With the digital and intelligent development of animal husbandry, individual pig identification has become a core link in precise breeding. Traditional methods such as ear tagging and DNA identification are difficult to meet the needs of modern breeding due to limitations including harm to pigs, high costs and poor operability. As an efficient and accurate biometric technology, pig face recognition has gradually become a research focus in this field and demonstrated great potential in animal husbandry. Based on the theory of deep learning, this paper first constructed a high-quality pig face dataset, then carried out structural reconstruction and refined tuning of key parameters for the basic VGG16 and ResNet50 models, and compared and analyzed their performance. The experimental results show that the combination of a high-quality dataset, model structure optimization adapted to task requirements and targeted adjustment of key parameters has significantly improved the performance of the pig face recognition model, with the optimal model achieving a recognition accuracy of $\mathbf{9 5. 6 \%}$ on the validation set. The research results provide new ideas and methods for the application of deep learning models in the agricultural field, and offer experimental references for the application of deep learning in individual livestock and poultry identification.

Deubiquitinating enzymes (DUBs) play essential roles in diverse plant biological processes, yet the ovarian tumor proteases (OTUs), a major DUB subfamily, have not been systematically characterized in wheat, and their functions in grain development remain unclear. Here, we identified 49 OTU genes (TaOTUs) in the hexaploid wheat genome and classified them into four subfamilies based on phylogenetic relationships, with nomenclature assigned according to homology. TaOTU6-7B was highly expressed during early and mid-grain development and was responsive to gibberellin and jasmonic acid. Its expression differed significantly between large-grain wheat Pindong34 (PD34) and small-grain wheat MY11847 at 7 and 11 days after flowering. To elucidate its function, we used CRISPR/Cas9 to generate loss-of-function mutants by knocking out the three homoeologs (TaOTU6-7A, -7B, and -7D). These mutants exhibited significantly increased grain width and weight relative to wild type. Moreover, TaOTU6-7B directly interacted with TaUBC13, whose expression was markedly elevated in the TaOTU6 knockout background, suggesting that TaUBC13 may positively regulate wheat grain size. Collectively, this study establishes the TaOTU gene family in wheat, reveals TaOTU6 as a negative regulator of grain width and weight, and provides valuable genetic resources and a theoretical foundation for high-yield wheat breeding.

Lyciumbarbarum is a valuable medicinal and edible crop, yet pesticide residues constrain its industrial development. Conventional detection methods are time-consuming and low-throughput, necessitating rapid, nondestructive alternatives. Using abamectin as a model pesticide, residue samples were prepared under natural conditions. Hyperspectral imaging (HSI) and volatile detection (VTD) were integrated to construct a multimodal nondestructive dataset. Two-dimensional correlation spectroscopy (2D-COS) combined with variable selection identified key spectral and volatile features. Ground-truth abamectin concentrations were determined by LC-MS/MS, enabling creation of a standardized spectral-volatile dataset. To address multimodal data structures and biological characteristics, a feature-decision fusion model (FDF-MT) was proposed, comprising a dual-branch feature extraction architecture with an attention mechanism. A shallow-deep fusion module enabled decision-level integration of optical and volatile responses for the classification of abamectin residue levels. Results showed significant differences between treated and control samples in spectral reflectance and volatile compounds. Combining 2D-COS with successive projection algorithm (SPA) selected 160 hyperspectral bands, accounting for 71.43% of the total hyperspectral bands, while uninformative variable elimination (UVE) identified 120 volatile features. The proposed FDF-MT achieved 96.79% accuracy, surpassing feature-level fusion by 1.61%. This decision-level fusion framework exploits cross-modal complementarity for rapid, nondestructive pesticide residue detection in L. barbarum.

To meet the agronomic requirements for multiple-seed-per-hill sowing of millet, broomcorn millet, and rapeseed under plastic-film mulching in Northwest China, an air-suction precision seed metering device was designed and evaluated through theoretical analysis and bench testing. The device integrates seed agitation, suction, staged cleaning, conveying, unloading, and placement, and is equipped with ten hill-forming units with a hill spacing of 16 cm. The airflow field and seed motion during operation were theoretically analyzed, and seed trajectories were captured using a 300 fps high-speed imaging system at an operating speed of 3 km/h. By adjusting the inlet negative pressure and suction-hole diameter, the preliminary stable operating ranges were determined. The results showed that the observed processes of seed agitation, adsorption, cleaning, unloading, and detachment were consistent with the proposed seed-motion model. The stable negative-pressure ranges were 0.61-3.54 kPa for millet, 0.83-4.15 kPa for broomcorn millet, and 0.74-3.59 kPa for rapeseed, while the corresponding suction-hole diameter ranges were 0.6-1.2 mm, 0.8-1.4 mm, and 0.7-1.2 mm, respectively. In 1000-hill bench tests, the qualified rates reached 90.6%, 92.7%, and 91.6% for millet, broomcorn millet, and rapeseed, respectively. These results demonstrate the feasibility of the proposed device and provide a basis for further optimization and field validation.

With the increasing scale of high-clearance sprayers, large-span flexible booms are prone to severe large-deformation nonlinear vibrations under complex vertical excitations encountered in field operation, thereby markedly compromising spray uniformity and structural safety. This study aims to elucidate the global nonlinear dynamics and chaotic evolution mechanisms of flexible spray booms under forced excitation. Based on Euler-Bernoulli beam theory and the von Kármán large-deformation assumption, and taking into account the stepped cross-sectional characteristics of the boom, the structure was equivalently modeled as a two-segment stepped cantilever beam composed of an inner boom and an outer boom. The nonlinear partial differential governing equation for the forced vibration of the system was then derived using Hamilton’s principle. The natural frequencies and mode shapes of the system were obtained analytically and compared with the results of three-dimensional finite element modal simulations in ANSYS, thereby validating the accuracy of the equivalent mechanical model. Considering the low-frequency-dominated vibration characteristics observed in field conditions, the partial differential equation was further reduced, via a single-mode Galerkin truncation, to a single-degree-of-freedom Duffing-type ordinary differential equation containing geometric nonlinear terms. The harmonic balance method was employed to derive the primary resonance frequency-response and force-response characteristics of the system, revealing hardening behavior, resonance-frequency shift, and multistable jump phenomena induced by large deformation. In addition, the sensitivities of the instability boundaries to damping and nonlinear stiffness were quantitatively characterized. Furthermore, the topological route to chaos under extreme operating conditions was systematically investigated using global bifurcation diagrams, the maximum Lyapunov exponent, time histories, two-dimensional phase portraits, and Poincaré maps. The results show that the chaotic response of large flexible spray booms is highly sensitive to the amplitude of external forcing and the damping coefficient. This study provides a solid theoretical basis for vibration-reduction-oriented structural optimization and nonlinear instability control of spray booms in modern high-clearance sprayers.

Lycium barbarum L. (L. barbarum) is a continuous-flowering and continuous-fruiting crop with thin and delicate fruit skin, necessitating multiple harvesting rounds. Currently, the harvesting of L. barbarum primarily relies on manual labor, leading to high labor demands and underscoring the urgent need for efficient, low-damage mechanized harvesting solutions. In this study, an air-suction double-roller harvesting device was designed. The harvesting mechanism operates by guiding L. barbarum into the picking zone via an airflow-induced suction stream, followed by gentle separation through a flexible double-roller structure. The fruit guiding process was modeled based on pneumatic attraction and analyzed through high-speed photography experiments. The results indicate that fruit velocity gradually increases during the guiding phase, with peak acceleration occurring at a specific location within this stage. A response-surface experiment was conducted to evaluate the effects of preloading force and roller speed on the picking rate of ripe fruit, the picking rate of unripe fruit, and the mechanical damage rate of ripe fruit. Parameter optimization resulted in optimal operating conditions with a preloading force of 0.903 N and a roller speed of 49.172 r/min. Under these conditions, validation experiments achieved a picking rate of ripe fruit of 93.7%, a picking rate of unripe fruit of 2.8%, and a damage rate of ripe fruit of 1.1%. This study provides a valuable reference for the design of vacuum-based harvesting devices for L. barbarum.

To provide fundamental data for the design and experimental evaluation of pneumatic precision seed metering devices for small-seed crops, the material characteristics of millet, broomcorn millet, and rapeseed were systematically investigated in this study. The moisture content, thousand-seed weight, density, principal dimensions, equivalent diameter, and angle of repose of the three seed types were determined through physical measurements. The Poisson's ratios of millet, broomcorn millet, and rapeseed were measured using a universal testing machine and were found to be 0.27, 0.33, and 0.29, respectively. In addition, the static and dynamic friction coefficients between each seed type and acrylic, stainless steel, and resin surfaces were obtained. The elastic moduli of millet, broomcorn millet, and rapeseed were measured using a texture analyzer as 351.76, 589.14, and 133.37 MPa, respectively, which were closely related to seed density and structural characteristics. To further clarify the influence of seed-surface collision behavior on seed motion trajectories during the seed metering process, a high-speed imaging test system was established to determine the coefficients of restitution of seeds with moisture contents of 15%, 20%, and 25% after collision with different material surfaces. The results showed that the coefficient of restitution was significantly affected by seed moisture content and decreased markedly with increasing moisture content. Among the three seed types, the coefficient of restitution decreased in the order of broomcorn millet, millet, and rapeseed. For each seed type, the coefficient of restitution decreased sequentially for collisions with acrylic, stainless steel, resin, and the corresponding seed surface. These results provide important data support for the structural design and simulation analysis of pneumatic precision seed metering devices.

To determine the optimal harvesting period for mechanized harvesting of Lycium barbarum L. (L. barbarum), fruits from different harvest batches within the same harvesting season were used as the experimental materials, and a comprehensive evaluation was conducted based on fresh-fruit ripeness and damage moisture content, and damage rate were measured on different sampling dates in two consecutive harvest batches (HP1 and HP2). Their variation patterns were analyzed, and a comprehensive evaluation model was established based on correlation analysis and principal component analysis to rank and optimize fruit quality across different sampling dates. The results showed that, with the progression of sampling dates, the ripeness of fresh L. barbarum fruit gradually increased, whereas damage resistance exhibited a stage-dependent variation pattern, and significant correlations were observed among the measured indicators. The dual-index weighted comprehensive evaluation indicated that the optimal harvesting period for both harvest batches was day 7, corresponding to an appropriate harvesting interval of 7 d. These results provide a theoretical basis for determining the harvesting period for mechanized L. barbarum harvesting.

To address the severe multiple-seed pickup problem during the seed-filling process of an air-suction seed metering device for small-seed crops with multiple seeds per hill, a combined seed-cleaning mechanism consisting of an upper seven-tooth seed-cleaning device and a lower seed-cleaning blade was developed based on an analysis of the causes of multiple pickup. Mathematical models of seed motion and force were established to describe the interaction between the seven-tooth seed-cleaning device and the seed population during the cleaning process. The installation position and adjustment mechanism of the device on the seed chamber housing were determined, and its tooth-profile parameters and major operating positions were theoretically analyzed. Accordingly, the design method and calculation models for the key parameters of the seven-tooth seed-cleaning device were established. A quadratic regression orthogonal rotational combination experiment was conducted using three factors affecting cleaning performance: the distance between the apex of the first tooth and the corresponding suction hole, the operating speed of the seed metering device, and the negative pressure. Regression equations were established and response surface analysis was performed. With the seed-cleaning qualification rate as the optimization objective, the optimal parameter combinations were obtained as follows: for millet, 3.36 mm, 3.59 km/h, and 1.43 kPa; for broomcorn millet, 3.49 mm, 4.22 km/h, and 2.11 kPa; and for rapeseed, 3.15 mm, 3.73 km/h, and 1.52 kPa. To reduce the influence of random error, 200 repeated bench tests were conducted for each seed type under its corresponding optimal parameter combination at operating speeds of 2.0-5.0 km/h. The seed-cleaning qualification rates for millet, broomcorn millet, and rapeseed were all above 90%, meeting the design requirements of the seed-cleaning mechanism. This study provides a theoretical basis and technical reference for seed-cleaning mechanisms for air-suction precision seed metering devices for small-seed crops with multiple seeds per hill.

To address the challenges of labor-intensive and inefficient manual harvesting of Lycium barbarum L. grown on a double-layer trellis, this study proposes a human–robot collaborative system. The system consists of a hand-held high-frequency rocker-oscillator harvester and an electric receiving cart. Mechanical property measurements of fruits and branches revealed detachment forces of 1.2 to 1.8 N for mature fruits and 2.5 to 3.0 N for immature fruits, with an average bending elastic modulus of 548.9 MPa for fruit-bearing branches, providing a basis for selective harvesting. A Box–Behnken design was employed to optimize vibration frequency, harvesting duration, and inter-rod spacing, and response surface analysis identified the optimal parameters as 18 Hz, 4.5 s, and 26 mm. Next, a central composite design optimized the receiving cloth tilt angle and radius to maximize collection efficiency, defined as the ratio of fruits collected by the cart to the total number of detached fruits. Field validation demonstrated a mature fruit harvesting rate of 96.2%, a mis-harvesting rate of 4.3%, a damage rate of 3.7%, and a collection efficiency of 96.1%, with relative errors below 0.5% compared to model predictions. These results confirm the system’s effectiveness and provide technical support for standardized, low-damage mechanized fruit harvesting of Lycium barbarum L..

Traditional gripping end-effectors often damage fruit due to structural complexity and uneven force distribution. To overcome these limitations, this study proposes a rotational end-effector that safely separates the stalk from the branch through a bending action. First, the design scheme and operating principle were established based on harvesting requirements. A rotational harvesting strategy was then formulated using statistical data, notably an average stalk length of 21.98 +/- 4.33 mm. Subsequently, a composite mechanical model was developed, which enabled the derivation of critical conditions for stalk separation through theoretical analysis. Next, finite element simulations were conducted to assess the bending effects and stress distribution, confirming the end-effector's capability to protect the apple's surface. Finally, the endeffector was integrated into a harvesting robot and underwent extensive field experiments. Key parameters such as motor torque, motor speed, and opening width were optimized using response surface methodology (RSM). Under conditions of a 31.43 rpm motor speed, 6.22 N & centerdot;m motor torque, and a 34.16 mm opening width, the theoretical harvesting success rate reached 90.17%, while the actual success rate was 89%, demonstrating a negligible discrepancy. These results indicate that the rotational endeffector effectively preserves apple quality and holds promising prospects for practical application.

Current Lycium barbarum L. vibration harvesting equipment exhibits low levels of intelligence and precision, often resulting in a trade-off between efficiency and fruit damage. This study proposed a ripe fruit region detection model, YOLO-RFR, specifically for precision vibration harvesting of L. barbarum. First, the ADown downsampling module was introduced to replace part of the conventional convolution layers. Then, the C3k2-AP module, inspired by the asymmetric padding strategy, was designed to replace the C3k2 module. Additionally, the GCHead detection head was constructed using group convolution. Finally, the EMA-Slide Loss function was developed to optimize the classification performance by combining the slide weighting function with Exponential Moving Average (EMA). The experimental results showed that the model achieved precision, recall, and mAP of 93.7%, 92.0%, and 97.0%, respectively, representing improvements of 4.0%, 4.4%, and 2.6% over the baseline. The parameter, floating-point operations (FLOPs), and model size were 1.7 M, 4.1 G, and 3.8 MB, respectively, corresponding to decreases of 34.6%, 34.9%, and 30.9% compared with the baseline. To further validate its practical feasibility, the improved model was deployed on an NVIDIA Jetson AGX Xavier embedded device, achieving an inference speed of 163 fps with TensorRT acceleration. In conclusion, the YOLO-RFR model demonstrated excellent performance in detection accuracy, model lightweighting, and deployment on embedded devices, providing strong technical support for the precision vibration harvesting of L. barbarum.

Apple picking is an inherently labor-intensive, time-consuming, and costly task, and robotic harvesting represents a potential alternative to address this challenge. This study presents the development and field evaluation of an integrated robotic system for apple harvesting, which combines machine vision, a dual four-degree-of-freedom (DoF) manipulator, and a mobile platform. The harvesting mechanism employed a streamlined 4-DoF manipulator driven by closed-loop stepper motors, incorporating a differential gear mechanism to execute yaw and pitch motions. Trajectory planning utilized linear interpolation with a harmonic acceleration/deceleration profile to ensure smooth end-effector movement. Fruit detection and localization within the canopy were performed by a stereo vision system running a lightweight deep neural network, achieving a mean hand-eye calibration accuracy of 4.7 ± 2.7 mm. Three negative-pressure driven soft end-effector designs—a suction soft end-effector (SSE), a grasping soft end-effector (GSE), and a suction-grasping soft end-effector (SGSE)—were assessed for their harvesting performance. Field trials conducted in a commercial spindle orchard demonstrated that the GSE achieved the highest performance, with a harvesting success rate of 80.80% among reachable fruits, a full-process success rate (from detection to collection) of 61.59%, an overall fruit damage rate of 10.89%, and an average single-fruit cycle time of 5.27 s. In contrast, the SSE and SGSE showed lower success rates (49.21% and 64.71%, respectively). This work provides a practical robotic harvesting solution. It validates the feasibility of a zoned, multi-manipulator harvesting strategy and delivers comparative data to guide the development of more efficient and robust harvesting robots.

As the core component of precision liquid fertiliser application systems, the anti-clogging mechanisms of liquid fertiliser distributors remain critical theoretical deficiencies. This study employed computational fluid dynamics (CFD) and discrete element method (DEM) to construct the flow fields of solid-liquid two-phase flow within the distributor. The accuracy of the simulation was validated through experimental testing. Based on the simulation process, by analysing the reasons for the formation of the flow fields, the study investigates the motion characteristics of liquid fertiliser in rotating flow fields and at fertiliser outlets under different rotational speeds. The basic movement trend of fibre particles was expounded. The mechanism by which fibre particles are smoothly discharged from the distributor was explored. The results indicate that the velocity distribution within the distributor's internal flow field is inversely correlated with the rotational radius. Across 180-540 rpm, negativepressure vortices can influence pulse-boosting effects within the flow field and the adsorption of particles onto baffles. The outlet pressures formed four synchronised groups (1,5,9/2,6,10/3,7,11/4,8,12), with pulse delays decreasing from 0.10 s (180 rpm) to 0.03 s (540 rpm) as speed increased, simultaneously enhancing synchronicity. The discharge of particles depends on the magnitude of velocity fluctuations, which can be modulated through rotor speed adjustments to prevent the retention of fibre particles and enhance discharge efficiency. These findings provide theoretical support for the design of high-viscosity fluid rotational distribution devices and have engineering guidance value for improving the anti-clogging performance of precision fertilisation equipment.

For the characteristic of infinite inflorescence, the aim of vibration harvesting of Lycium barbarum L. (L. barbarum) is to picking ripe fruits and leaving unripe fruits, flowers and leaves. Therefore, it is essential to understand the dynamic response of each component during vibration. In this study, the 3D models of a branch, ripe fruit, unripe fruit, flower and leaf were established respectively, which were assembled into a branch-fruit-flower-leaf system model based on the growth characteristics of L. barbarum. The vibration harvester was designed based on the hedge agronomy and simplified as rods, and a rigid-flexible coupling model was established using the kinetic analysis and the finite element method. The dynamic response of branch-fruit-flower-leaf during vibration harvesting was obtained when the rods continuously excited the branch. Results showed that the maximum acceleration of fruit-flower-leaf during vibration harvesting was in the same order of magnitude. And according to the formula of inertia force, the detachment force and mass were calculated and found that the detachment acceleration of ripe fruit was much smaller than that of unripe fruit, flower, and leaf. This study presents a theoretical basis for achieving the harvesting target. Scripts were written to simulate the vibration detachment process of ripe fruit, and high-speed photography was used for experimental verification. The results indicate that the simulation error was 9.15 %, demonstrating that the rigid-flexible coupling simulation is more effective. The field test showed that the picking rate of ripe fruit of 82.69 %, the picking rate of unripe fruit of 3.13 %, and the damage rate of ripe fruit of 4.06 %. The research provides a new analytical approach and theoretical basis for researching vibration harvesting.

Organic chemistry is an important basic course in the field of natural science, and its experimental teaching is also an important part of this course. How can students acquire practical ability in experimental practice within the limited class hours? Our organic chemistry experimental teaching team has developed a teaching model using density functional theory (DFT) calculation-aided experimental teaching of organic chemistry. In this paper, taking the ethyl acetate synthesis experiment as an example, we provided a case study using DFT calculation-aided experimental teaching of organic chemistry. The reaction mechanism of ethyl acetate was studied using Gaussian 09 software. The changes in reaction energy barrier and carbonyl carbon structure were also studied. Our results show that the rate-determining step is the nucleophilic addition. The reasonable raw material addition procedure is that the glacial acetic acid and anhydrous ethanol should first be added to a flask, followed by adding sulfuric acid slowly. DFT calculation can explain clearly the mechanism of esterification reaction in a graphic form, which is not only helpful for the students to better grasp the key points of the experiment but also beneficial for deeply understanding the esterification reaction. It provides some important guidance and reference for the teaching activities of organic chemistry in the university and high school.

The Qinghai–Tibet Plateau (QTP), a critical hydrological regulator for Asia through its extensive glacier systems, high-altitude lakes, and intricate network of rivers, exhibits amplified sensitivity to climate-driven alterations in precipitation regimes and ice mass balance. While the Gravity Recovery and Climate Experiment (GRACE) and its Follow-On (GRACE-FO) missions have revolutionized monitoring of terrestrial water storage anomalies (TWSAs) across this hydrologically sensitive region, spatial resolution limitations (3°, equivalent to ~300 km) constrain process-scale analysis, compounded by mission temporal discontinuity (data gaps). In this study, we present a novel downscaling framework integrating temporal gap compensation and spatial refinement to a 0.25° resolution through Gated Recurrent Unit (GRU) neural networks, an architecture optimized for univariate time series modeling. Through the assimilation of multi-source hydrological parameters (glacier mass flux, cryosphere–precipitation interactions, and land surface processes), the GRU-based result resolves nonlinear storage dynamics while bridging inter-mission observational gaps. Grid-level implementation preserves mass conservation principles across heterogeneous topographies, successfully reconstructing seasonal-to-interannual TWSA variability and also its long-term trends. Comparative validation against GRACE mascon solutions and process-based hydrological models demonstrates enhanced capacity in resolving sub-basin heterogeneity. This GRU-derived high-resolution TWSA is especially valuable for dissecting local variability in areas such as the Brahmaputra Basin, where complex water cycling can affect downstream water security. Our study provides transferable methodologies for mountainous hydrogeodesy analysis under evolving climate regimes. Future enhancements through physics-informed deep learning and next-generation climatology–hydrology–gravimetry synergy (e.g., observations and models) could further constrain uncertainties in extreme elevation zones, advancing the predictive understanding of Asia’s water tower sustainability.

Compared with manual harvesting of Lycium barbarum, existing large-scale harvesters have achieved a certain degree of efficiency improvement. Nevertheless, their intermittent operation mode remains a bottleneck restricting overall harvesting performance. To address this issue, a continuous vibration-based L. barbarum harvesting device was developed in this study. Plackett–Burman experiments indicated that vibration angle, vibration frequency, and the spacing between the upper and lower vibrating rods were the primary factors affecting the harvesting performance. Further parameter optimization experiments were carried out by considering the harvesting rate of ripe fruits, the mis-harvesting rate of unripe fruits, and the damage rate of ripe fruits. The optimal parameter combination was determined as a vibration angle of 46°, a vibration frequency of 9 Hz, and a spacing of 62 mm between the upper and lower vibrating rods. Based on these parameters, performance verification tests were conducted. The results showed that the harvesting rate of ripe fruits reached 85.40%, the mis-harvesting rate of unripe fruits was 4.61%, and the damage rate of ripe fruits was 3.19%. These findings provide technical and equipment support for the development of continuous mechanized harvesting of L. barbarum.

During the tillering stage of wheat, the distribution of weeds in the field is irregular, often showing single plants or clusters. Current precision spraying systems are mainly suitable for locating and spraying single-plant vegetation, which usually leads to the system missing or under-spraying when dealing with clustered weeds. In this study, a precision spraying control method is proposed to reduce the effect of camera frame rate on weed localization failure through three sets of position determination regions, and to address the effect of solenoid valve response frequency on precision spraying by controlling the spray nozzle to continuously spray herbicides on clustered weeds through a velocity-adaptive dynamic overlap region. To improve the accuracy of weed detection, GCGS-YOLO is proposed as a weed target detection model, and we integrate the Global Context (GC) attention mechanism with the traditional C3 module to optimize the backbone feature extraction network, and introduce the GSConv module to improve the neck network. The improved models P, R, mAP and F1 were 88 %, 84.6 %, 92.2 % and 86.3 %, which were 3 %, 3.1 %, 2.7 % and 3.1 % higher compared to the original model. The precision spraying algorithms and systems were integrated in a test bed and sprayer to carry out the tests. The tests showed that the recognition rate and spraying rate on the test bed could reach >98 % at different speeds. The results of the field test showed that the recognition rate and spray application rate of the sprayer were 91.2 % and 96.1 %, respectively, at a speed of 0.2 m/s. The research results can reduce the waste of herbicide, improve the efficiency of weeding, and provide reference for large-scale precision weeding.

Accurate classification of wolfberry geographical origin is essential for assessing its nutritional and medicinal properties. A multimodal convolutional neural network (MTCNN) with a cross-attention mechanism was proposed to effectively fuse spectral and image features, achieving a test accuracy of 99.88 %. Traditional feature extraction methods such as and principal component analysis (PCA) were combined with support vector machine (SVM) and k-nearest neighbor (KNN) classifiers. The SVM model using gradient boosting decision tree (GBDT) -extracted spectral features and fused image data achieved the highest accuracy of 96.68 %. The comparison results highlight the superior performance of multimodal deep learning over conventional methods, demonstrating its potential for robust agricultural product traceability, authentication, and quality assurance. Beyond achieving higher accuracy, our model improves upon previous architectures by reducing computational complexity through the use of a simplified attention mechanism and enhancing interpretability, making the model more efficient and accessible for practical applications.