reflect the plant's growth status and potential yield.Therefore,accurately acquiring images of maize ears in the field across different growth stages is crucial for breeding research and yield prediction.Traditional field detection of maize ears relies heavily on manual labor,which is not only inefficient and labor-intensive but also struggles to meet the high-throughput demands of modern precision breeding programs.There is an urgent need for efficient,automated detection technologies that can operate reliably under real-world field conditions.To address the requirement for efficient acquisition of maize ears phenotypic traits in field breeding work,the objective of this research is to develop a robust object detection solution suitable for large-scale field environments.An im-proved CornYOLO model based on the YOLO11n(You Only Look Once)architecture was designed to enhance the detection accura-cy and efficiency of maize ears in complex field environments.[Methods]Image data were acquired using an unmanned ground vehi-cle(UGV)equipped with a high-resolution panoramic camera,which traversed multiple experimental plots under varying lighting and growth conditions.A dataset containing 1 152 annotated samples was constructed,covering diverse ear morphologies and occlusion scenarios.Dynamic data augmentation techniques were applied during training to enhance the model's generalization capability.Three key enhancements were introduced to the YOLO11n detection framework.First,a cross stage partial network with dynamic pointwise spatial attention(C2PDA)module was designed to replace the cross stage partial with pointwise spatial attention(C2PSA)module in the YOLO11 backbone network.This module enhanced spatial discriminability and channel sensitivity in feature representation through the collaborative integration of a dynamic channel weighting mechanism and position-aware modeling.It significantly im-proves the model's performance in identifying maize ears under challenging field conditions such as occlusion of stems and leaves and multi-scale target distribution.Second,the spatial pyramid pooling-fast(SPPF)module in the original model was replaced with an fea-ture refinement module(FRM)to optimize multi-scale feature fusion.The FRM functions via directional feature decomposition and an adaptive attention mechanism.It captures fine-grained spatial structural information through horizontal and vertical bidirectional pooling and combines spatial-channel cooperative attention for dynamic feature calibration,thereby improving recognition accuracy across varying ear sizes and complex backgrounds.Finally,the unified intersection over union(UIoU)loss function was introduced to optimize bounding box regression accuracy.UIoU is an innovative loss function that emphasizes weight allocation among prediction boxes of different qualities.It adaptively adjusted the weight of each prediction box's loss term based on the IoU value or its monoton-ic function,assigning higher weights to lower-quality predictions to prioritize their optimization,while reducing weights for high-qual-ity boxes to prevent over-optimization.[Results and Discussions]Experimental results demonstrate that CornYOLO achieved a mAP@50 of 89.3％on the validation set,with the F1-Score increasing by 2.5 percentage points.Compared to widely used lightweight models including YOLOv3-tiny,YOLOv5n,YOLOv8n,YOLOv10n,real-time detection transformer(RT-DETR)and YOLO13n,CornYOLO showed significantly superior detection performance in complex field environments,with mAP@50 improvements of 2.2,1.9,1.8,5.7,12.6 and 2.4 percentage points,respectively.These results fully validate that CornYOLO can efficiently and accurately extract maize ear images under field conditions,providing a technical foundation for precise phenotypic evaluation and yield prediction.Fur-thermore,ablation studies were conducted:Introducing the C2PDA module improved the model's mAP@50 by 0.5 percentage points and the F1-Score by 0.5 percentage points.However,after incorporating the FRM module,which successfully enhanced multi-scale detection performance and increased the F1-Score by 1.5 percentage points,the integration of these two modules resulted in the genera-tion of a small number of low-quality detection boxes.The original loss function was inefficient in optimizing such boxes,resulting in no improvement in mAP@50 after the modification.To address this issue,the UIoU loss function was introduced.By dynamically ad-justing weight assignments based on prediction quality,it significantly improved the regression performance for low-quality detection boxes,thereby enhancing the localization accuracy and convergence stability of the model in dense target scenarios.The final CornY-OLO model exhibited excellent overall performance:Compared to the original YOLO11n,the F1-Score increased by 2.5 percentage points and mAP@50 improved by 1.1 percentage points.The experimental results fully demonstrate that CornYOLO effectively en-hances the detection capability for maize ears in complex field environments compared to the baseline YOLO11n model.[Conclusions]The CornYOLO model proposed in this study incorporates three key components:C2PDA,FRM,and UIoU,which enhances model convergence and localization performance in dense and occluded scenes,enables the model to effectively and precisely identify maize ears under practical conditions,thereby providing reliable technical support for phenotypic analysis and yield prediction in maize breeding.Future work will focus on extending the model to other crop types and further optimizing inference efficiency for real-time deployment on mobile platforms.

Accurate quantification of crop residue cover(CRC)is crucial for monitoring and evaluating conservation tillage practices,yet it poses a significant image segmentation challenge.The subtle visual distinctions between fragmented residue and soil,compounded by variable illumination and shadows in field imagery,often lead to poor segmentation performance.To overcome these limitations,we introduce RCTUnet,a novel deep learning architecture designed for robust crop-residue-soil segmentation and precise CRC estimation.RCTUnet's architecture synergistically integrates three key components:(1)a ResNet50 backbone for deep,multi-scale feature extraction;(2)a convolutional block attention module(CBAM)to adaptively focus on salient residue features across both channel and spatial dimensions;and(3)a transformer-based global context fusion module(GCFM)to model long-range spatial dependencies,which is critical for interpreting heterogeneous residue patterns.We evaluated RCTUnet on a dataset of 1220 field-acquired images spanning four typical crop rotations.Experimental results show that,compared to traditional models:(1)RCTUnet achieves significantly higher crop-residue-soil segmentation accuracy than classic models including Unet,Unet++,DeepLabV3,segmentation network(SegNet),and fully convolutional network(FCN),with improvements of 3.24%,3.42%,4.88%,8.28%,and 6.05%in overall accuracy,respectively;(2)RCTUnet yields superior residue-soil segmentation performance,with increases in residue recall of 7.67%,7.37%,14.09%,27.05%,and 16.91%,respectively;(3)RCTUnet shows enhanced CRC estimation accuracy,achieving a root mean square error(RMSE)of 4.875,representing a 45.5%improvement over Unet(RMSE=8.941).These results demonstrate the efficacy of our hybrid approach,which combines deep hierarchical features,dual-domain attention,and global context modeling.RCTUnet provides a robust and reliable tool for automated CRC assessment,advancing the capabilities of in-field agricultural monitoring.

Accurate aboveground biomass estimation with quantified uncertainty is essential for precision agriculture, enabling risk-aware decision-making and strategic model improvement. Existing approaches predominantly provide point estimates without uncertainty quantification, limiting their operational utility for trustworthy Artificial Intelligence (AI) deployment. This study presents a Multi-modal Attention-based Uncertainty Quantification Network (MA-UQNet), which achieves superior prediction accuracy (R2 = 0.856) with well-calibrated uncertainty (97.18% coverage) for wheat aboveground biomass estimation through integrated multi-modal attention, growth stage-specific processing, and epistemic-aleatoric uncertainty decomposition. The framework integrates hyperspectral remote sensing with environmental variables via joint attention mechanisms that adapt to phenological variations. Model development employed a decade-spanning dataset (2012-2022, 1272 samples) collected under factorial combinations of nitrogen rates (0-270 kg/ha), irrigation levels (0-384 mm), and wheat cultivars across four growth stages. Temporal extrapolation validation using chronological partitioning (2012-2019 for training and 2020-2021 for testing) demonstrated robust generalization, substantially outperforming Random Forest (R2 = 0.751, coverage = 76.61%) and nine representative baselines, including Bayesian Neural Networks (R2 = 0.805, coverage = 38.31%). Uncertainty decomposition revealed epistemic uncertainty to be moderately dominant (53%) relative to aleatoric uncertainty (47%), indicating that strategic data collection offers greater potential for uncertainty reduction than improving measurement precision alone. These findings provide validated tools for uncertainty-aware biomass estimation in precision agriculture.(c) 2026 The Authors. Publishing services by Elsevier B.V. on behalf of KeAi Communications Co., Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

In three-dimensional applications of the Moving Particle Semi-Implicit (MPS) method, free-surface identification is commonly based on particle number density, which can lead to misclassification of surface particles and consequently degrade the accuracy of pressure field predictions. To address this issue, an improved formulation for computing the relative position divergence (RPD) is first developed to achieve better consistency with theoretical values. Based on geometrical considerations and test cases involving internal cavities, a set of surface identification criteria suitable for three-dimensional conditions is established. Building upon the improved RPD formulation, a new free-surface particle detection method is proposed by coupling the relative position divergence with an approximate surface normal vector estimation (RPD + NV). Validation using geometrical models demonstrates that the proposed method can accurately identify surface particles on convex and concave boundaries as well as within cavity-containing configurations. Further assessments through a series of MPS simulations—including hydrostatic, dam-break, and damped oscillation cases—show that both the improved RPD method and the combined RPD + NV scheme effectively reduce pressure oscillations. In particular, the RPD + NV method significantly decreases false negatives and false positives in surface particle identification, leading to smoother and more physically consistent pressure fields, without introducing a noticeable increase in computational cost. Therefore, the proposed surface particle detection method provides a robust and efficient solution for three-dimensional MPS simulations.

Agricultural sustainable development for food and environmental security relies largely upon our understanding of the functions of an agricultural ecosystem and their relationships. This study aims to define functional zones by identifying the relationships among the individual functions of the Guangdong Province’s agricultural land resources for sustainable development. Therefore, a specific Bayesian belief network (BBN) was built to identify the trade-off/synergy relationships between functions, and the density peaks clustering (DPC) algorithm was used to perform function-oriented zoning. As a result, significant synergies of food production function are identified with other functions, while a trade-off exists with food cleanliness. The 1 km grid scale set as the zoning unit can be more appropriate for a land use plan to consider functional relationships and specific management options. Ultimately, the Guangdong agricultural land is zoned into five functional zones (green agriculture, tropical agriculture, productive agriculture, ecological agriculture, and urban agriculture) with specific options and stakeholders’ policies that are suggested to weaken existing and potential trade-offs accordingly. Evidently, coupling BBN with DPC demonstrates an enhanced capability to optimize an agricultural land use plan by balancing multifunctional synergies/trade-offs for advancing sustainable agricultural development.

Accurate and timely crop yield estimation is essential for ensuring food security. Although deep learning methods integrating multi-source remote sensing and meteorological data have been widely applied in yield prediction, existing studies still face limitations in capturing crop growth temporal dynamics, quantifying the full-season temporal contributions of multi-source features, and assessing prediction uncertainty induced by climate variability. These limitations constrain model accuracy, interpretability, and cross-regional generalization capability. To address these challenges, this study proposes an interpretable yield estimation framework that combines a Temporal Convolutional Network (TCN) with Long Short-Term Memory Networks (LSTM) in a serial architecture (TCN–LSTMs). Using multi-source time-series data, we conduct county-level winter wheat yield prediction and attribution analysis across Henan Province, revealing the temporal cumulative effects of key growth-related variables and the uncertainty introduced by late-season climate fluctuations. The results show that the TCN–LSTMs model (R2=0.81, RMSE = 574.26 kg/ha) achieves notably better predictive performance than baseline models. Specifically, it outperforms CNN–LSTM (R2=0.78) and TCN–LSTMp (R2=0.75), as well as attention-based architectures like Temporal Fusion Transformer(TFT) (R2=0.74) and Transformer (R2=0.69), reducing the RMSE by 6.1% to 21.1%. Leave-one-out cross-validation and additional experiments in the Huang–Huai–Hai region further verify the robustness and generalization ability of the proposed model. The relative prediction error (RE) is mainly concentrated between -10% and 10%, and accurate yield estimation can be achieved one month in advance. The attribution analysis based on Integrated Gradients (IG) indicates that early-season soil temperature and moisture affect emergence rates; mid-season canopy photosynthesis and soil moisture influence spike number and grain-setting rate; and late-season photosynthetic efficiency and water availability are critical for grain filling and thousand-kernel weight. The period from mid-March to late April provides the most informative signals for yield prediction. Uncertainty analysis further shows that early-stage stable signals, such as vegetation indices, help reduce predictive uncertainty, whereas mid- and late-season uncertainty fluctuates due to climate variability and management differences. Overall, the proposed TCN–LSTMs framework not only improves the accuracy and robustness of winter wheat yield prediction but also enhances interpretability and early warning capability. This study provides an effective pathway for regional crop yield estimation and food security assessment.

Leaf area index (LAI) is an important structural parameter of crops and it is usually estimated non-destructively using reflectance spectra from various reflectometers. Prevailing models, often trained on single-crop and singleyear data, lack generalizability. As rotation crops with similar morphology, rice and wheat present an opportunity to develop generalized models; however, their spectral response patterns are not well compared, and adaptable multi-year, multi-crop LAI models remain scarce. To bridge this gap, we developed a generalized LAI estimation model for both crops by integrating physically-based simulation with data-driven deep learning. Key steps included canopy spectral simulation, data augmentation, and model construction with a 1D-CNN and transfer learning. The PROSAIL model was employed to simulate canopy reflectance spectra, with crop growth stages stratified into two phenological phases: sowing-heading stage (LAI: 0.01-5, increment: 0.2) and headinggrouting stage (LAI: 3-8, increment: 0.2). To enhance ecological fidelity, the LSMM was integrated to simulate mixed spectral scenarios involving soil background, water interactions, and spike contributions, while 5% Gaussian noise was systematically introduced to approximate real-world environmental variability. The results showed that the R2 values of the SMOTE-1D-CNN model for the different datasets (four rice and two wheat) ranged from 0.62 to 0.87, and the RMSE values ranged from 0.55 to 1.22. The model achieved a relatively high R2 (0.79 f 0.09) for rice LAI estimation but exhibited a larger RMSE (0.8 f 0.29). For wheat, the R2 was slightly lower (0.74 f 0.17), while the RMSE was smaller and more stable (0.56 f 0.01). These discrepancies reflect how crop characteristics or data distribution may influence estimation accuracy. SMOTE is used as a data enhancement to reduce the "high underestimation" phenomenon of the model, and the model performance is stabilized when the multiplicity of the sample size (n) is greater than or equal to 5. And the model input feature importance is only related to the original sample (the original unenhanced dataset) and does not change with "n". This study demonstrates that a hybrid methodology, fusing physically-based simulation with deep learning, offers significant potential for robust, multi-crop LAI inversion, providing novel insights and technical support for crop monitoring and management.

Lodging substantially disrupts the canopy architecture of winter wheat and introduces considerable uncertainty in unmanned aerial vehicle (UAV)-based estimation of aboveground biomass (AGB). To investigate the influence of lodging on biomass estimation, a field experiment was conducted at the Yuanyang experimental station in Xinxiang, Henan Province, China. The experiment comprised four nitrogen application levels (N0, N8, N15, and N22) and four wheat cultivars (XM26, YM15, YN186, and YN908), encompassing 48 experimental plots. Multispectral UAV imagery was acquired at 14, 21, 28, and 35 days after flowering. A total of 18 vegetation indices were derived and combined with field-measured AGB to develop machine learning estimation models. The results indicated that lodging significantly altered the relationship between vegetation indices and AGB, with clear phenological stage-dependent differences. Prior to lodging, most vegetation indices exhibited strong correlations with AGB (r>0.5). Following lodging, indices sensitive to canopy structural changes, such as the difference vegetation index (DVI) and ratio vegetation index (RVI), showed markedly reduced correlations with AGB. In contrast, red-edge and green-based indices, including the normalized difference red-edge index (NDRE) and chlorophyll index green (CIgreen), maintained relatively stable relationships with biomass. Among the tested models, Random Forest Regression (RFR) achieved the highest estimation accuracy (R2=0.718,rRMSE=11.5%), outperforming Gaussian Process Regression (GPR) and Support Vector Regression (SVR). SHAP analysis further revealed that red-edge and green spectral band indices, such as the nonlinear index (NLI), modified green-red vegetation index (MGRVI), and green-red difference vegetation index (GRDVI), contributed most to AGB estimation and exhibited pronounced nonlinear effects, while DVI and RVI played comparatively minor roles. Moreover, a notable interaction effect was identified between NLI and MGRVI. Overall, the findings highlight that the mid-flowering stage represents a critical window for UAV-based biomass monitoring. Red-edge and green spectral indices demonstrated greater robustness under lodging conditions and provided valuable guidance for improving biomass estimation and yield assessment in lodging-prone wheat production systems.

Crop lodging significantly impacts wheat yield and quality, necessitating rapid and accurate identification methods for post-disaster response and agricultural insurance assessment. While extreme weather events causing lodging have increased in frequency, conventional semantic segmentation approaches face limitations in global context perception. This study introduces UssNet, an innovative algorithm integrating contextual and spatial awareness by combining UNet's semantic segmentation strengths with State Space Models (SSM). We selected UNet as our base architecture due to its proven effectiveness with limited training data, encoder-decoder structure with skip connections that preserve critical spatial information for lodging detection, and optimal balance between computational complexity and performance. UssNet implements local auxiliary mechanisms with SSM, enabling selective scanning of feature maps from multiple directions to achieve linear computation of long sequences and comprehensive extraction of global contextual information. To address class imbalance challenges and improve recognition of small lodging areas, we incorporate the Focal Loss function. Additionally, we replace ReLu with GeLu activation to mitigate the "dead ReLu" phenomenon while maintaining overfitting suppression benefits. Experimental results demonstrate UssNet's superior performance, achieving a pixel accuracy (PA) of 0.971, mean intersection over union (mIoU) of 0.931, recall of 0.85, and F1score of 0.82 on the test dataset. Comparative analysis against state-of-the-art models confirms UssNet's enhanced capability in capturing global context information, providing an efficient approach for wheat lodging monitoring with valuable applications in yield estimation and disaster management.

Timely and accurate prediction of stem dry biomass (SDB) is crucial for monitoring crop growing status. However, conventional biomass estimation models are often limited by the influence of crop growth phase, which significantly restricts their temporal and spatial transferability. This study aimed to develop a semi-mechanistic stem biomass prediction model (PVWheat-SDB) using phenological variable (PV) to accurately predict winter wheat SDB across different growth stages. The core of the model is to predict SDB using PV under remote-sensed canopy vegetation indices (VIs) constraint. The results demonstrated that VIs can quantify the variations in stem growth equations under different planting conditions and varieties. The developed a PVWheat-SDB model using normalized difference red edge (NDRE) and accumulated growing degree days (AGDD) performed well for SDB predictions, with R2 , RMSE, nRMSE and MAE values of 0.88, 75.48 g/m2, 8.04 % and 55.36 g/m2 for the validation datasets of field spectral reflectance, and 0.82, 81.76 g/m2, 11.22 % and 62.82 g/m2 when transferred to unmanned aerial vehicle (UAV) hyperspectral images. Furthermore, the model can not only estimate SDB at the current growth stage, but also predict SDB of subsequent phenological stages. The growth stage stacking strategy indicated that the model accuracy improves significantly as more growth stages are incorporated, especially during the reproductive stages. These results all highlight the robustness and transferability of the PVWheat-SDB model in accurately predicting SDB across different growing seasons and growth stages. (c) 2025 The Authors. Publishing services by Elsevier B.V. on behalf of KeAi Communications Co., Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Fruit tree diseases often show vertical heterogeneity within the canopy, however, most remote sensing based monitoring methods rely on observations from a single vertical viewing angle and assume a uniform disease distribution. This simplification reduces the accuracy of disease severity diagnosis, especially facing similar spectral response caused by different disease vertical patterns at early stages. To overcome these limitations, we propose a framework that combines dual-angle remote sensing with a three-dimensional radiative transfer model (3D RTM) to jointly diagnose canopy disease vertical distribution and severity. The LargE-Scale remote sensing data and image Simulation framework (LESS) was used to simulate canopy scenarios with different disease severities and vertical distribution, enabling a systematic analysis of spectral responses under different viewing zenith angles. Based on the simulated canopy reflectance and the spectral responses to vertical disease distribution, we constructed a novel dual-angle vegetation index (DAVI) to retrieve a vertical distribution index (VDI) that represents the relative vertical position of disease within the canopy. The retrieved VDI was introduced as a knowledge-based constraint into a hierarchical linear model for disease severity index (DSI) estimation. This integration constitutes a knowledge-guided disease severity diagnosis framework (KG-DDS), which was validated using field measurements and demonstrated its applicability under UAV-based observations. Our results show that the proposed DAVI is sensitive to disease vertical distribution and enables stable retrieval of the VDI across different canopy structures, with R2 exceeding 0.85 for simulated data and approximately 0.65 for field measurements. By explicitly incorporating vertical distribution information, the proposed KG-DDS improves the mean R2 of DSI retrieval from 0.35 to 0.67, significantly enhancing disease severity estimation accuracy. Validation with both simulated and field data confirms the effectiveness of the framework, while also revealing that its transferability depends on the reselection of angle–band combinations and recalibration under specific disease and canopy conditions. These findings underscore the importance of linking spectral responses to vertical canopy disease distribution for knowledge-guided disease severity diagnosis, offering new opportunities for reliable canopy-scale disease assessment and early detection.

Precise monitoring of wheat phenology (BBCH scale) is essential for agricultural optimization, yet UAV-based single-phase monitoring encounters spectral ambiguities where multiple vegetation indices correspond to identical growth stages. A dual-mode framework integrating time-series reconstruction with hybrid deep learning was developed to resolve this limitation. UAV multispectral and digital imagery (333 plots, 2023-2024) enabled reconstruction of daily-resolved vegetation indices, color/texture features, and BBCH stages using Gaussian, PCHIP, and linear fitting to mitigate environmental noise. Synthetic datasets incorporating Gaussian noise (5-100 % relative intensity) simulated field variability. Feature selection was optimized through Competitive Adaptive Reweighted Sampling (CARS) and Variance Inflation Factor (VIF). Hybrid CNN-GRU and CNN-LSTM architectures surpassed standalone networks by resolving spectral ambiguities in single-phase data and leveraging temporal patterns during time-series analysis. Time-series models attained maximum accuracy under noise-free conditions (CNN-GRU: R2 = 0.90-0.98, RMSE = 3.61-7.65 BBCH units), with accuracy decreasing proportionally to noise intensity. Conversely, single-phase models demonstrated peak performance at 20 % noise intensity (CNN-GRU: R2 = 0.56-0.70, RMSE = 15.33-17.22 BBCH units), achieving optimal balance between robustness and practicality for real-time farm monitoring. Extreme noise (100 %) distorted feature distributions (7.25-8.73x expansion), validating controlled augmentation. A novel Rate of Phenological Development (RPDW) -quantified as the slope of BBCH progression-was derived to inform breeding programs, while the noiseoptimized single-phase approach enables resource-efficient phenology tracking for family farms. This work bridges methodological innovation (adaptive noise strategies, hybrid architectures) with scalable solutions for precision agriculture, advancing UAV-based phenology monitoring in both academic and applied contexts. (c) 2025 The Authors. Publishing services by Elsevier B.V. on behalf of KeAi Communications Co., Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

The maize tassel represents one of the most pivotal organs dictating maize yield and quality. Investigating its phenotypic information constitutes an exceedingly crucial task within the realm of breeding work, given that an optimal tassel structure is fundamental for attaining high maize yields. High-throughput phenotyping technologies furnish significant tools to augment the efficiency of analyzing maize tassel phenotypic information. Towards this end, we engineered a fully automated multi-angle digital imaging apparatus dedicated to maize tassels. This device was employed to capture images of tassels from 1227 inbred maize lines falling under three genotype classifications (NSS, TST, and SS). By leveraging the 3D reconstruction algorithm SFM (Structure from Motion), we promptly obtained point clouds of the maize tassels. Subsequently, we harnessed the TreeQSM algorithm, which is custom-designed for extracting tree topological structures, to extract 11 archetypal structural phenotypic parameters of the maize tassels. These encompassed main spike diameter, crown height, main spike length, stem length, stem diameter, the number of branches, total branch length, average crown diameter, maximum crown diameter, convex hull volume, and crown area. Finally, we compared the GFC (Gaussian Fuzzy Clustering algorithm) used in this study with commonly used algorithms, such as RF (Random Forest), SVM (Support Vector Machine), and BPNN (BP Neural Network), as well as k-Means, HCM (Hierarchical), and FCM (Fuzzy C-Means). We then conducted a correlation analysis between the extracted phenotypic parameters of the maize tassel structure and the genotypes of the maize materials. The research results showed that the Gaussian Fuzzy Clustering algorithm was the optimal choice for clustering maize genotypes. Specifically, its classification accuracies for the Non-Stiff Stalk (NSS) genotype and the Tropical and Subtropical (TST) genotype reached 67.7% and 78.5%, respectively. Moreover, among the materials with different maize genotypes, the number of branches, the total branch length, and the main spike length were the three indicators with the highest variability, while the crown volume, the average crown diameter, and the crown area were the three indicators with the lowest variability. This not only provided an important reference for the in-depth exploration of the variability of the phenotypic parameters of maize tassels but also opened up a new approach for screening breeding materials.

Radiative transfer models (RTMs) describe how light is absorbed, scattered, and transmitted within leaves by simulating mechanistic light propagation processes. The PROSPECT model is based on measurable parameters (the leaf biochemical content) and a non-measurable parameter (the leaf anatomical structure represented by the leaf structure parameter (N)). The effect of N on the optical properties of leaves has been investigated through a number of local and global sensitivity analyses. Other studies have directly evaluated the effect of the leaf anatomical structure on spectral reflectance, particularly in the near infrared region. However, the relationship between N and the anatomical structure is unclear. In this study, we leveraged eLeaf, a ray tracing-based 3D rice leaf simulator, to establish relationships between leaf anatomical features and spectral properties, enabling us to replace N in the PROSPECT-4 model with measurable leaf anatomical parameters and develop the RSPECT model. The leaf thickness at minor vein, leaf thickness at bulliform cells, mesophyll thickness at minor vein, and distance between two minor veins could be used to predict N effectively. The RSPECT model achieved spectral simulation accuracy comparable to PROSPECT-4 and was more suitable for parameter inversion of the physical and chemical properties of rice leaves, with relative root mean square errors of 7.4 % for chlorophyll content, 5.6 % for equivalent water thickness, and 7.5 % for dry matter content. In conclusion, RSPECT improves radiative transfer modeling by integrating measurable anatomical features and provides a framework for extending this approach to other vegetation types.

Accurate and real-time monitoring true leaf area index (LAI) is an essential for assessing crop growth status and predicting yields. Conventional LAI inversion approaches have been constrained by insufficient data representativeness and environmental variability, particularly when applied across interannual variations and different phenological stages. This study presented a novel methodology integrating three-dimensional radiative transfer modeling (3D RTM) with knowledge-guided deep learning to address these limitations. We developed a knowledge-guided convolutional neural network (KGCNN) architecture incorporating 3D canopy structural physics, enhanced through transfer learning (TL) techniques for cross-temporal adaptation. The KGCNN model was initially pre-trained on synthetic datasets generated by the large-scale remote sensing scattering model (LESS), followed by domain-specific fine-tuning using 2021 field measurements, and culminating in cross-year validation with 2022-2023 datasets. Our results demonstrated significant improvements over conventional approaches, with the 3D RTM-based KGCNN achieving superior performance compared to 1D RTM implementations (PROSAIL + CNN + TL). Specially, for the 2022 dataset, the overall R2 increased by 0.27 and RMSE decreased by 2.46; for the 2023 dataset, the overall RMSE decreased by 1.62, compared to the PROSAIL ​+ ​TL method. Our method (3D RTM ​+ ​KGCNN ​+ ​TL) delivered superior LAI retrieval accuracy on the two-year datasets compared to LSTM ​+ ​TL, RNN ​+ ​TL, and 3D RTM ​+ ​RF models. This study also introduced an effective 3D scene modeling strategy that integrates scenarios representing the measured data range with additional synthetic scenes generated through random combinations of structural parameters. By incorporating detailed 3D crop structural information into the KGCNN network and fine-tuning the model with measured data, the approach significantly enhanced the model's adaptability to varying data distributions across different years and growth stages. This approach thus improved both the accuracy and stability of true LAI retrieval.