IntroductionSafeguarding the yield and quality of field crops against abiotic stresses is critical for large-scale agricultural production and precision agronomy. This study addresses the challenges of detecting concealed fruit stress and overcoming label scarcity in unmanned aerial vehicle (UAV) multispectral monitoring, using blossom-end rot (BER) in processing tomatoes as a case study.MethodsWe developed a leaf–fruit synergistic Roll and Color Disease Index (RCDI), integrating fruit incidence with visible canopy phenotypic features to characterize the combined canopy–fruit stress status associated with BER. A three-level screening strategy was used to identify the optimal spectral feature set. A Multi-model Collaborative Cyclic Self-Training (MCC-ST) framework was subsequently developed to address the limited availability of severity-labeled samples.ResultsThe combination of GRVI, NDVI, and SAVI was identified as the optimal spectral feature set, achieving stable within-dataset binary classification performance of approximately 97% in repeated cross-validation. Under 30 random-seed repeated stratified three-fold cross-validations, MCC-ST + DT and MCC-ST + RF achieved RCDI-based BER severity-grading accuracies of 85.00% ± 0.31% and 85.07% ± 0.42%, respectively. Compared with the corresponding original DT and RF models, MCC-ST improved repeated-validation accuracy by 16.19–4.09 percentage points.DiscussionThe RCDI helps bridge the observational gap between canopy signals and concealed fruit stress, while MCC-ST alleviates the bottleneck associated with label scarcity. The proposed approach provides a promising framework for crop abiotic-stress monitoring under limited-label conditions.

Traditional agricultural traceability systems often suffer from centralized data management, fragmented lifecycle records, weak tamper resistance, and limited scalability in multi-stakeholder supply chains. To address these issues, this study proposes a Hyperledger Fabric-based agricultural quality traceability framework that combines consortium blockchain, smart contracts, and an improved Group-Aggregation PBFT (GA-PBFT) consensus mechanism. The framework models key supply-chain stages, including cultivation, harvesting, processing, warehousing, and quality inspection, and uses modular smart contracts to standardize the registration, transfer, reception, processing, and lifecycle assessment of traceability data. To improve consensus efficiency, GA-PBFT integrates reputation-based node grouping with BLS aggregate signatures, reducing inter-node communication complexity from O(N2) to O(N) while preserving Byzantine fault tolerance. A smart contract-based cotton traceability prototype was implemented on Hyperledger Fabric to verify the feasibility of the proposed framework. Experimental results show that GA-PBFT maintains lower communication overhead and better scalability than traditional PBFT as the number of nodes increases. System benchmark tests further indicate that the average latency of write transactions is 0.09 s, whereas read-only transactions have near-zero latency, demonstrating that the prototype can support high-concurrency traceability queries and data uploads. The proposed framework provides a practical and extensible solution for trustworthy agricultural supply-chain traceability and offers technical support for digital quality supervision in the agricultural sector.

Cotton is a globally cultivated vital economic crop, highly susceptible to Verticillium wilt (VW), a disease caused by the fungus Verticillium dahliae. This infection leads to significant losses in biomass, lint yield, and fiber quality. Currently, the detection and control of VW in cotton primarily rely on visual inspection and fungicidal treatments, which are time-consuming, labor-intensive, and environmentally polluting. This study integrates hyperspectral imaging and chlorophyll fluorescence imaging techniques, while adopting a four-class classification (healthy, slight, moderate, and severe) for the accurate identification of slight infection samples—corresponding to the early disease stage—thereby facilitating the early detection of Verticillium Wilt (VW) in cotton via multi-source imaging fusion. By extracting hyperspectral data and chlorophyll fluorescence induction dynamic curves from cotton leaves with different VW infection grades, we investigate the phenotypic characteristic changes of cotton under varying VW infection levels. Using pixel-level, feature-level, and decision-level feature fusion strategies to integrate hyperspectral imaging and chlorophyll fluorescence imaging data, we identify cotton’s phenotypic characteristic information and construct a multi-input, single-output deep learning architecture based on such fusion strategies. This architecture amplifies the early phenotypic changes of VW-infected cotton and improves the timeliness and accuracy of VW detection and warning in cotton. The results show that compared with single imaging modeling, traditional machine learning and deep learning approaches using fusion strategies achieve overall improvements in detection accuracy and stability. Notably, the multi-source data feature-level fusion deep learning architecture with Resnet-1D as its main backbone exhibits significant enhancement. In this architecture, the detection accuracy for different VW infection levels in cotton reaches 99.57% on the training dataset and 97.88% on the testing dataset, respectively. The accuracy of early VW detection in cotton exceeds 99%. This research provides technical support for the green prevention and control of cotton VW, as it helps effectively reduce cotton yield losses while minimizing fungicide overuse, thus enabling precise regional spraying control. The code is available at https://github.com/TianyingYan/MISO.

Accurate monitoring of nitrogen nutrition is critical for optimizing cotton production. Traditional machine learning-based inversion models have limited effectiveness for precision monitoring. Multisource fusion models for small samples were developed in this study to achieve enhanced accuracy through fitting and data complementarity. Cotton plants subjected to different nitrogen treatments were investigated. A two-year pot experiment was conducted to collect main-stem leaf images to construct an image pretraining dataset for model transfer. In a field experiment conducted over one year, main-stem leaf data were collected using hyperspectral, chlorophyll fluorescence, and digital camera sources, thereby providing a multisource dataset for training monitoring models. Two architectures—a neural network (NN) and an interpretable deep forest (DF), which are suitable for small-sample spectral, fluorescence, image color, and texture-sequence features—were constructed to improve the accuracy of nitrogen content inversion. Additionally, a two-dimensional sliding-window processing method was introduced into the DF multigranularity scanning module, and a transfer-learning-based two-dimensional convolutional NN was employed to directly model small-sample two-dimensional images. Building upon the outcome, multilayer fusion models were constructed, with corresponding fusion strategies designed for homogeneous sequence inputs and heterogeneous image–sequence inputs. The results showed that NN and DF can effectively handle limited sample sizes and outperform traditional machine learning models. Among the fusion models, the optimal secondary decision-level fusion model achieved an R² of 0.926 on the independent test set, indicating good performance under small-sample conditions. This study provides a methodological reference for the precise monitoring of crop phenotypic parameters under small-sample conditions.

Nitrogen, an essential nutrient for cotton growth, requires accurate and timely assessment for effective fertilizer application. Despite advances in sensing technology and models, traditional machine learning monitoring models exhibit limited accuracy in assessing nitrogen content. Agronomic sample collection is challenging, and small datasets are unsuitable for conventional deep learning (DL) methods. Moreover, monitoring at specific time points cannot capture dynamic nitrogen changes within the crop, and time lags in decision-making can lead to mismatches between nitrogen supply and crop demand. Therefore, improving monitoring accuracy with small agronomic samples and effectively predicting future nitrogen content changes are crucial. Here, we used a hyperspectral technology to collect data, focusing on "Xinluzao53" cotton and established four nitrogen concentration gradients. Approximately 30 days after emergence, we conducted destructive and non-destructive sampling of the main stem leaves at regular intervals. To train the monitoring model, destructive sampling involved collecting hyperspectral data, followed by leaf cutting for nitrogen determination, whereas nondestructive sampling involved collecting hyperspectral data over time without leaf damage. We then constructed DL monitoring models suitable for small samples to estimate nitrogen levels. The optimal monitoring model was applied to non-destructive sampling, and the resulting nitrogen content time-series was cleaned and used as input for prediction models. The one-dimensional convolutional neural network monitoring model developed in this study achieved optimal accuracy, and the improved ensemble time-series prediction models demonstrated better predictive performance than single time-series models. These findings offer valuable insights for monitoring phenotypic parameters with limited sample sizes and predicting future changes.

Nitrogen is crucial for crop growth, development, yield, and quality. Traditional nutrition monitoring relies on single data sources; however, spatial coverage and information limitations hamper the accuracy of such monitoring methods. The recently developed unmanned aerial vehicle (UAV) remote sensing technology has emerged as an efficient and convenient method of crop nutrition monitoring, which allows the integration of data from sources, such as hyperspectral and digital images, resulting in comprehensive and multi‐angular insights. This study is aimed at enhancing crop monitoring accuracy by integrating multiple data types obtained at the UAV scale, using ‘Xinluzao 53’ cotton as an experimental subject. Nitrogen content was obtained via hyperspectral and digital imaging and the features of the two data sources analyzed by constructing four machine learning models: Ridge (RR), back-propagation neural network (BPNN), random forest (RF), and Bagging, which were integrated with the multilevel data fusion methods to obtain nutrition information. The results indicated optimal efficacy for RF together with the UAV ‘spectrum-image’ feature-level fusion framework, with a validation set R2 of 0.915 and RMSE of 1.562, while the optimal decision-level fusion framework was found to be Bagging, with a validation set R2 of 0.923 and RMSE of 1.488. UAV-based ‘spectral-image’ multilevel fusion frameworks were found to enhance the accuracy of monitoring, with the optimum decision-level fusion evaluation indices providing crucial theoretical support for precision agriculture in the future.

Cotton is one of most important economic crops in the world. Cotton yield has been significantly affected by frequent infestations of Verticillium wilt (VW). Currently, most detection methods for cotton VW are implemented based on the clearly visible symptoms, leading to delayed interventions and control. Therefore, early detection of cotton VW is crucial for minimizing economic losses. However, existing early detection methods for cotton VW face substantial challenges due to the subtle nature of early-stage symptoms and the limited availability of data, which result in considerable error. To address this, an early detection method for cotton VW by integrating Generative Adversarial Networks (GANs) with hyperspectral imaging technology was proposed, focusing primarily on cotton hyperspectral data augmentation. GANS-based spectral enhancement model (Spe-GAN) and GANS-based spatial-enhancement model (Spa-GAN) were developed to capture subtle early-stage symptoms of VW under limited data from both spectral and spatial perspectives. Compared to traditional machine learning methods (RF and SVM) and deep learning methods (LSTM and ResNet18), the proposed Spe-GAN and Spa-GAN achieved better detection performance, with accuracy rates of 94.52 % and 91.78 %, respectively. Moreover, this study also explored the underlying reasons for the superiority of the proposed method from various perspectives, further enhancing the model interpretability. The data augmentation method proposed in this study provided a new perspective and opened up possibilities for achieving the early detection of cotton VW and other plant diseases.

To address the limitations of traditional cotton leaf nitrogen content estimation methods, which include low efficiency, high cost, poor portability, and challenges in vegetation index acquisition owing to environmental interference, this study focused on emerging non-destructive nutrient estimation technologies. This study proposed an innovative method that integrates multi-color space fusion with deep and machine learning to estimate cotton leaf nitrogen content using smartphone-captured digital images. A dataset comprising smartphone-acquired cotton leaf images was processed through threshold segmentation and preprocessing, then converted into RGB, HSV, and Lab color spaces. The models were developed using deep-learning architectures including AlexNet, VGGNet-11, and ResNet-50. The conclusions of this study are as follows: (1) The optimal single-color-space nitrogen estimation model achieved a validation set R2 of 0.776. (2) Feature-level fusion by concatenation of multidimensional feature vectors extracted from three color spaces using the optimal model, combined with an attention learning mechanism, improved the validation R2 to 0.827. (3) Decision-level fusion by concatenating nitrogen estimation values from optimal models of different color spaces into a multi-source decision dataset, followed by machine learning regression modeling, increased the final validation R2 to 0.830. The dual fusion method effectively enabled rapid and accurate nitrogen estimation in cotton crops using smartphone images, achieving an accuracy 5–7% higher than that of single-color-space models. The proposed method provides scientific support for efficient cotton production and promotes sustainable development in the cotton industry.

Accurate monitoring of cotton water status is crucial for optimizing irrigation management and improving water-use efficiency in precision agriculture. UAV-based remote sensing offers high-resolution, flexible, and efficient data acquisition for agricultural monitoring, presenting significant potential for assessing crop water stress. However, existing approaches often treat spectral and texture features separately, overlooking their complementary nature across resolutions. This increases model complexity and reduces generalizability across phenological stages. To address these limitations, we propose a Dual-Cycle Cognitive Learning (DCCL) framework that integrates multi-resolution vegetation indices and texture features through a two-stage interpretable training pipeline. In the first stage, all extracted features are fed into a random forest model, and their contributions are quantified using SHapley Additive exPlanations (SHAP). The top 20 SHAP-ranked features are further refined using Recursive Feature Elimination (RFE) to select the 10 most informative features. These features are reintroduced into a pretrained model to form a distilled final monitoring model in the second stage, enhancing interpretability and cross-scale monitoring accuracy. Knowledge distillation further facilitates feature integration across different resolutions and growth stages, eliminating the need for manual feature engineering. Experiments conducted on real UAV datasets demonstrate the effectiveness of the proposed DCCL framework. It achieved a training R² of 0.9577 and an RMSE of 0.0026, while maintaining a cross-validation R² of 0.6514 on unseen datasets. In contrast, the baseline random forest model yielded a training R² of 0.9484 but a considerably lower cross-validation R² of 0.4268. These results confirm the improved robustness and generalizability of our approach under real-world field conditions. The DCCL framework offers a scalable, interpretable, and high-precision solution for UAV-based cotton water status monitoring, with significant potential to support sustainable irrigation strategies and intelligent crop management in large-scale agricultural systems.

Verticillium wilt (VW) is one of the most common and devastating diseases in cotton production, and early diagnosis is very important to alleviate the damage of VW. Recent studies have shown that early diagnosis and prevention of soil-borne diseases can be achieved by detecting spectral changes related to chlorophyll fluorescence and transpiration. However, there are no systematic studies to report the heterogeneity of photosynthetic characteristics and their spectral responses of plant leaves at the early stage of VW. In this study, the spatial heterogeneity characteristics in chlorophyll fluorescence of cotton leaves during the incubation period of VW were discussed, and the pixel-level inversion of the heterogeneity characteristics of leaf chlorophyll fluorescence was realized with hyperspectral imaging information, aiming to realize the early diagnosis of VW of cotton. The results showed that the chlorophyll fluorescence parameters Y(NPQ) (quantum yield of regulated energy dissipation) and NPQ/4 (non-photochemical quenching/4) values of cotton increased and the Y(II) (effective quantum yield of photosystem II) decreased significantly during the asymptomatic period of VW, indicating heterogeneity in photosynthetic capacity of leaves in the early stage of VW, i.e., VW developed from leaf margins to leaf center, and leaf margin was the area where chlorophyll fluorescence changed firstly. Furthermore, the multi-task learning model constructed with vegetation index and wavelet features accurately inversed the pixellevel heterogeneous characteristics of leaf Y(NPQ) and Y(II). The spectral information had the best inversion performance for the local heterogeneous regions of Y(II), with a classification accuracy of 85.6 %, a Kappa coefficient of 0.71, an r2 (coefficient of determination) of 0.66, and a RMSE (root mean square error) of 0.06. According to the inversion results of the local heterogeneous region of Y(II), the accurate diagnosis of early-stage VW was realized, with an accuracy of 87.4 % and a Kappa coefficient of 0.75. This study will provide a new method for the early prevention and control of VW.

The remote sensing-based estimation of the vertical attenuation coefficient K is of great significance to increase the accuracy of the estimation of crop canopy nitrogen vertical distribution by remote sensing technology. However, the multiple-angle information is susceptible to interference from specular reflection, which greatly limits the accuracy and stability of K estimation. In this research, the cotton canopy multiple-angle spectrum and polarization were acquired. Then, the spectral reflectance in the red- and blue-edge regions were combined to construct multiple-angle vegetation indices (MAVIs) using diffuse reflection component and total reflectance separately, and the MAVIs were used to estimate K. The estimated K was used to invert the nitrogen content of different vertical layers (upper, middle, and lower layers) of cotton canopy. Finally, the inversion results were compared with the inverted nitrogen content by the constructed multi-angle vegetative indices. The results showed that removing the specular reflection component from the total reflectance significantly increased the K estimation accuracy. The K estimation accuracy of MAVIs was higher than that of single-angle vegetation indices. Among the MAVIs, MNDVIR-B (-30,45,45,45,0) had the highest K estimation accuracy, and the R2 for the different growth season was in the range of 0.816-0.871. The estimated K by the MNDVIR-B (-30,45,45,45,0) accurately inverted the nitrogen content of different vertical layers of cotton canopy, which was significantly higher than the R2 of the estimation of different-layer nitrogen directly using the MAVIs. This study will provide a new method for accurately monitoring the vertical nitrogen status of crop canopy.

Cotton is susceptible to Verticillium wilt (VW) during its growth. Early and accurate detection of VW can facilitate targeted pesticide treatment and reduce the potential spread of the disease. However, accurately detecting VW in cotton before symptoms appear (the asymptomatic period) after infection by Verticillium dahliae remains challenging. This study proposes an early detection method for cotton wilt disease using hyperspectral imaging and recurrence plots (RP) combined with machine learning techniques. First, spectral curves were collected and analyzed under three conditions of cotton plants: healthy, asymptomatic, and symptomatic. Then, the one-dimensional spectral curve was transformed into two-dimensional recurrence plots to enhance the detail differences in the original spectral curve of cotton plants in various states. Hyperspectral recurrence plots contain rich texture information; fifteen texture features were extracted from the spectral recurrence plots using the Gray-Level Gradient Co-occurrence Matrix (GLGCM). Eleven of these texture features showed a strong correlation with the class labels of the cotton plants. In order to reduce redundant information between features, principal component analysis (PCA) was used to extract the first five principal components, which explained 99.02% of the information from the 11 features. The final principal component dataset was then input into KNN, SVM, ELM, and XGBoost classifiers to assess the accuracy of early detection of VW in cotton. The results showed that the XGBoost model, based on the first five principal components obtained from the texture features, achieved accuracy, precision, recall, and F1-score of 96.3%, 95.6%, 96%, and 95.8%, demonstrating a high classification capability. The results of this study confirm the feasibility of converting spectral curves into recurrence plots and extracting image texture features for the accurate identification of VW in cotton during the asymptomatic period. This method also provides a new strategy for early disease detection of cotton and other plants in the future.

Cotton aphid ( Aphis gossypii Glover) severely impacts cotton yield and quality, necessitating effective detection and control measures. Traditional manual detection methods are inefficient, highlighting the need for rapid and accurate detection. In order to realize rapid and accurate cotton aphid detection, we proposed a novel attention mechanism named Bi-Enhanced Attention Mechanism (BEAM), aiming at improving the performance of the YOLOv8-s model. Furthermore, we employed a domain-adaptive transfer learning strategy by pre-training our enhanced network model on a public forestry pest dataset and fine-tuning it on a custom-built cotton aphid dataset. In this study, we evaluate our approach using the mean Average Precision (mAP) at different Intersection over Union (IoU) thresholds. Experimental results demonstrated that our approach achieved excellent performance in detecting cotton aphids. Specifically, our approach, which includes an enhanced YOLOv8-s model with a bi-enhanced attention mechanism and a domain adaptive transfer learning strategy, achieved an mAP of 58.1% at an IoU threshold range of 0.5 to 0.95 (mAP@0.5:0.95), 95.4% at an IoU threshold of 0.5 (mAP@0.5), and 64.8% at an IoU threshold of 0.75 (mAP@0.75). Compared to the baseline method, which utilizes the original YOLOv8-s model with standard training procedures, there was an improvement of 4% in mAP@0.5:0.95, 1.3% in mAP@0.5, and 8.1% in mAP@0.75. This research introduces a novel method for accurate detection of cotton aphids in the field, which is crucial for effective pest management and timely intervention.

•Soil TN estimation accuracy using vis-NIR spectra is higher than pXRF spectra.•Feature selection & geographical stratification improve soil TN estimation accuracy.•CARS is the optimal feature selection method for multi-sensor data fusion models.•SO-PLS fusion method increases model accuracy by fully using multiple sensor data. Soil TN estimation accuracy using vis-NIR spectra is higher than pXRF spectra. Feature selection & geographical stratification improve soil TN estimation accuracy. CARS is the optimal feature selection method for multi-sensor data fusion models. SO-PLS fusion method increases model accuracy by fully using multiple sensor data.

IntroductionRapid and accurate estimation of leaf area index (LAI) is of great significance for the precision agriculture because LAI is an important parameter to evaluate crop canopy structure and growth status.MethodsIn this study, 20 vegetation indices were constructed by using cotton canopy spectra. Then, cotton LAI estimation models were constructed based on multiple machine learning (ML) methods extreme learning machine (ELM), random forest (RF), back propagation (BP), multivariable linear regression (MLR), support vector machine (SVM)], and the optimal modeling strategy (RF) was selected. Finally, the vegetation indices with a high correlation with LAI were fused to construct the VI-fusion RF model, to explore the potential of multi-vegetation index fusion in the estimation of cotton LAI.ResultsThe RF model had the highest estimation accuracy among the LAI estimation models, and the estimation accuracy of models constructed by fusing multiple VIs was higher than that of models constructed based on single VIs. Among the multi-VI fusion models, the RF model constructed based on the fusion of seven vegetation indices (MNDSI, SRI, GRVI, REP, CIred-edge, MSR, and NVI) had the highest estimation accuracy, with coefficient of determination (R2), rootmean square error (RMSE), normalized rootmean square error (NRMSE), and mean absolute error (MAE) of 0.90, 0.50, 0.14, and 0.26, respectively. DiscussionAppropriate fusion of vegetation indices can include more spectral features in modeling and significantly improve the cotton LAI estimation accuracy. This study will provide a technical reference for improving the cotton LAI estimation accuracy, and the proposed method has great potential for crop growth monitoring applications.

Accurate monitoring of nitrogen nutrition is crucial for improving cotton yield and quality, as well as the ecological environment. The mainstream method for monitoring nutrition is to establish traditional machine learning (ML) models using a single data source. However, this approach has limitations such as limited access to feature information, model-fitting problems, and limited generalization. Deep learning (DL), on the other hand, has shown promise in complex nonlinear modeling tasks due to its flexible structure. However, it has its limitations, such as the fact that agronomic sample collection and testing are usually labor- and material-intensive, resulting in sample sizes that are too small to meet its training conditions. Therefore, there is an urgent need for DL models that can effectively integrate features from multiple data sources and accurately monitor crop nitrogen content, especially in scenarios with small samples. In this study, we conducted indoor pot experiments using the cotton variety Xinluzao 53 and subjected it to six nitrogen treatments. The data sources for our analysis included hyperspectral and digital images of the cotton leaves. To enhance representation learning capabilities, we enriched the multi-class base learners within each layer of the deep forest (DF) model and introduced skip connections. These enhancements improved the quality of inversion for both hyperspectral and digital image datasets. We then developed image-spectral fusion models, which combined the DF structure with stacking ensemble learning. Our focus was on three levels of fusion: feature-level fusion, decision-level fusion, and secondary decision-level fusion. This approach aimed to further enhance the accuracy and stability of nitrogen content inversion. The DF model satisfied the training condition for small samples. Compared to traditional ML algorithms and the original DF algorithm, the improved DF model achieved an increase in validation set R2 of 13.4–28.5% and 10.9–14.9%, respectively. These findings highlight the enhanced accuracy and stability of the improved DF model. Additionally, compared to the optimal inversion model using two single data sources, the “Image-Spectral” three-level fusion models exhibited improvements in validation set R2 of 8.6–9.3%, 10.5–11.2%, and 11.8–12.5% for feature-level, decision-level, and secondary decision-level fusion, respectively. The improved DF and three-level fusion model collectively contributed to the increased accuracy of cotton nitrogen content inversion. Among these models, the secondary decision-level fusion model demonstrated the most marked improvement. This methodology provides valuable insights into monitoring crop phenotypic parameters in situations with limited sample sizes.