Near-infrared spectroscopy (NIRS) has been gained extensive application in the field of flour adulteration detection due to its rapid and non-destructive analytical capabilities. Nevertheless, conventional methodologies merely offer simplistic simulations of adulteration scenarios, thereby failing to accurately represent the intricate conditions of multi-type and multi-source adulterants in actual production and circulation. To overcome this limitation, the present study constructed a multi-source composite adulteration data system: based on a handheld NIR spectrometer, multiple types and brands of wheat flour and common adulterants (cassava flour, gypsum powder, talcum powder) were systematically collected, and samples were selected for modeling and heterologous testing according to the weighted similarity (WS) criterion, preserving spectral diversity while controlling experimental costs. In conjunction with the proposed lightweight multi-task deep learning (DL) model Adulterated Flour Unmixing Net Multitask (AFUNet-MT), end-to-end joint analysis of adulteration types and component abundances was achieved. Compared with traditional machine learning (ML) methods and State-ofthe-art (SOTA) DL models, AFUNet-MT exhibited superior comprehensive performance in 5-fold cross-validation: the classification accuracy reached 0.9816 f 0.0036, and the overall R2 value of component abundance prediction reached 0.9564 f 0.0059. In heterologous testing composed of edge brands, AFUNet-MT still maintained a classification accuracy of 0.9560 f 0.0021 and an abundance estimation R2 value of 0.8992 f 0.0050, with a residual predictive deviation (RPD) value of up to 5.7695, fully demonstrating its generalization ability and stability under cross-brand and cross-category adulteration. Regarding computational performance analysis, the proposed model exhibited low memory and high sample throughput per unit time, thereby furnishing practical technical support for embedded deployment with handheld NIR devices and rapid screening in agricultural field applications.

Biomarkers are vital for assessing food quality, but real-time and continuous monitoring of internal metabolic dynamics in plant leaves remains challenging for electrochemical sensors, primarily due to the extremely limited sap content. Herein, an integrated electrode-electrolyte sensing platform is developed for real-time detection of indole-3-acetic acid (IAA). The sensing interface features laser-induced graphene electrodes modified with MXene and molecularly imprinted polymers for selective recognition of IAA, combined with a gelatin-based hydrogel electrolyte containing sucrose, glycerol, and KCl. This semi-solid hydrogel provides adhesion, hydration, and conductivity, replacing volatile liquid electrolytes. The platform demonstrates high specificity and sensitivity toward IAA in the range of 5–200 μM with a detection limit of 0.137 μM. It successfully achieves continuous, real-time monitoring of endogenous IAA in lettuce for up to 72 h, revealing a clear diurnal rhythm. This strategy advances plant physiology research and supports data-driven crop management.

The performance of electrochemical sensors is prone to signal degradation caused by biofouling in complex biological fluids. Overcoming these biofouling remains a critical hurdle for their large-scale commercial applications. In this work, the antifouling effect of three antifouling membranes were compared using indole-3-acetic acid (IAA) in lettuce as sample. The Polydopamine-poly (sulfobetaine methacrylate) (PDA-PSBMA) membrane demonstrated exceptional resistance to IAA oxidation products and other contaminants. To further enhance the accuracy of the sensors, six Machine Learning (ML) algorithms were employed to predict post-contamination standard curves of the sensor after multiple uses in real samples. The Random Forest (RF) model exhibited optimal predictive performance, enabling effective calibration of current data of IAA from lettuce samples at different growth stages (mature and vegetative). After the application of the antifouling membranes helped by the ML methods, the RSD values of the sensor in lettuce samples decreased from 21.93% to 4.16% (mature stage) and from 27.65% to 6.17% (vegetative stage). Our work provides an effective technical approach that ensures the accuracy of electrochemical sensor even after repeated use in complex biological samples, which represents a significant, data-driven enhancement over traditional antifouling strategies. This approach can be readily adapted by other researchers to extend its utility to a broader range of biological samples or analytes.

Saline-alkali stress is a major abiotic constraint on wheat growth and productivity, creating a need for nondestructive, sensitive, and physiologically interpretable phenotyping methods for fine-grained stress detection. In this study, we propose a dual-tower deep-fusion framework that integrates low-field nuclear magnetic resonance (LF-NMR) relaxometry with proton density weighted imaging (PDWI) texture analysis for the classification of saline-alkali stress levels in wheat seedlings. Specifically, 1D-T2 and 2D-T1-T2 relaxation spectra were combined with 3D-PDWI-derived grey-level co-occurrence matrix (GLCM) texture features to jointly characterise water-state dynamics and spatial structural heterogeneity. Based on an SE-ResNet backbone, early fusion, midlevel fusion, and dual-tower late fusion strategies were systematically compared under a unified evaluation protocol. The results showed that LF-NMR captured the conversion of free water into semi-bound and bound water, together with disruption of cellular compartmentalisation, whereas PDWI texture features reflected the progression of tissue structural damage from a spatial perspective. These multimodal signals were consistent with physiological and biochemical measurements. Among all models, the proposed DT-LF-SE-ResNet achieved the best test accuracy of 95.93 %, exceeding the single modality baselines by 35.00 % and 30.37 %, and clearly outperforming early fusion (71.43 %) and mid-level fusion (87.56 %). Feature importance analysis and Grad-CAM further indicated that the performance gain arose from the synergistic representation of water population redistribution and increased microstructural heterogeneity. These findings demonstrate the mechanistic and predictive advantages of dual-tower multimodal fusion and provide an explainable phenotyping framework for fine grained crop stress assessment.

In maize seed production, an accurate assessment of the pollination status of maize tassels enables efficient artificial pollination and enhances grain setting rates. With the expansion of large-scale seed production and annual increases in labour costs, information technology solutions are urgently required for monitoring pollination and growth stages. To address this challenge, machine vision technology was employed in this study to identify the pollination status of maize tassels, and a pollination status-detection model for maize tassels was proposed based on an improved YOLOv8s. The backbone of this model incorporated the Dilated Reparam Block, which enhances the feature-extraction capability through dilated convolution, thereby better capturing detailed information. The neck of this model integrates the Bidirectional Feature Pyramid Network, which efficiently fuses features of different scales to further enhance the detection capability of the model for multi-scale targets. The results showed that the proposed model was better than the control, with an AP50 of 95.6% for pollinating tassel detection, reducing the model size by 38.3% and the number of parameters by 40% compared to those with the original YOLOv8s. Moreover, the detection speed surpassed 100 frames per second, which meets the real-time requirement of the actual detection task. In the maize pollination stage-detection experiment, the proposed model achieved an R2 of 0.87 in predicting the pollinating tassel ratio. Based on the pollinating tassel ratio of male parent maize in seed production fields, researchers can precisely determine whether maize has entered the peak pollination stage. This study provides reliable technical support for decision-making in the assisted pollination of maize.

Nitrogen is an essential nutrient for plant growth and is predominantly present in soil and nutrient solution as ammonium (NH4+) and nitrate (NO3-). In this work, novel all-solid ion-selective electrodes were developed for the detection of NH4+ and NO3- using chitosan and black phosphorus combined with ferric oxide magnetic nanoparticles as the solid contact. Electrochemical characterization revealed near-Nernstian responses, with the NH4+-selective electrode exhibiting a slope of 57.8 +/- 1.5 mVdecade-1 and the detection limit of 1 x 10(-5.4 )M, while the NO3--selective electrode showed a slope of -58.5 +/- 0.5 mVdecade(-1 )and the detection limit of 1 x 10(-6.2) M. Surface modification improved hydrophobicity, with the contact angle increasing by 30.8 degrees compared with the unmodified electrodes. To enhance measurement accuracy, a mixed improved dung beetle algorithm-optimized backpropagation neural network was employed to establish concentration prediction models. The resulting models achieved mean square errors of 0.0214 for NH4+ and 0.1933 for NO3-, with recovery errors controlled within 3%. These findings demonstrate that the proposed electrodes, coupled with intelligent modeling, provide a reliable approach for long-term, online, and precise monitoring of nitrogen in agricultural and environmental systems.

Gluten content and strength are key indicators of whole wheat flour processing quality and their rapid non-destructive detection is crucial for quality screening and wheat variety breeding. This paper proposed a covariance guided stratified cluster sampling (CGSCS) algorithm for partitioning multi-indicator small sample sets. Furthermore, a novel analytical framework combining a surface model with machine learning algorithms is presented to simultaneously detect gluten content and strength indicators. The results showed that the surface model combined with CARS-LASSO achieved optimal prediction, with test $R^2$ values of 0.98, 0.93, 0.89, 0.84, and 0.83 and RPD values of 6.73, 3.77, 3.08, 2.47, and 2.40 for WGS, GI, DGC, WGC, and WGC (14\% MB), respectively. The surface model constructed in this study, incorporating the core principles of machine learning algorithms, provides a novel approach for the quantitative and rapid simultaneous detection of gluten characteristics in whole wheat flour.

Saline-alkali stress significantly impacts wheat growth and yield, making it essential to understand its effects on wheat seedlings. This study established stress treatments of varying intensities and employed Low-Field Nuclear Magnetic Resonance (LF-NMR) and Non-invasive Micro-test Technology (NMT) to measure moisture dynamics and net ion fluxes in seedling. Grey prediction models and machine learning algorithms were applied to predict moisture parameters and net K+ and Na+ fluxes. Results indicated that saline-alkali stress inhibited seedling growth and altered water distribution patterns, characterised by decreased bound water and increased free water content. As stress intensity increased, peak relaxation times for bound, semi bound, and free water fractions all declined in the transverse relaxation spectra, indicating decreased water molecule mobility, enhanced binding, and intensified confinement. Furthermore, the distinction between bound and semi-bound water peaks diminished progressively under moderate and severe stress. Ion flux measurements revealed that saline-alkali stress reduced K+ absorption capacity and decreased Na+ efflux in wheat roots, with Na+ competing for K+ binding sites, leading to Na+ toxicity and disruption of physiological mechanisms. Model performance comparisons demonstrated that the Grey Model (GM (1,1)) achieved superior predictive accuracy for moisture parameters, with Concordance Correlation Coefficient (CCC) of 0.98 under severe stress. The Variational Mode Decomposition-Long Short-Term Memory (VMD-LSTM) model exhibited exceptional capability for predicting net ion fluxes, particularly for Na+ under severe stress (CCC = 0.96). This research enhances understanding of physiological responses of wheat seedlings under saline-alkali stress and demonstrates the applicability of predictive models for assessing moisture and ion dynamics.

Moisture content of maize is a key indicator in maize production, processing and storage. In order to realize the rapid and accurate detection of maize grain moisture content, the original output signals such as input/output voltage and phase angle at 26 frequency points within the range of 10 kHz to 1.25 MHz were measured by using an oscilloscope, a signal generator and a self-made parallel plate capacitor. The six types of electrical parameters of maize under different moisture content conditions, namely capacitance, resistance, dielectric constant, dielectric loss factor, dissipation factor and quality factor, were calculated. Feature frequencies were selected using the Successive Projections Algorithm (SPA) and Competitive Adaptive Reweighted Sampling (CARS) algorithm, and modeling was performed using the Support Vector Regression (SVR) and Random Forest Regression (RF). The experimental results showed that the coefficient of determination R 2 of the constructed SPA-SVR model reaches 0.989, and the root mean square error (RMSE) is 0.93. This method broke through the limitations of the existing detection technology that underutilizes the range of the frequency domain, and by establishing a nonlinear relationship between the electrical characteristics of the wide-frequency domain and the moisture content, it provided an effective means of rapid and accurate detection of the maize grain moisture content.

Real-time in situ monitoring of magnesium ions (Mg2+) concentration in plants is essential for understanding ion transport and advancing precision agriculture. However, real-time in situ monitoring using all-solid-state ion-selective electrodes (ASS-ISEs) remains challenging because a water-layer tends to form at the interface between the sensing membrane and the metallic substrate, which leads to substantial potential drift. Here, we present an interfacial-engineering strategy for the microneedle electrode fabrication process, along with a method that enables accurate and stable in situ monitoring of Mg2+ in the stem xylem of tomato seedlings. Through comparative evaluation of candidate solid-contact materials, a graphene composite waterborne coating was identified as the optimal solid contact; it suppresses water-layer formation primarily by enhancing interfacial compactness rather than hydrophobicity alone. The optimized electrode delivered a near-Nernstian slope of 29.00 ± 0.20 mV/dec, a linear range of 1.0 × 10-5-1.0 × 10-1 mol/L, a detection limit of 1.0 × 10-6.97±0.06 mol/L, a low drift of ∼0.3 mV/h, a charge-transfer resistance of 0.74 MΩ, and a storage lifetime of ∼13 months. Under both magnesium-deficient and high-magnesium stress conditions, the in situ monitoring data agreed well with the trends measured by inductively coupled plasma optical emission spectrometry (ICP-OES), confirming the reliability of the electrode. Overall, this work offers mechanistic insight into water-layer suppression in ASS-ISEs and highlights the potential of microneedle electrodes as a versatile platform for smart agriculture and plant science.

Rutin, a natural flavonoid with anti-inflammatory and antioxidant properties, is widely found in plants and foods. To meet the demand for in situ detection of Ru in soft plant tissues, in this paper, a thin laser-induced graphene (LIG) electrode was prepared on polyimide (PI) and transferred to Ecoflex substrate, obtaining LIG/Ecoflex electrodes with good flexibility and tensile resistance. To further improve the electrochemical catalytic performance, nanocomposites of hafnium diselenide (HfSe2), carboxylated single-walled carbon nanotubes (COOH-SWCNT), carboxylated graphene (COOH-GR) and nafion were also modified on the surface of LIG by a one-step dropping method. The HfSe2-COOH-SWCNT-COOH-GR-nafion/LIG/Ecoflex sensor could detect Ru in the range of 1 μM-700 μM under different pH values (4.5-7.4). It was also successfully used to in situ detection of Ru content in tomato leaves under different salt stress. The as-prepared sensor has important practical application prospects in the monitoring of plant physiological information in situ.

Serine (SER) is a functional amino acid that promotes plant growth and development, and plays an important role in plant stress response. In this paper, an ultrasensitive electrochemical sensor based on molecularly imprinted method (MIP) has been constructed to detect SER in plants in situ and in vivo. Composite material of titanium carbide (MXene), zeolitic imidazolate frameworks (ZIF-8) and thionin (Thi) was used to improve the conductivity and electroactive area of the sensor. Dopamine (DA) was used as the functional monomer to form the MIP. The prepared MIP/Thi-ZIF-8-MXene/SPE sensor showed an ultra-sensitive SER detection response with a detection range of 1 nM-4 mM (R2 = 0.995) and a detection limit of 0.69 nM (S/N = 3). It was also used to detect free SER in lettuce leaves under salt stress, which verified the practicality of the sensor. The MIP-based SER sensor has important application prospects for detecting the physiological status of plants in situ.

Ion-selective electrodes (ISEs) are pivotal tools for real-time, non-destructive monitoring of ionic dynamics in living plants, addressing key challenges in agriculture, plant physiology, and environmental science. This review presents recent advancements in ISE-based in vivo detection technologies, with an emphasis on sensor architectures tailored for plant tissues, including screen-printed planar electrodes, flexible electrodes, microneedle electrodes, and microfluidic devices. These systems enable precise in situ quantification of essential ions and trace elements, providing valuable insights into fundamental physiological processes such as nutrient uptake, stress responses, and signal transduction. Building on current ISE fabrication techniques, the review is aimed at developing a more cohesive theoretical framework for their application in plant systems. Future directions focus on synergistic integration of wearable plant sensors to build comprehensive real-time monitoring systems, which could advance understanding of plant–environment interactions and help address global food security challenges.

Maize is one of the most significant food crops in the world, and the vigor of maize seeds is a crucial indicator of seed quality. Therefore, it is of paramount importance to accurately and non-destructively detect the vigor of single maize seeds. In this study, hyperspectral images of the endosperm side and embryo side of a single maize seed were collected, and the feasibility of using hyperspectral imaging for vigor detection of single maize seeds was investigated. Due to the differences between the two sides of the maize seed spectra, four data fusion proposals (Mean, Concat, Stack, and Parallel) were designed and utilized to establish single maize seed vigor detection models in order to effectively utilize the spectral data from both sides of the maize seed. Meanwhile, feature engineering methods were used to assist modeling. The findings indicate that data fusion exhibits a superior capacity for the detection of single maize seed vigor when compared to using single-side spectral data. Additionally, the convolutional neural network demonstrates remarkable capabilities in feature extraction and resilience to noise. Feature engineering can further enhance the model performance. Suitable preprocessing algorithms can be used to reduce noise in the original spectra and alleviate the problem of noise amplification by the data fusion approaches. Characteristic wavelength extraction can eliminate redundant information in the original spectra and reduce the parameters of the models. The experiment results demonstrated that the fusion of spectra from both sides of the seed can be successfully used for single maize seed vigor detection and provided a potential method for accurate quality detection of seeds.

Strigolactones (SLs) play a crucial role in regulating plant growth and development. However, the soaking concentration significantly affects the growth of wheat seeds. Currently, there is a lack of rapid and accurate detection methods for this purpose. The objective of this study is to develop a rapid and accurate detection method for wheat seeds treated with varying concentrations of SLs, utilizing dual-view hyperspectral data fusion combined with deep learning techniques. Wheat seeds were soaked in different SLs concentrations, and hyperspectral data were collected from both the embryo and endosperm surface. The spectral data were preprocessed using a combination of Savitzky-Golay (SG), Second Derivative (ddA), and Multiplicative Scatter Correction (MSC). To effectively utilize the spectral data from both sides of the seeds, Parallel, Concat, and Stack data fusion strategies were employed. Detection was performed using Self-Built Convolutional Neural Network (SCNN), Adaptive Boosting (AdaBoost), and Gradient Boosting Decision Tree (GBDT) models. Results showed that the SG-MSC preprocessing combination demonstrated the best performance across all models. Compared to single-view spectral data, dual-view data improved the detection performance of the models. Furthermore, the Stack fusion strategy effectively avoided information redundancy and loss when processing dual-view data, outperforming both Concat and Parallel fusion strategies. The SCNN-SG-MSC-Stack model is the optimal model, achieving Accuracy, Precision, Recall, and F1-score values of 99.26%, 99.27%, 99.26%, and 0.99, respectively. This study demonstrates that combining dual-view hyperspectral data fusion and deep learning provides an efficient and reliable method for detecting different SLs concentrations in seed soaking, offering new insights for rapid evaluation.

In situ detection of plant ion signals faces technical limitations in terms of real-time capability, minimal invasiveness, and data analysis. Therefore, the development of sensors for in vivo plant detection and construction of time-series prediction models to analyze the dynamic patterns of ion concentrations in plants are imperative. This study presents a microneedle electrode system for potassium ion (K+) sensing, which is applied to real-time in situ detection in lettuce. The microneedle ion-selective electrodes (ISEs) fabricated herein exhibited a rapid potentiometric response (within < 15 s), with concentration responses adhering to the Nernst equation. During in vivo plant detection, the system captured instantaneous ion-signal changes upon exogenous application without influencing subsequent plant growth. This study demonstrates the pioneering application of time-series prediction (nonlinear autoregressive neural network model) to analyze in vivo K+ signals in lettuce, accurately forecasting ion concentration dynamics over time and identifying the transition pattern from signal fluctuation to stabilization. The integration of microneedle ISE-based in situ plant monitoring with time-series prediction represents a crucial and reliable approach to agricultural sensor innovation, providing a novel paradigm for precision agriculture and plant stress response research.

Glutathione S-transferases (GSTs) are essential multifunctional enzymes. In the face of abiotic stresses such as drought and heavy metal exposure, plants utilize GSTs for detoxification and antioxidant defense, as these enzymes facilitate the conjugation of glutathione (GSH) with toxic compounds. Specific details of this process, however, remain unknown. This study identified 118 Avena sativa GST (AsGST) genes within the A. sativa genome and classified them into five subfamilies: Tau, Phi, Zeta, Lambda, and EF1Bγ. Phylogenetic analysis revealed that AsGSTs exhibit significant similarity to corresponding GST categories in Arabidopsis thaliana and Oryza sativa, indicating a possible common ancestor. Gene structure and conserved motif analysis demonstrated that AsGST genes within the same subfamily shares similarities in the number and positioning of exons and introns, as well as in motif composition, suggesting that these genes may perform analogous biological functions in A. sativa. The promoter regions of the identified genes are enriched with various cis-acting elements that play roles in plant growth and development, stress response, and hormone signaling. Transcriptomic analysis and real-time quantitative PCR (RT-qPCR) validation indicated that the expression of four AsGST genes (AsGSTU12, AsGSTU13, AsGSTU14, and AsGSTU15) was significantly up-regulated in the roots of A. sativa under both PEG-induced drought stress and CdCl2-induced cadmium stress. These genes likely regulate reactive oxygen species (ROS) levels by catalyzing their scavenging through glutathione (GSH) substrates, and may also participate in ABA signaling and the maintenance of osmotic homeostasis. Under cadmium stress, these genes may mitigate cadmium toxicity by enhancing the chelation and sequestration of cadmium via GSH or through its compartmentalization, as evident from the subcellular localization studies. This study systematically described the GST gene family in A. sativa, characterized its expression patterns and potential functions in response to drought and cadmium stress, and confirmed the essential role of the AsGST gene family in mediating stress responses. The findings enhance our understanding of the mechanisms underlying stress tolerance and offer valuable genetic resources for breeding stress-tolerant A. sativa. The work also provides a theoretical framework and identifies gene targets for the development of stress-resistant A. sativa varieties.

Electrochemical sensors are devices that convert chemical signals into electrical signals, having been widely applied in various fields. However, traditional electrochemical sensors are prone to interference from various issues, which leads to inaccurate measurement results. The development of artificial intelligence (AI) technologies offers new approaches to address these issues. Among these, machine learning (ML) techniques can analyze large volumes of sensor data, identify complex patterns and relationships, and thereby enhance the accuracy and stability of sensors. This paper provides a review of the latest research over the past five years on ML applications addressing challenges such as nonlinear sensor signal relationships, low-concentration accuracy, signal drift, and interference resistance. It also summarizes the performance of various algorithms in different application scenarios. Finally, the paper discusses the challenges faced by ML technologies in improving the accuracy of electrochemical sensors and outlines future development directions.