It is of great significance to conduct efficient and accurate pest detection during grain storage. To address the problems of low detection accuracy and frequent false or missed detections encountered by traditional algorithms under complex backgrounds, an improved RT-DETR detection model is proposed in this paper. Based on the existing multi-scale enhancement module and gated partial attention mechanism, a Multi-scale Enhanced Gated Partial Attention Network (MEGPANet) is constructed to replace the original backbone network, thereby enhancing the model’s ability to extract pest edge contours and local features. A Diverse Branch Block (DBB) is introduced to replace the original RepBlock module, strengthening the model’s cross-scale feature fusion capability. A Conditional Position Encoding Generator (PEG) is incorporated into the Attention-based Intra-scale Feature Interaction (AIFI) module to dynamically inject positional information, further improving the localization accuracy and detection precision of tiny pests. Experimental results reveal that the precision, mAP@0.5, and mAP@0.5:0.95 of the improved model are increased by 0.9%, 5.5%, and 3.4%, respectively. While improving detection accuracy, the computational complexity of the model is also significantly reduced. Compared with other mainstream object detection models, the improved RT-DETR model proposed in this paper shows certain advantages in detection accuracy and overall performance, indicating that the model can better adapt to pest detection tasks in complex stored-grain scenarios.
To address issues in traditional paddy freshness detection methods such as low efficiency, strong subjectivity, sample damage, and reliance on professional equipment, this paper proposes an improved multi-scale attention-enhanced convolutional network model (LC-ConvNeXt) based on artificially aged paddy samples for paddy freshness detection. This model uses ConvNeXt as the backbone network, introduces a multi-scale feature fusion module (Lite-Ms Conv), and simultaneously extracts local textures, color variations, and global contextual features of paddy images through convolutional branches of different scales, enhancing the model's ability to represent fine-grained differential features. At the same time, a convolutional block attention module (CBAM) is introduced, which adaptively weights features from both channel and spatial dimensions, strengthening the model's focus on key discriminative regions while suppressing background noise interference. A dataset is constructed using three types of artificially aged paddy samples. Comparative experiments are conducted with mainstream models including MobileNetV3, ResNet50, DenseNet121, and ConvNeXt, and the model's performance is validated through ablation experiments and Grad-CAM visualization analysis. Experimental results show that the LC-ConvNeXt model achieves a classification accuracy, macro-average precision, macro-average recall, and macro-average F1-score of 97.80%, 97.74%, 97.78%, and 97.76%, respectively, with a single-sample inference time of 8.65 ms, achieving high detection accuracy while also meeting real-time requirements. The results indicate that the proposed method can achieve rapid, non-destructive detection of paddy freshness under artificially aged sample conditions, providing a technical reference for intelligent detection and quality safety evaluation of grain storage.
Hyperspectral imaging provides a new measurement approach for the non-destructive detection of moldy maize through non-contact and high-resolution spectral information acquisition.However,in practical applications,mildew-sensitive information is often hidden in high-dimensional spectral features and is therefore difficult to extract accurately.Moreover,the scarcity of mildew samples with costly manual annotations further limits the stability and generalization ability of detection models.To address these issues,a generalization-enhanced spectral convolutional network is proposed in this study for hyperspectral imaging-based detection of moldy maize based on deep learning theory.Using regional spectral features extracted from hyperspectral images of maize kernels as input,the proposed model constructs a residual feature-refined spectral convolution module to extract deep discriminative features related to mildew,and further incorporates Mixup data augmentation and label smoothing at the network input and training stages to improve classification performance and generalization ability under small-sample conditions.To verify the effectiveness of the proposed method,induced-mildew and naturally mildewed maize hyperspectral datasets were constructed,and the method was compared with existing spectral feature extraction methods under a two-stage framework of"feature extraction-classification validation."The results showed that the optimal accuracy and F1 score of the proposed method reached 94.76%and 94.77%on the naturally mildewed maize dataset,and 93.82%and 93.79%on the induced-mildew maize dataset,respectively.This method can provide effective technical support for the rapid screening and non-destructive detection of moldy maize.
Emergency food refers to specialized food products designed to meet basic human survival and nutritional needs during sudden emergencies such as disasters and accidents. These products are characterized by nutritional balance, convenience, suitability for storage and transportation at room temperature, long shelf life, ease of digestion and absorption, and the ability to quickly restore physical strength. In recent years, China has made certain results in emergency incident management and has established a system for the stockpiling and distribution of emergency food. However, due to the late start of research in this field, there remain gaps compared to advanced countries in the emergency sector regarding key technical aspects such as product variety, taste, shelf life, packaging, and nutritional properties. Furthermore, the functional positioning of these products is not sufficiently defined, and the standardization system remains incomplete. The current status of the emergency food industry and standardization in countries such as the United States, Japan, and the United Kingdom is elaborated, and the development history and shortcomings of China’s military, earthquake relief, and civilian emergency food sectors is analyzed. In response to the current state of standardization and future development needs for emergency food, a framework for China’s emergency food standardization system across four dimensions is constructed: process stages, standard types, product categories, and application attributes. It clarifies the content of foundational general standards, product standards, control and management standards, and testing standards. Based on this framework, recommendations are offered, aiming to provide new insights for the standardized and regulated development of the country’s emergency food industry and to enhance the nation’s capacity for emergency material supply.
To explore the optimal preparation process for selenated konjac oligoglucomannan(Se-KOGM)with high antioxidant activity,three konjac oligoglucomannan(KOGM)with different molecular weight ranges were used as raw materials.Se-KOGM was synthesized via the HNO3-Na2SeO3 method,with selenium content and in vitro antioxidant capacity serving as evaluation criteria.The preparation process was optimized through single-factor experiments and response surface methodology,and the product was structurally characterized by UV and infrared spectroscopy.The results of the single-factor experiments showed that selenization time,selenization temperature,nitric acid concentration,and sodium selenite dosage all significantly affected the selenium content of Se-KOGM,and the selenium content of Se-KOGM was negatively correlated with the molecular weight of the reactant KOGM.In vitro experiments indicated that Se-KOGM possessed effective scavenging activity against·OH,DPPH·and O2-,and the antioxidant activity of the product prepared from low-molecular-weight KOGM was higher.The results of the response surface experiments showed that the order of influence of the main factors was:sodium selenite dosage>nitric acid concentration>selenization time>selenization temperature.The optimal preparation conditions were:selenization time of 8.75 h,selenization temperature of 70.05℃,HNO₃ concentration of 0.07%,and Na₂SeO₃ dosage(mNa2SeO3/mKOGM)of 0.89.Under these conditions,Se-KOGM with a selenium content of 4.98 mg/g was obtained.Ultraviolet and infrared spectroscopy analysis showed that Se-KOGM exhibited a distinct absorption peak at 197 nm and a characteristic Se=O absorption peak at 609.93 cm-¹,resulting in the preparation of an organic selenium food ingredient with high selenium content and high antioxidant activity.
The extensibility of Lamian dough is most important for chain noodle restaurants but it tends to decrease during the resting processing in current Lamian production. This study investigated change law and mechanism in the extensibility of sodium metabisulfite (SMBS)-treated Lamian dough during resting time (0, 10, 20, 30, 60 and 90 min). The results showed that with prolonged resting time, the extension distance of the dough significantly decreased (P<0.05) from 58.01 mm to 30.09 mm, while the breaking force significantly increased (P<0.05). Both the storage modulus (G′) and loss modulus (G″) showed an upward trend. The sulfhydryl (—SH) content significantly decreased (P<0.05), whereas the disulfide bond (S—S) content significantly increased (P<0.05). The glutenin macropolymer (GMP) content increased significantly (P<0.05), and hydrogen bonds showed a fluctuating upward trend. Hydrophobic interactions increased from 0 to 30 min but decreased from 30 to 90 min. During resting, the oxidation of —SH to S—S bonds, together with hydrogen bonds and hydrophobic interactions, promoted GMP reformation and decreased SDS-extractable protein, leading to reduced extension distance and increased breaking force. By elucidating gluten protein changes in Lamian dough during resting, this study provides a theoretical basis for developing pre-made dough products.
Traditional methods for environmental monitoring and control in granary still face certain limitations in terms of multi-source information fusion, complex environmental modeling, and closed-loop decision-making, which may lead to delayed early warnings, crude control, and insufficient system coordination. Utilizing data-driven methods for monitoring, predicting, and controlling storage environments is a key technical approach to enhancing the level of smart grain storage. This paper systematically summarizes methods for acquiring and preprocessing grain storage datasets, explores recent advancements in the application of traditional machine learning, deep learning, and deep reinforcement learning (including Deep Q-Network, Deep Deterministic Policy Gradient and Multi-agent Deep Reinforcement Learning) for grain condition monitoring, grain temperature prediction, and environmental control, and analyzes the major problems in current research, including inconsistent data standards, insufficient cross-warehouse model generalization, difficulties in verifying the safety of control strategies, and the complexity of system integration. Furthermore, the paper outlines future research directions, such as multimodal sensing, safety-aware reinforcement learning, multi-agent coordination, and lightweight digital twins.
As a key link in the production operations of grain storage enterprises, the grain conveying system often faces the challenge of coupled complex operating conditions in its actual operation. Traditional manual inspections and simple sensor alarms have limitations like incomplete monitoring, poor real-time performance, and difficulty in detecting early abnormalities, easily causing production interruptions or even safety accidents. Data-driven technology provides a new solution for its condition monitoring and anomaly early warning, serving as an important technical means to ensure safe and stable operation and improve enterprises’safety production level. Firstly, the basic composition and common abnormal states of grain conveying systems are elaborated, multi-dimensional signal perception technologies such as vibration, temperature, and sound are summarized, data preprocessing processes including cleaning and denoising are introduced, and three major feature extraction strategies (time domain, frequency domain, and time-frequency domain) are discussed; Secondly, focusing on the application progress of machine learning methods represented by support vector machines, Gaussian mixture models, and ensemble learning, as well as deep learning methods represented by long short-term memory networks, gated recurrent units, and Transformers in condition monitoring and anomaly early warning. Finally, it analyzes the shortcomings of current research in multi-condition adaptation, small sample processing, and practical scenario implementation, and looks forward to future research directions such as multi-sensor collaborative acquisition, development of special algorithms for grain depots, and multi-system interaction.
In-depth mining and analysis of the massive data reported in the literature on grain, oil and food enable the systematic integration of existing research resources and findings, thereby providing a robust foundation to guide subsequent studies. However, traditional data-mining approaches are typically highly dependent on expert knowledge, resulting in inefficiency and susceptibility to subjective judgment, which limits their versatility and practical utility. Large language models (LLMs) have recently emerged as a transformative approach, offering distinct advantages in reasoning and generation, thereby presenting novel and efficient pathways for domain knowledge synthesis and scientific innovation discovery. In response to the current lack of systematic data-mining frameworks in the field, this paper first outlines the fundamental technical principles and workflow of LLM-based data mining, which can assist in tasks such as domain literature screening, data extraction, and information integration, providing foundational support for data extraction and database construction in the grain, oil and food field. Furthermore, by coupling extracted data with knowledge generation, the framework enables downstream applications such as system-level deployment and the development of task-specific models, promoting the utilization of domain data across multi-task platforms and real-world scenarios. This paper reviews the technologies of LLMs in data mining and applications in the field of grain, oils and food sector, offering comprehensive analysis of workflow, disciplinary applications, and faced challenges. It aims to provide researchers with actionable guidelines and practical references for the effective utilization of LLMs in conducting large-scale literature-based data mining.
Ganoderma lucidum spore oil (GLSO) is a fat-soluble bioactive substance extracted from cell-wall-broken Ganoderma lucidum spore powder, and it holds broad application prospects in the fields of functional foods and pharmaceuticals. This paper provides a systematic review of research progress regarding its preparation technology, bioactive components, and multidimensional pharmacological mechanisms. Supercritical CO2 extraction has become the mainstream method for industrial production due to its green and efficient nature and the absence of solvent residues. GLSO is rich in monounsaturated fatty acids, triglycerides, steroids, and triterpenoids, among which, monounsaturated fatty acids, primarily oleic acid, serve as the key molecular basis for its immunomodulatory and antitumor activities. Although triterpenoid components are regarded as important indicators for evaluating Ganoderma quality, existing detection technologies still have limitations, and it is currently inappropriate to use them as the sole standard for quality control. In terms of pharmacological mechanisms, GLSO exhibits significant multiple biological activities, including antitumor, immunomodulatory, and antioxidant effects. Its antitumor mechanism demonstrates multi-target synergistic characteristics. Although GLSO holds promising prospects for application, it still faces challenges such as technical bottlenecks in triterpenoid detection, a lack of clinical evidence, and limited product forms. In the future, the realization of the industrialization of GLSO will require the establishment of standardized production processes and precise quality monitoring systems. Through clinical trials, this will drive high-quality development in both precision nutrition and pharmaceutical-grade applications.
In this paper, oat flour with different gelatinization degrees (30%, 50%, 70%, 90%) was prepared by controllable pre-gelatinization treatment using a drum stir-frying thermal processing technique. of oat kernels with different tempering water contents. Oat protein was extracted from the oat flour with different pre-gelatinization degrees, and the effect of stir-frying gelatinization on the structural characteristics and functional properties of oat protein was systematically analyzed. 30P, 50P, 70P, and 90P represent oat protein extracted from oat flour with gelatinization degrees of 30%, 50%, 70%, and 90%, respectively. The results showed that with increasing pregelatinization degree, the oat protein underwent a transition from a naturally dispersed state to a depolymerized and homogenized state (30P and 50P), then to a crosslinked network structure (70P), and finally formed a dense aggregated structure (90P). During the process of stir-frying, aggregates with a molecular weight greater than 100 kDa were formed. Moderate pre-gelatinization induced by tempering water content reduced the α-helical structure and increased the β-sheet structure of oat protein, and 50P has the highest β-sheet content, reaching 38.25%. The intrinsic fluorescence intensity of pregelatinized protein was enhanced, and the maximum emission wavelength shifted towards the longer wavelength direction. Furthermore, it improved the solubility (highest at 50P, 21.07%), water-holding capacity (highest at 70P, 2.84 g/g), emulsifying activity (highest at 50P, 22.03 m²/g) and emulsion stability of oat protein, while reducing the foaming capacity and foam stability. The results indicate that moderate pre-gelatinization treatment could regulate the microscopic morphology and structure of oat protein, thereby influencing its functional characteristics, which could provide a theoretical basis for the precise processing and functional application of oat protein.
Finished grain storage in High-Bay Racks is an emerging form of grain storage in China in recent years. However, during the integration of informatization and intelligent applications, technical issues often arise from the fragmentation of multi-source heterogeneous data, resulting in difficulties in data sharing, low efficiency in multi-system collaboration, inconsistency between grain condition monitoring and quality preservation, and complexity in inventory inspection. These problems have become pain points and challenges in the industry's intelligent transformation. This paper addresses these issues by conducting research on data-driven intelligent technologies. Through systematic analysis of the characteristics of finished grain storage and the core requirements of intelligentization, a data-driven intelligent technical architecture for finished grain storage in High-Bay Racks is proposed. A five-layer integrated platform system is constructed, comprising the perception layer, control layer, system layer, platform layer, and application layer, with "data hub + business middle platform" as its core. The core functions of multi-source heterogeneous data fusion and the data hub are analyzed. The paper discusses how to achieve multi-source heterogeneous data fusion and high-quality assetization through establishing a unified information classification, coding, and master data alignment mechanism. An intelligent application of finished grain storage in High-Bay Racks was implemented in a grain reserve enterprise in the Guangdong-Hong Kong-Macao Greater Bay Area. A comparative analysis of application effectiveness shows that the data-driven intelligent technology for finished grain storage in High-Bay Racks fully realizes high-quality data collection, sharing, exchange, and analysis, providing a replicable and scalable data-driven intelligent paradigm for finished grain storage.
Pea protein isolate(PPI)was used to prepare edible packaging films via a pH-shifting combined with heat treatment in the present study.The effects of pH(8～12)and heat treatment temperature(70～90℃)on PPI solubility and the physicochemical properties of the resulting films were systematically investigated.The results showed that the optimal modification conditions were pH of 10 and heat treatment temperature of 90℃,yielding the highest PPI solubility(67.14%).Alkaline treatment at pH 10 induced unfolding protein molecules,disrupting hydrophobic interactions and van der Waals forces.Heat treatment at 90℃further exposed charged residues within the protein molecules,significantly increasing the absolute value of the zeta potential,thereby enhancing intermolecular electrostatic repulsion and inhibiting protein aggregation.The pea protein film prepared under the optimal conditions exhibited superior mechanical properties,water resistance,and barrier performance:tensile strength was 4.7 MPa,elongation at break was 182.4%,moisture content and water solubility was as low as 17.69%and 17.10%,respectively,the water vapor transmission rate was 415.1 g/(m2·24 h),and the oxygen transmission rate was only 2.1 cm3/(m2·24 h·0.1 MPa).Correlation analysis indicated that the mechanical properties of the pea protein film,such as tensile strength and elongation at break,showed a highly significant positive correlation with pea protein solubility,whereas water vapor transmission rate and moisture content showed highly significantly and negatively correlated with PPI solubility.Overall,PPI solubility markedly influenced the film performance.The results provides a theoretical basis and practical guidance for the industrial application of plant-based edible packaging films.
To addresses the difficulty of non-destructive detection of early infection by the hidden pest sitophilus zeamais in wheat kernels, the practical differences in moisture content during wheat harvesting and storage is considered. Three groups of wheat samples with different moisture levels were selected, and hyperspectral imaging data were collected. Based on machine learning methods, mean spectral analysis, comparison of different preprocessing methods, comparison of different classification models, and recognition analysis of different infestation stages were conducted. A mixed-moisture-aware hierarchical and fusion SVM model, namely MAHF-SVM, was proposed and a classifier was constructed to validate the model performance. The results showed that moisture content significantly affected the spectral distribution of wheat, and was an important factor affecting identification performance. Among different preprocessing methods and classification models, SNV-SVM showed relatively strong overall performance among conventional models, achieving an accuracy of 89.35% and a recall of 88.63% under mixed-moisture conditions. At different infestation stages, the recognition performance in the pupal and late-larval stages was generally better than that in the egg and early-larval stages, and overall classification performance decreased with increasing moisture content. The proposed MAHF-SVM model effectively mitigated spectral information confusion caused by moisture variation. The MAHF-SVM model achieved an accuracy of 91.90% and a recall of 92.08%, effectively mitigating spectral information confusion caused by moisture variation and demonstrating improved model performance.
Transglutaminase (TGase) can significantly enhance the emulsifying activity of soy protein isolate (SPI) through covalent cross-linkin. However, the cross-linking efficiency is significantly restricted by the aggregation state of SPI. In this experiment, the effects of different pretreatment methods (heat treatment, ultrasonic treatment, and DTT treatment) on the structure and emulsifying activity of TGase-cross-linked SPI were compared. The results of sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE), Fourier transform infrared spectroscopy (FTIR), and fluorescence spectroscopy indicates that the pretreatments induced the full unfolding of protein structure, exposing more hydrophobic regions and -SH groups, thereby enhancing the sensitivity of SPI to TGase. The effects of the three pretreatments exhibited distinct effects: ultrasonic treatment efficiently dissociated the SPI structure and exposed active sites by virtue of cavitation effect, providing the most sufficient reaction interface for TGase-cross-linking and thus achieving the optimal effect; heat treatment and DTT treatment played an auxiliary role in cross-linking, and their effects were relatively moderate. After structural unfolding, the hydrophobic residues of SPI were more easily adsorbed at the oil-water interface, increasing emulsion viscosity and consequently improving emulsifying capacity. Meanwhile, the covalent bonds catalyzed by TGase further strengthened the interfacial protein film, effectively inhibiting oil droplet aggregation and emulsion stratification, and significantly enhancing the emulsion stability. In conclusion, TGase-cross-linking can alter the secondary structure of SPI and promote molecular aggregation. Combined with the synergistic effect of pretreatments in exposing active sites, it can significantly improve the emulsifying properties of SPI, especially conducive to the enhancement of emulsion stability, which provides theoretical support for the development of plant protein-based emulsion stabilizers.
To improve the problems of high chemical consumption and serious environmental pollution in the traditional refining process of Camellia oleifera seed oil, this study explored a high-pressure microfluidization (HPM)-mediated nano-degumming and deacidification process, with LC-MS/MS used to analyze changes in lipid composition before and after treatment. The results showed that under conditions of 0.10% acid addition, 1% water addition, 0.10% excess alkali usage, and a homogenization pressure of 50 MPa, the total phosphorus content in the nano-degummed and deacidified oil (NO) was reduced to 2.39 mg/kg, with a degumming rate of 95.32%. The acid value and peroxide value both met national standards, indicating that this process enables efficient degumming and deacidification. LC-MS/MS analysis identified 567 lipids in the crude Camellia oleifera seed oil, among which the difficult-to-remove non-hydrated phospholipids—phosphatidic acid (PA) and phosphatidylethanolamine (PE)—accounted for 49.78% and 31.23%, respectively. After HPM treatment, the removal rates of PA, PE, and phosphatidylcholine (PC) all exceeded 99%, indicating its high efficiency in removing both hydrated and non-hydrated phospholipids. This is attributed to the nanoscale droplet system formed by the cavitation effect of HPM, which, with its ultra-large specific surface area and high interfacial reaction activity, rapidly disrupts the metal chelation structure of non-hydrated phospholipids and achieves synchronous separation of phospholipid micelles through soapstock adsorption, thereby overcoming the limitations of traditional degumming. Analysis showed that HPM treatment exerted no significant effects on the composition and content of fatty acid in Camellia oleifera seed oil, and did not cause oxidation of unsaturated fatty acids. In summary, the HPM-mediated nano- degumming and deacidification process not only achieves efficient degumming and deacidification but also has no significant negative impact on lipid components, maintaining oil quality. This provides a green and efficient new solution for the refining of Camellia oleifera seed oil.
Soybean holds strategic significance for China's edible oil and protein feed security,yet the industry faces challenges including low yield,high production costs,poor profitability,and heavy import dependence.This study aims to analyze the structural evolution and cost-benefit characteristics of China's soybean industry since 2000,providing references for improving industrial efficiency and quality.Based on national data from 2000-2024,this study employs comparative advantage and cost-benefit analysis methods to systematically examine the industry from multiple dimensions,including production,spatial distribution,trade,and cost structure.Results show that although cultivation area expanded from 9,306.58 thousand hectares in 2000 to 10,325.04 thousand hectares in 2024,and output rose from 15.409 million tons to 20.6478 million tons,the yield(2.00 tons/hectare)remains significantly lower than that of major producing countries such as the United States and Brazil.The industrial layout is characterized by"strengthening in the Northeast,rising in the Southwest,and shrinking in traditional production areas,"with the scale advantage index in the Northeast consistently above 2.0.Import dependency has long exceeded 80%,forming a"high-import,low-export"pattern.Total costs rose sharply,with the share of land costs increasing from 20.06%to 50.84%.Consequently,net profits have been negative since 2014,falling to-350.80 CNY/mu in 2024—a clear"increased production without increased income"dilemma.Despite expansion,profitability of China's soybean industry continues to decline due to resource constraints,cost pressures,and international competition.Moving forward,priorities should include optimizing regional layout,accelerating yield-focused technology,strengthening policy support,and developing value-added processing to build a more efficient,secure,and sustainable soybean industry system.
The traditional manual inspection mode in grain depot is plagued by low efficiency, subjectivity-interfered inspection results, high safety risks for on-site operators and delayed response to grain condition abnormalities, which makes it difficult to meet the refined management and control needs of modern grain depots. To address these pain points, this paper designs a multi-technology integrated inspection and early warning robot system for grain depots, realizing all-weather, intelligent and automatic inspection and accurate early warning in grain depot areas. Taking the Ackermann-structured drive-by-wire chassis as the hardware carrier, the system integrates RTK positioning and 16-line LiDAR to construct a multi-source SLAM navigation map, adopts the A* algorithm for global path planning and combines the TEB algorithm for dynamic obstacle avoidance, and ensures the accuracy of trajectory tracking and speed stability of the robot through the pure pursuit algorithm and PID control algorithm. Equipped with a high-definition PTZ camera, the system realizes real-time identification and on-site voice early warning of 12 types of grain depot-specific violations involving personnel, operations, environmental equipment and other aspects based on the YOLO model. A 5G cloud management platform is built to realize robot status monitoring, remote scheduling, early warning information upload and multi-terminal push through the MQTT protocol, and integrates multi-system data such as grain condition, pest condition, gas detection and fire early warning to form an integrated management and control closed loop. The system was tested at Sinograin Heze Direct Depot, achieving a mapping accuracy within 5 cm, an average inspection path deviation of less than 15 cm, an average accuracy of 92.3% in violation identification and a false alarm rate of only 2.6%. The cloud platform realizes real-time data update and timely early warning response. Test results show that the system effectively improves the efficiency and accuracy of grain depot inspection, reduces labor costs and the incidence of production safety accidents, and cuts down grain deterioration losses. It provides reliable technical support for grain storage security and has good engineering application and promotion value.
To develop an efficient and visual composite indicator film for food freshness detection, this study utilized purple sweet potato anthocyanin as a pH-sensitive indicator, chitosan and soluble starch as film-forming substrates, and glycerol as a plasticizer. Four gradient mass fractions of purple sweet potato anthocyanins (0%, 20%, 26%, and 33%) were designed, and a series of indicator films were successfully prepared using the casting method. The gradient effects of different addition levels on the structural, mechanical, and pH-responsive properties of the films were investigated. The results showed that the prepared films exhibited uniform thickness and excellent flexibility. The addition of purple sweet potato anthocyanins significantly increased the color difference (∆E) of the films, with the tensile strength reaching a maximum of 19.336 MPa. Moreover, the films demonstrated sensitive pH responsiveness, enabling rapid visual indication of environmental pH changes through color variation. Based on these findings, a “chitosan-soluble starch-purple sweet potato anthocyanin” ternary composite indicator film was developed. When applied to monitor the freshness of Litopenaeus vannamei at 4 ℃, the film color gradually changed from deep purple to yellow-green after 5 days. Multidimensional dynamic correlation analysis was conducted between the film’s color change (∆E) and the pH value, total volatile basic nitrogen (TVB-N) content, and sensory evaluation of the shrimp samples. The results showed high consistency among these parameters, confirming the favorable detection effectiveness and sensitivity of the developed film. This study provides a basis for the visual detection of Litopenaeus vannamei freshness and offers insights for the material design and performance optimization of natural polysaccharide-based intelligent packaging films.
The determination of the population level of stored grain pests plays a crucial role in the safe storage of grain,risk level assessment,and the formulation of subsequent granary prevention and control management strategies.Specifically,grain with a main pest density of less than 2 individuals per kilogram is classified as basically pest-free grain,3-10 individuals per kilogram as generally infested grain,and more than 10 individuals per kilogram as severely infested grain.Accurate counting of stored grain pests is thus a core prerequisite for implementing scientific prevention and control.Current research on grain pest counting either requires extensive manual data annotation or relies on supervised annotation methods for model training.Furthermore,existing methods suffer from poor model generalization ability and insufficient counting accuracy when facing practical scenarios where grain pests are tiny,highly similar to the grain background,diverse in species,and prone to variation.To address these issues,the PestCount-dinov2 grain pest counting model is proposed,a training-free and unsupervised model designed to tackle current technical challenges.The model employs the unsupervised DINOv2ViT-L/14 as the global feature extraction backbone,while leveraging YOLOv8 to complete grain pest region screening and feature extraction.These two components collaborate to generate grain pest features,eliminating the need for manual annotation required by traditional methods.Based on the fused features,the model further generates a counting density map to achieve accurate counting of grain pests.Validation results on a self-constructed small-scale stored grain pest dataset and a public tomato dataset demonstrate that this scheme provides a low-cost and high-precision counting method for grain pest monitoring,which can be flexibly adapted to the counting scenarios of stored grain pests.
