A method and system for detecting grain conditions to ensure food quality and safety

Abstract

This invention discloses a method and system for detecting grain conditions and ensuring grain quality and safety, relating to the field of grain quality detection technology. The method involves multi-target three-dimensional sampling of grain piles inside grain transport vehicles to obtain target sampling points, and collecting corresponding multi-dimensional sensor data at each target sampling point; performing joint feature extraction and fusion analysis to construct multi-dimensional feature vectors and conduct preliminary grain risk assessment and early differentiation of risk root causes, classifying and storing target grain samples; during grain storage, real-time collection of gas and temperature data from the grain silo is conducted for predictive analysis of grain conditions, generating a dynamic grain condition risk structure; risk correlation analysis is performed based on the dynamic grain condition risk structure to generate abnormal operation early warning signals; simultaneously, abnormal behavior detection is performed during grain storage, and grain silo environmental control is implemented, solving the problem of low effectiveness in grain quality risk assessment in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of grain quality testing technology, and in particular to a method and system for detecting grain conditions to ensure grain quality and safety. Background Technology

[0002] To ensure the stability and safety of grain quality during storage, and to monitor and analyze key parameters such as temperature, humidity, moisture, pest density, and gas composition within the grain pile and storage environment during the grain storage process, grain condition monitoring is a technical means of monitoring and analyzing these parameters. Current technologies primarily rely on integrated sensor networks and monitoring systems. By deploying sensors inside the grain silo to detect temperature, humidity, gases (such as carbon dioxide and phosphine), pests, and microbial activity, the physical and biochemical parameters of the grain storage environment are collected in real time. Combined with video monitoring and sampling inspections, the data is uploaded to a central platform, where big data analysis is used to intelligently assess the condition of the grain pile, enabling early warnings of abnormal conditions such as mold and pests. This aims to improve the efficiency and safety of grain quality monitoring.

[0003] For example, the invention patent with publication number CN119827494A discloses a grain quality testing device, method, control device, and storage medium, including: transmitting a target grain sample to a first discharge port and a second discharge port in two directions for appearance quality testing, bulk density testing, internal quality testing, and safety index testing; wherein the appearance testing involves extracting the contour by acquiring image data of the target grain sample and determining the contour area, perimeter, and aspect ratio; the internal quality testing involves acquiring infrared spectral data of the target grain sample and determining the characteristic spectrum to determine the internal quality testing result; and the safety index testing involves acquiring images of the sample's nano-biosensor array and determining the fission data of characteristic volatile compounds.

[0004] However, due to the limited spatial density of sensor deployment, when the sensors deployed in the grain warehouse acquire the characteristics of the grain, the diffusion and migration of heat, moisture and gas in the grain pile is a slow process. When the sensors detect an anomaly, it is usually because the quality of the grain has changed locally, and they cannot fully capture the changes in real time and carry out corresponding automated intervention and control.

[0005] Furthermore, when integrating and processing multi-source parameters of grain acquired by sensors, there is a lack of efficient integration and in-depth correlation analysis, making it difficult to accurately distinguish different risk sources such as pests and mold. This is because when analyzing multi-source parameters acquired by sensors, such as temperature, relative humidity, and carbon dioxide concentration at a certain point, linear correlation is often used. When temperature and carbon dioxide rise simultaneously, it indicates an abnormality. However, different risk scenarios such as pests, mold, and increased grain respiration cause highly similar and overlapping temperature, humidity, and gas parameter changes in their initial stages in the original data space. It is difficult to distinguish the characteristics of different risk sources from these similar changes, which further leads to the problem of low effectiveness in grain quality risk assessment during the grain quality detection process. Summary of the Invention

[0006] To address the technical problem of low effectiveness in assessing grain quality risk in existing technologies, this invention provides a method and system for detecting grain conditions related to grain quality and safety. The technical solution is as follows: On the one hand, a method for detecting grain conditions to ensure food quality and safety is provided. This method includes: during the grain transport vehicle's entry into the warehouse, performing multi-target three-dimensional sampling of the grain pile inside the transport vehicle to obtain target sampling points, and collecting corresponding multi-dimensional sensor data at each target sampling point; constructing a multi-dimensional feature vector to characterize the early state of the grain at the target sampling points through joint feature extraction and fusion analysis of the multi-dimensional sensor data, and performing preliminary grain risk assessment and early differentiation of risk sources based on the multi-dimensional feature vector; classifying and storing the target grain samples corresponding to each target sampling point based on the results of the early differentiation of risk sources; simultaneously, during grain storage, collecting real-time gas and temperature data from the grain warehouse, performing predictive analysis of grain conditions, and generating a dynamic grain condition risk structure; performing risk correlation analysis based on the dynamic grain condition risk structure, generating abnormal operation early warning signals, and simultaneously detecting abnormal behavior and controlling the grain warehouse environment during storage.

[0007] On the other hand, a grain condition monitoring system for grain quality and safety is provided. This system includes: a target sampling and acquisition module, used to perform multi-target three-dimensional sampling of grain piles inside grain transport vehicles during the grain transport vehicle's entry into the warehouse to obtain target sampling points, and to collect corresponding multi-dimensional sensor data at each target sampling point; a preliminary grain analysis and judgment module, used to construct a multi-dimensional feature vector characterizing the early state of grain at the target sampling points through joint feature extraction and fusion analysis of the multi-dimensional sensor data, and to perform preliminary grain risk judgment and early differentiation of risk root causes based on the multi-dimensional feature vector; a grain condition predictive analysis module, which, based on the results of early differentiation of risk root causes, classifies and stores the target grain samples corresponding to each target sampling point, and simultaneously collects real-time gas and temperature data from the grain warehouse during grain storage, performs predictive analysis of grain conditions, and generates a dynamic grain condition risk structure; and a risk correlation analysis module, used to perform risk correlation analysis based on the dynamic grain condition risk structure, generate abnormal operation early warning signals, and simultaneously detect abnormal behavior and regulate the grain warehouse environment during grain storage.

[0008] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: 1. This invention applies a multi-target three-dimensional sampling method to the grain storage process. By combining real-time storage data with a historical risk database for spatial mapping and overlay analysis, the risk probability of each basic grid unit within the three-dimensional space of the grain pile is calculated. This process, through data-driven intelligent analysis, identifies potential risk areas historically prone to mold and pests. Based on risk probability ranking and considering spatial uniformity constraints, not only are the points with the highest risk probability selected as initial targets, but supplementary point screening ensures the rationality and coverage of the sampling points' distribution in three-dimensional space. This approach allows limited sampling resources to be prioritized for locations with problems and reflecting the overall grain condition. Compared to existing technologies with limited sensor deployment density and the inability to comprehensively and instantly detect the interior of the grain pile upon storage, this method can scan the internal state of the grain pile upon entry. This not only increases the probability of problem detection but also ensures that the collected samples representatively reflect the early condition of the entire batch of grain, laying a solid and reliable data foundation for subsequent in-depth analysis and precise intervention. This enhances the initiative and predictability of grain quality and safety management at the source.

[0009] 2. By performing precise time-stamp-based synchronization calibration between near-infrared spectral sequences and high-frequency acquired temperature data sequences, a spatiotemporally aligned data stream is formed. Near-infrared spectroscopy contains molecular information about the internal chemical composition of grains, serving as an indicator of changes in chemical quality; while temperature sequences characterize the thermodynamic phenomena of heat generation and transfer accompanying processes such as grain respiration, microbial metabolism, or pest activity. Through temporal alignment, spectral fluctuation characteristics and temperature evolution characteristics within the same time window are coupled and analyzed. Deep feature parameters such as the first derivative of the spectrum, the standard deviation of spectral fluctuation, the linear slope of temperature change, and the standard deviation of temperature fluctuation are extracted and, after normalization, concatenated into a unified multidimensional feature vector. This vector, from the two dimensions of dynamic changes in chemical properties and thermodynamic process response, collaboratively characterizes the state evolution pattern of grains on a microscopic timescale. This deep integration approach allows patterns that were originally difficult to distinguish on a single data dimension, such as the slight temperature rise caused by the initial activity of pests and the spectral changes of specific metabolites, as well as the spectral absorption and local heat accumulation corresponding to the characteristic volatile substances released in the initial stage of mold growth, to be separated and highlighted in a multi-dimensional fusion feature space. This provides a richer characterization basis for the subsequent accurate differentiation of different risk sources.

[0010] 3. By dynamically linking the risk and hazard category labels identified during the warehousing stage with the monitoring and control during the storage stage, differentiated key monitoring is implemented for grains of different risk categories during storage. During real-time monitoring, the grain storage space is intelligently divided into multiple vertical monitoring zones based on the vertical stacking depth of the grain pile. This considers the differences in quality evolution and risk types of grain at different depths during storage due to variations in ventilation conditions, temperature and humidity gradients, and gas diffusion rates. For each vertical monitoring zone, a predicted risk index is calculated based on dynamic correlation analysis of temperature and gas concentration data, generating a regional risk assessment value. When the risk in a certain zone exceeds the standard, the control strategy activates the most targeted, pre-set control plan based on the zone's depth and common risk characteristics. This layered identification, zoned assessment, and differentiated control management allows environmental intervention measures to directly target specific grain pile locations and risk types exhibiting risk signs. This avoids the energy waste, drug residues, or adverse effects on normal grain quality caused by overall ventilation in existing technologies, achieving optimal storage results with minimal intervention while ensuring overall grain safety.

[0011] 4. By constructing a complete process spanning the entire lifecycle of grain warehousing, storage, monitoring, and control, intelligent 3D sampling and multi-source data fusion analysis at the time of warehousing enable risk classification of the initial state of the grain. The classification results then guide storage layout and monitoring. During storage, real-time environmental data is continuously collected for predictive analysis of grain conditions, generating a dynamically evolving risk structure map. Finally, spatial correlation analysis is performed based on the risk map, triggering automated environmental control commands by region and type. The entire process is interconnected; the analysis results of the previous stage provide input and basis for decision-making in the next stage, and the execution effect of the next stage can be fed back through continuous monitoring data for model verification and optimization. It not only possesses initial analysis and judgment capabilities but also adaptive capabilities in long-term operation, thereby improving the level of grain storage safety assurance in grain storage management. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A flowchart illustrating a method for detecting grain conditions to ensure grain quality and safety, provided in an embodiment of the present invention; Figure 2 This is a flowchart corresponding to the target sampling collection provided in the embodiments of the present invention; Figure 3 The flowchart corresponding to the preliminary grain analysis and judgment provided in the embodiments of the present invention; Figure 4 The flowchart corresponding to the risk association analysis provided in the embodiments of the present invention; Figure 5 This is a schematic diagram of the structure of a grain condition detection system for grain quality and safety provided in an embodiment of the present invention; Figure 6 A graph showing the relationship between actual ventilation humidity and corresponding predicted values ​​provided in this embodiment of the invention; Figure 7 This is a graph showing the relationship between the actual nitrogen concentration and the corresponding predicted value provided in an embodiment of the present invention. Figure 8 This is a graph showing the relationship between the actual heat dissipation rotation speed and the corresponding predicted value, provided in an embodiment of the present invention. Figure 9 A diagram showing the relationship between regional risk assessment values ​​and grain warehouse environmental control models provided in this embodiment of the invention. Detailed Implementation

[0014] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0015] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0016] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0017] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0018] Embodiment 1 of the present invention provides a method for detecting grain condition to ensure grain quality and safety, such as... Figure 1 The flowchart shown illustrates a grain condition detection method for grain quality and safety. The method's processing flow can include the following steps: During the grain transport vehicle's entry into the warehouse, multi-target three-dimensional sampling is performed on the grain pile inside the transport vehicle to obtain target sampling points, and corresponding multi-dimensional sensor data is collected at each target sampling point; through joint feature extraction and fusion analysis of the multi-dimensional sensor data, a multi-dimensional feature vector is constructed to characterize the early state of the grain at the target sampling points, and preliminary grain risk assessment and early differentiation of risk root causes are performed based on the multi-dimensional feature vector; based on the results of the early differentiation of risk root causes, the target grain samples corresponding to each target sampling point are classified and stored. Simultaneously, during grain storage, real-time gas and temperature data from the grain warehouse are collected for predictive analysis of grain conditions, generating a dynamic grain condition risk structure; risk correlation analysis is performed based on the dynamic grain condition risk structure to generate abnormal operation early warning signals, while abnormal behavior detection and grain warehouse environmental control are conducted during grain storage.

[0019] like Figure 2 The diagram shows a flowchart of the target sampling collection provided in an embodiment of the present invention. The specific process of multi-target three-dimensional sampling is as follows: acquiring the warehousing data during the grain transport vehicle's entry into the warehouse, that is, acquiring the three-dimensional point cloud data of the grain transport vehicle during the entry into the warehouse, and performing spatial mapping and overlay analysis with the warehousing data stored in the historical risk database to obtain the risk probability of each basic sampling grid. The basic sampling grid is a unit grid in the three-dimensional space of the grain pile corresponding to the grain transport vehicle during the entry into the warehouse that is adapted to the sampling operation range.

[0020] The historical risk database is a structured knowledge base for warehousing quality risks. Its construction is a continuous learning and accumulation process. The core of the database is to record and associate the spatial location of risks with the characteristics that cause them. The data in the historical risk database mainly comes from historical cases of multiple grain depots monitored over a long period of time. When a monitored grain depot is confirmed to have a risk disaster (the types of risk disasters include quality risks such as pests, mold, and localized high temperatures) during storage through preset personnel inspections or by triggering risk processes, the risk case is recorded. The recorded content includes spatial information (three-dimensional coordinates of the location where the risk occurred), risk labels (clear risk types, such as localized heat accumulation), associated traceability information (automatically backtracking and associating the grain entry information corresponding to the risk point, including: the approximate vehicle type of the source vehicle, such as vans and semi-trailers, grain type, entry time, and the basic sampling grid obtained when the batch of grain entered the warehouse and the status of each grid at that time, such as whether there were any rapid detection anomalies), as well as the temperature and humidity change curves in the warehouse and ventilation operation records for a period of time before the risk occurred.

[0021] The specific process of spatial mapping overlay analysis is as follows: When a grain transport vehicle is detected entering the warehouse, its three-dimensional point cloud data is acquired using lidar and a basic sampling network is generated. Using the current vehicle type and grain type as query conditions, all historical risk situations of the same type are retrieved from the historical risk database. Based on the historical risk situation, for each basic sampling grid point of the current grain pile, the three-dimensional spatial straight-line distance between it and each historical risk point is calculated. The contribution value of each basic sampling grid point is obtained by inverse distance weighted (IDW) interpolation. The sum of the contribution values ​​of each basic sampling grid point is recorded as the risk probability of each basic sampling grid.

[0022] Based on risk probability, the basic sampling grids are sorted from high to low, and the basic sampling grid with the highest risk probability is selected as the initial target sampling point. If there are basic sampling grids with the same maximum risk probability, they are also used as initial target sampling points. Based on the spatial uniform distribution constraint rule, spatial supplementation is performed on the initial target sampling points to obtain the final target sampling points, and joint feature extraction and fusion analysis are then performed. The spatial uniform distribution constraint rule is used to ensure that the distribution of the target sampling point set within the grain pile is statistically representative. Specifically, the target sampling points must cover the upper / middle / lower and front / back of the grain pile to avoid the sampling area being concentrated in a localized area.

[0023] Specifically, the joint feature extraction and fusion analysis process is as follows: During the sampling operation, the near-infrared spectrometer and digital temperature sensor on the intelligent sampling probe collect the near-infrared spectral sequence and temperature data sequence of the target sampling point during the sampling operation; the near-infrared spectral sequence is obtained during the sampling process at a first preset sampling frequency, and the near-infrared spectral sequence includes the spectrum at the time of acquisition and the first timestamp of the corresponding spectral point storage acquisition time; the first preset sampling frequency is the result of summing and averaging the frequencies of historically acquired near-infrared spectral sequences in historical sampling processes.

[0024] The temperature data sequence is obtained during the sampling process at a second preset sampling frequency not lower than the first preset sampling frequency. The temperature data sequence includes the temperature at the time of sampling and the second timestamp of the corresponding temperature data point storage time. The second preset sampling frequency is the result of summing and averaging the frequencies of historical temperature data sequences collected during historical sampling processes.

[0025] Based on the time dimension corresponding to the first and second timestamps, the near-infrared spectral sequence and temperature data sequence are time-synchronized to obtain an aligned data stream. The specific process of time-synchronization calibration is as follows: based on the received first timestamp and its corresponding data stream, and the second timestamp and its corresponding data stream, a linear interpolation algorithm is used to calculate the aligned data stream for the two data streams, which can effectively compensate for the small time offset caused by the difference in fixed sampling frequency.

[0026] The aligned data stream is divided into short-time data slices based on a preset slice duration threshold, and a multi-dimensional feature vector is constructed. The preset slice duration threshold is jointly determined by the probe insertion speed and the smallest spatial scale of the grain pile to be analyzed. For example, if the probe speed is 0.3 m / s, a slice duration of 0.2 seconds corresponds to the path of the probe moving about 6 cm in the grain pile. The short-time data slice represents the smallest data unit of the aligned data stream within a preset time window, containing synchronous spectral and temperature features.

[0027] Furthermore, a multidimensional feature vector is constructed. The specific process is as follows: For each short-time data slice, spectral feature parameters are extracted to characterize the dynamic changes in the chemical state of grain. The spectral feature parameters include the first derivative curve of the near-infrared spectral subsequence within a preset band range (the first derivative is calculated by applying the convolution kernel of the filter to the smoothed near-infrared spectral sequence to obtain the curve characterizing the rate of spectral change; the filter is the Savitzky-Golay filter, which can effectively suppress noise while preserving the original shape features of the spectrum, such as peak position and peak width, to the maximum extent), and the standard deviation of the corresponding spectral subsequence is statistically analyzed. The preset band range is the characteristic absorption band of the component most relevant to grain quality and safety selected by the preset personnel.

[0028] From the same short-time data slice, temperature characteristic parameters for characterizing grain thermodynamic processes are extracted. These parameters include the slope of a linear regression model representing the average temperature change rate over the slice's time interval (using the least squares method, with time as the independent variable and temperature as the dependent variable, the resulting slope represents the average temperature change rate), and the standard deviation of the corresponding temperature subsequence obtained through statistical analysis. The first derivative amplifies subtle chemical composition changes in the spectrum; the linear slope quantifies heat generation / dissipation trends; and the standard deviations of both reflect the intensity of fluctuations. These parameters synergistically capture the unique, weak, and coupled chemical and thermal signals of early-stage risks.

[0029] After normalizing all the extracted spectral and temperature feature parameters, they are concatenated into a one-dimensional array in ascending order of timestamps. This array represents the multi-dimensional feature vector of the grain state during that specific time interval (corresponding to a microscopic path in the grain pile). It simultaneously encodes the coupled information of the near-infrared spectrum and temperature of the grain at that point, providing highly informative input for subsequent differentiation of different risk sources and enabling preliminary grain risk assessment. Figure 3 The flowchart shown is for the preliminary grain analysis and judgment provided in the embodiment of the invention. Specifically, the multidimensional feature vector is input into a preset sampling state-probability distribution network structure, the risk probability of the target grain sample corresponding to the risk category is output, and early risk differentiation is performed. If the risk probability of any basic sampling grid exceeds the preset first risk probability threshold, the target grain sample corresponding to the basic sampling grid is judged to have an early risk, and the corresponding risk disaster category label and the corresponding risk probability are output. The first risk probability threshold is the 95th percentile (P95) of the highest risk probability value in the historical risk classification. This means that 95% of the empirically verified safe grain predictions have a risk probability lower than this value. The specific value is fine-tuned by the preset personnel according to the current application scenario and is not fixed.

[0030] If the risk probability of all basic sampling grids does not exceed the preset first risk probability threshold, but the risk probability of a certain basic sampling grid exceeds the preset second risk probability threshold, then the target grain sample corresponding to the basic sampling grid is determined to be a risk to be verified, and the risk disaster category label and corresponding risk probability are output in the historical risk database; the second risk probability threshold is determined by the median (50th percentile) or the lower quartile (25th percentile) of the highest risk prediction probability in the historical risk classification.

[0031] If the risk probability of all target grain samples does not exceed the preset second risk probability threshold, then the target grain samples in the short-time data slice are determined to have no early risk and are marked as risk-free. After the early risk is identified, the grains are classified and stored, and the value of the first risk probability threshold is greater than the value of the second risk probability threshold.

[0032] The specific process for obtaining the preset sampling state-probability distribution network structure is as follows: using historically accumulated multidimensional feature vectors with clearly labeled risk categories as training samples, multi-classification algorithms such as support vector machines, random forests, or deep neural networks are trained and cross-validated, enabling the model to learn the unique patterns corresponding to different risk sources (such as early mold, early pest activity, and increased grain respiration) in the multidimensional feature space jointly constructed by spectral fluctuation features and temperature evolution features (referring to the dynamic process of temperature change extracted from temperature data sequences that can quantify the temperature change), thereby enabling the model to classify risk sources based on newly extracted fusion features.

[0033] like Figure 4 The diagram shows a flowchart of the risk association analysis provided in this embodiment of the invention. The specific process for classifying grain storage is as follows: During the grain storage stage, based on the risk hazard category label of the target grain sample in the grain pile in the current grain warehouse, the temperature data and carbon dioxide concentration of each monitoring point are obtained at fixed points according to the preset collection cycle and spatial sampling density. The preset collection cycle and spatial sampling density are set by preset personnel according to the requirements in the relevant technical specifications.

[0034] The correlation coefficient between the first derivative of temperature data and the first derivative of carbon dioxide concentration was calculated using the Pearson correlation coefficient formula. Simultaneously, the time offset between the two data sequences was measured. After normalization, the deunited correlation coefficient and time offset were obtained. The time offset represents the difference in the starting time between the two data sequences. The time offset was then divided by a preset time offset to obtain the time offset rate, which is the summed and averaged result of historical time offsets in historical risk association analysis. In historical grain storage, a multinomial regression was performed with the correlation coefficient and time offset rate as independent variables and the historical grain storage risk pattern as the dependent variable. The normalized regression coefficients are the weighted coefficients corresponding to the correlation coefficient and time offset.

[0035] The product of the correlation coefficient and its weight coefficient, and the product of the time offset rate and its weight coefficient, are summed to generate a predicted risk index. This index is then matched and filled into the spatial location of the corresponding monitoring and collection point for risk correlation analysis. Specifically, the accumulation depth of the grain pile in the current grain warehouse is obtained for each monitoring and collection point. The accumulation depth represents the vertical length of the grain pile in the current grain warehouse from the grain surface.

[0036] If the accumulation depth is less than the preset first accumulation depth, the area within the grain warehouse to which the corresponding monitoring point belongs is recorded as the first longitudinal monitoring area; if the accumulation depth is between the preset first accumulation depth and the preset second accumulation depth (including when it is equal to the first accumulation depth and when it is equal to the second accumulation depth), the area within the grain warehouse to which the corresponding monitoring point belongs is recorded as the second longitudinal monitoring area.

[0037] If the pile depth is greater than the preset second pile depth, the area within the grain warehouse to which the corresponding monitoring point belongs is recorded as the third longitudinal monitoring area. The first and second pile depths are the results of statistical analysis. In the process of historical risk correlation analysis, the 95th percentile of the historical first and second pile depths is selected as the preset first and second pile depths. These three longitudinal monitoring areas are divided based on the depth of the grain pile, corresponding to the surface, middle, and bottom layers, respectively, with the first pile depth being higher than the second pile depth. Grain piles at different depths face drastically different physical and biological risk mechanisms: the surface layer is easily affected by environmental fluctuations, the middle layer is the core reflecting the overall condition of the grain pile, and the bottom layer is prone to accumulating moisture and heat. This division can identify the location of the root cause of the risk and implement highly targeted differentiated environmental control strategies, thereby achieving precise intervention in local risks.

[0038] The upper quartiles of the predicted risk indicators for all monitoring points within each longitudinal monitoring area are obtained through statistical analysis, and these upper quartiles are used as the regional risk assessment values ​​for the longitudinal monitoring areas. If the regional risk assessment value is greater than the preset regional risk indicator threshold, it indicates that the corresponding longitudinal monitoring area is unqualified, and the grain storage environment is regulated. The preset regional risk indicator threshold is represented by the result of summing and averaging the regional risk assessment values ​​during the historical risk correlation analysis process. Conversely, if the threshold is lower than the threshold, it indicates that the corresponding longitudinal monitoring area is qualified, and a command indicating no abnormal behavior is sent.

[0039] When the regional risk assessment value of a certain longitudinal monitoring area exceeds the preset regional risk indicator threshold, the specific process of grain warehouse environmental control is as follows: When the regional risk assessment value of the first longitudinal monitoring area is unqualified, the current ventilation and humidity adjustment value of the grain warehouse is calculated based on the deviation of the regional risk assessment value and the pre-built dew point tracking and control relationship. The humidity of the grain warehouse environment is then adjusted according to this adjustment value to reduce abnormal grain problems caused by temperature and humidity fluctuations. The pre-built dew point tracking and control relationship is established through a combination of historical data learning and physical rules. Historical cases are collected, including the air temperature inside the warehouse, relative humidity (used to calculate the real-time dew point temperature), grain pile surface temperature, regional risk assessment deviation, and the final effective ventilation and humidity adjustment amount. The regional risk assessment deviation is the first longitudinal monitoring value. The difference between the regional risk assessment value and the preset regional risk assessment value is used to represent the risk assessment area. Based on this data, multiple linear regression or decision tree algorithms are used for training, with regional risk assessment deviation as input features and historically effective ventilation and humidity adjustment amounts as learning objectives. The essence of this training process is to allow the algorithm to summarize the mapping relationship between regional risk assessment deviation and the ventilation and humidity adjustment amounts that have been verified in practice and can stabilize the environment from historical pairings (regional risk assessment deviation and corresponding verified effective ventilation and humidity adjustment amounts). The final model can output a parameter recommendation value based on historical best practices according to the real-time input risk deviation.

[0040] When the regional risk assessment value of the second longitudinal monitoring area is unqualified, the actual accumulation depth of the monitoring collection point in that area is input into a preset gas adjustment mapping relationship to solve for the nitrogen concentration adjustment value in the grain silo. The nitrogen supply is then adjusted according to this value to suppress grain respiration and reduce grain anomalies. The preset gas adjustment mapping relationship is a mathematical model trained on historically validated effective gas adjustment case data. Each historical case data package contains two parts: first, state characteristics (including the accumulation depth, regional risk assessment deviation, and average temperature of the grain pile at that time); second, the effective control action taken for that state (i.e., the nitrogen concentration adjustment value). The training process of the model (such as support vector regression or a neural network) essentially learns the complex mapping rules between state characteristics and effective control actions. Therefore, when new monitoring data is input, the model recommends a nitrogen concentration adjustment value that is most likely to succeed under similar historical conditions, based on the learned rules.

[0041] When the regional risk assessment value of the third longitudinal monitoring area is unqualified, the target heat dissipation speed of the directional ventilation duct is calculated and determined by combining the actual stacking depth of the area with the risk assessment deviation. The operating parameters of the grain silo's directional ventilation duct are then adjusted according to this heat dissipation speed to enhance heat dissipation from deep grain piles and effectively reduce the risk of heat accumulation. The target heat dissipation speed is determined using an empirical formula based on heat load estimation; the core parameters are the current area's risk assessment deviation (reflecting the intensity of thermal anomalies) and stacking depth (affecting thermal resistance and airflow resistance). These two parameters are substituted into a pre-calibrated formula (speed base × (1 + risk assessment deviation coefficient) × depth compensation coefficient) to directly calculate the required target speed. The specific form and parameters of the empirical formula are determined by nonlinear regression fitting of data from historical effective control cases. The specific process is as follows: collect historically verified effective fan speed adjustment amounts and their corresponding accumulation depth and regional risk assessment deviation data; use depth and risk deviation as independent variables and effective speed as dependent variable, and determine the formula through regression algorithm fitting. Similarly, the coefficients in the formula are obtained by nonlinear regression fitting through analysis of the statistical correlation between regional risk assessment values ​​and control intensity (such as speed adjustment amounts) in historically verified effective control cases by preset personnel; ensure that the calculated speed can provide sufficient air volume to offset the heat load at that depth.

[0042] Because the first, second, and third longitudinal monitoring areas correspond to the surface, middle, and bottom storage areas in the grain warehouse, respectively, the surface grain pile is the only interface where condensation risk occurs due to direct heat and mass exchange with the air inside the warehouse, and is greatly affected by environmental fluctuations. In the middle grain pile, because the environment is relatively stable, any risks that arise often stem from the grain itself and biochemical processes such as respiration of microorganisms. The diffusion rate of gas (carbon dioxide) in the grain pile is much slower than heat transfer. By regulating the nitrogen concentration to reduce the oxygen content, the biological activity of the entire middle layer can be inhibited from a biochemical perspective. This is an intervention targeting volumetric and process-related biological risks.

[0043] In the bottom storage area, the bottom grain pile is a place where heat and moisture tend to accumulate. Due to gravity, hot and humid air is difficult to rise and diffuse, and the risks are mostly physical heat and moisture accumulation, which can further lead to mold growth. By adjusting the heat dissipation speed of the directional ventilation duct, the airflow intensity and ventilation efficiency in this local area can be enhanced, and the accumulated heat and moisture can be discharged through forced convection.

[0044] Taking the transportation of newly harvested wheat from a van to a grain depot for storage as an example, during the sampling process, historical risk distribution data for the same vehicle type and grain type are used to predict and calculate that the rear and corners of the van are areas with a higher probability of risk. Based on this result, the intelligent sampling equipment prioritizes the placement of sampling points in these high-risk areas, while supplementing sampling points in other parts of the grain pile according to the uniformity rule. In this operation, the sample obtained from a high-risk grid point in the right rear corner of the van was the first to show an abnormal signal in subsequent analysis.

[0045] In Embodiment 1 of this invention, through fusion feature analysis of the abnormal sample during the data analysis and diagnostic phase, it was found that the near-infrared spectrum exhibited a weak but persistent distortion in a specific wavelength band, and the temperature at this location showed an abnormal nonlinear increasing trend during probe insertion. After the constructed fusion feature vector was input into the prediction model, the output indicated a high probability of a combination of early fungal activity and localized insect infestation risk at this location.

[0046] In the classification, storage, and intelligent control phase, based on the aforementioned precise assessment, this batch of grain marked as high-risk was directed to a dedicated grain warehouse equipped with enhanced monitoring and active environmental control. At the initial storage stage, the dynamic grain risk map showed a coordinated upward trend in temperature and carbon dioxide concentration in the local area where this batch of grain was located. Targeted control strategies were then initiated: firstly, ventilation was adjusted by introducing moderately cooled and dehumidified air to disrupt the microenvironment conducive to mold growth; simultaneously, the airflow velocity was increased in the ventilation branches corresponding to the risk area to eliminate localized heat accumulation. When pest activity was not detected as being suppressed, low-dose circulating fumigation was further initiated in the grain warehouse area, with gas concentration monitored and controlled in a closed-loop manner in real time.

[0047] Through continuous operations involving precise detection, early diagnosis, and targeted intervention, the risks carried by the stored grain were effectively contained before they developed into large-scale risks. The treatment process was highly targeted, avoiding the high energy consumption and grain quality damage caused by whole-warehouse fumigation or cooling, and achieving the goal of maintaining grain quality to the greatest extent while ensuring grain storage safety. This demonstrates a new paradigm of intelligent management that transforms risk control from a passive response to a proactive, precise, and predictable approach.

[0048] like Figure 5 The diagram shown is a structural schematic of a grain condition monitoring system for grain quality and safety provided in an embodiment of the present invention. The grain condition monitoring system for grain quality and safety provided in this embodiment of the present invention includes the following modules: The target sampling and acquisition module is used to perform multi-target three-dimensional sampling of the grain pile inside the grain transport vehicle during the grain transport vehicle's entry into the warehouse to obtain target sampling points, and to collect corresponding multi-dimensional sensor data at each target sampling point.

[0049] The preliminary grain analysis and judgment module is used to construct a multidimensional feature vector to characterize the early state of grain at the target sampling point by performing joint feature extraction and fusion analysis on multidimensional sensor data, and to make preliminary grain risk judgment and early differentiation of risk sources based on the multidimensional feature vector.

[0050] The grain condition prediction analysis module, based on the early differentiation of risk sources, classifies and stores the target grain samples corresponding to each target sampling point. At the same time, during the grain storage process, it collects gas and temperature data of the grain warehouse in real time, performs grain condition prediction analysis, and generates a dynamic grain condition risk structure.

[0051] The risk correlation analysis module is used to perform risk correlation analysis based on the dynamic grain condition risk structure, generate abnormal operation early warning signals, detect abnormal behavior during grain storage, and control the grain storage environment.

[0052] In Example 1, during the monitoring of grain storage, the grain warehouse may experience fundamental problems (such as continuous exposure to extreme external weather) causing multiple longitudinal monitoring areas to simultaneously exceed risk limits for an extended period. This means that when all longitudinal monitoring areas within the grain warehouse exhibit excessive risk, the regional risk assessment values ​​for all longitudinal monitoring areas exceed the preset regional risk indicator threshold. In this case, the specific process of grain warehouse environmental control in Example 2 is as follows: The regional risk assessment values ​​of each longitudinal monitoring area are input into the pre-built grain warehouse environment control model, and the global adjustment parameters of the grain warehouse (including ventilation humidity, nitrogen concentration and heat dissipation speed) are output. At the same time, the risk status of the grain warehouse is issued on the designated platform. Within the preset monitoring time window, the regional risk assessment values ​​are collected in real time and the risk change rate is calculated. If the risk change rate is greater than the corresponding preset value, the current global adjustment parameters are maintained. Otherwise, the preset personnel are immediately prompted to investigate the grain warehouse environment to confirm the control effect and potential hazards.

[0053] The preset monitoring time window is the average of historical monitoring time windows for grain warehouse environmental control. The grain warehouse environmental control model is trained using a supervised learning framework, combined with deep neural networks and reinforcement learning for optimization. The specific process is as follows: a training dataset is constructed based on historical grain warehouse control records, where the input features are multi-dimensional vectors; the output labels are the optimal combination of control parameters (ventilation humidity, nitrogen concentration, and heat dissipation speed) after verification; a deep fully connected network is used to initially fit the nonlinear mapping relationship between features and labels; to further improve the dynamic adaptability of the model, reinforcement learning is introduced for online fine-tuning: the control is modeled as a Markov decision process, with the rate of risk change as the reward signal. A deep deterministic policy gradient algorithm is used to allow the model to learn action strategies for automatically adjusting parameters under different risk states in the simulation environment, ultimately forming a grain warehouse environmental control model that can output adaptive control parameters based on real-time risk conditions and has continuous optimization capabilities.

[0054] In this embodiment, an integrated grain storage environment control model outputs a globally coordinated combination of environmental parameters (ventilation, nitrogen filling, and heat dissipation) to rapidly and powerfully intervene in the grain storage environment, aiming to control the evolution of risks from a macroscopic perspective. Simultaneously, a reinforcement learning fine-tuning mechanism is introduced, which dynamically judges the effectiveness of the control based on the real-time rate of risk change; this avoids the mutually exclusive or inefficient problems that arise from using multiple local strategies in a global emergency, ensuring the overall stability of grain storage safety with the highest priority.

[0055] like Figure 6 The diagram shown illustrates the relationship between actual ventilation humidity and corresponding predicted values ​​provided in an embodiment of the present invention. Figure 7 The figure shown is a graph illustrating the relationship between the actual nitrogen concentration and the corresponding predicted value provided in an embodiment of the present invention. Figure 8 The figure shown is a graph illustrating the relationship between the actual heat dissipation rotation speed and the corresponding predicted value provided in an embodiment of the present invention. Figure 6 , Figure 7 and Figure 8 These scatter plots illustrate the difference between the actual adjusted values ​​of global parameters and the predicted values ​​under ideal conditions in the grain storage environment control model. The scatter points in the plots represent the real-time ventilation humidity, real-time nitrogen concentration, and real-time heat dissipation speed collected. The actual and predicted values ​​of ventilation humidity, nitrogen concentration, and heat dissipation speed are compared respectively. The closer the scatter points are to the red diagonal, the smaller the deviation between the predicted and actual values, reflecting the accuracy and reliability of the model in controlling the core parameters.

[0056] like Figure 9The diagram shows the relationship between the regional risk assessment value and the grain warehouse environmental control model provided in this embodiment of the invention. The horizontal axis represents the monitoring timestamp, and the vertical axis represents the specific value of the regional risk assessment value. When the grain warehouse environmental control model intervenes to regulate the grain warehouse environment, the output of the globally coordinated environmental parameter combination can optimize and control the regional risk assessment value. It should be further understood that the values ​​in the figure are only illustrative and do not represent that the grain warehouse environmental control model will intervene under these values. The dashed line in the figure indicates that when a risk disaster is detected in the grain warehouse, the regional risk assessment value will increase with the monitoring time. The solid line indicates that the regional risk assessment value increases with time. After the grain warehouse environmental control model intervenes, the regional risk assessment value will decrease with time, indicating that the regional risk has been effectively controlled.

[0057] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0058] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0059] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0060] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0061] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0062] In the embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0063] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0064] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0065] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0066] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for detecting grain condition to ensure grain quality and safety, characterized in that, Includes the following steps: During the process of grain transport vehicles entering the warehouse, multi-target three-dimensional sampling is performed on the grain pile inside the grain transport vehicles to obtain target sampling points, and corresponding multi-dimensional sensor data is collected at each target sampling point. By performing joint feature extraction and fusion analysis on the multidimensional sensing data, a multidimensional feature vector is constructed to characterize the early state of grain at the target sampling point, and preliminary grain risk assessment and early differentiation of risk sources are made based on the multidimensional feature vector. Based on the early differentiation of risk sources, the target grain samples corresponding to each target sampling point are classified and stored. At the same time, during the grain storage process, gas and temperature data of the grain warehouse are collected in real time to conduct predictive analysis of grain conditions and generate a dynamic grain condition risk structure. Based on the dynamic grain condition risk structure, risk correlation analysis is performed to generate abnormal operation early warning signals. At the same time, abnormal behavior is detected during grain storage and the grain storage environment is controlled. The specific process for the classified grain storage is as follows: During the grain storage stage, based on the risk hazard category label of the target grain sample in the grain pile in the current grain warehouse, temperature data and gas concentration data of each monitoring and collection point are obtained at fixed points according to the preset collection cycle and spatial sampling density. Calculate the correlation coefficient between the first derivative of the temperature data and the first derivative of the gas concentration data, and simultaneously measure the time offset between the two types of data sequences; After weighted coupling of the correlation coefficient and the time offset, a predicted risk index is generated. This index is then matched and filled into the spatial location of the corresponding monitoring and collection point for risk correlation analysis.

2. The grain condition detection method for grain quality and safety as described in claim 1, characterized in that, The specific process of the multi-target three-dimensional sampling is as follows: Data on grain transport vehicles entering the warehouse is acquired and spatially mapped and overlaid with data stored in the historical risk database to obtain the risk probability of each basic sampling grid. The basic sampling grid is a unit grid in the three-dimensional space of the grain pile corresponding to the sampling operation range during the grain transport vehicle entering the warehouse. Based on the risk probability, each basic sampling grid is sorted from high to low, and the basic sampling grid with the highest risk probability is taken as the initial target sampling point. If there are basic sampling grids with the same maximum risk probability, they are also taken as the initial target sampling points. Based on the spatial uniform distribution constraint rule, the initial target sampling points are spatially supplemented and filtered to obtain the final target sampling points, and joint feature extraction and fusion analysis are performed.

3. The grain condition detection method for grain quality and safety as described in claim 2, characterized in that, The joint feature extraction and fusion analysis process is as follows: Near-infrared spectral and temperature data sequences of the target sampling points during the sampling process were collected. The near-infrared spectral sequence is obtained during the sampling process at a first preset sampling frequency, and the spectral points corresponding to each near-infrared spectral sequence are stored with the first timestamp of the acquisition time. The temperature data sequence is obtained during the sampling process at a second preset sampling frequency not lower than the first preset sampling frequency, and the temperature data points corresponding to each temperature data sequence are stored as the second timestamp of the sampling time. Based on the time dimension corresponding to the first timestamp and the second timestamp, the near-infrared spectral sequence and the temperature data sequence are time-series synchronized and calibrated to obtain an aligned data stream; Using a preset slice duration as the slice duration interval, the aligned data stream is divided into short-time data slices within the slice duration interval, and a multi-dimensional feature vector is constructed. The short-time data slice represents the smallest data unit of the aligned data stream within a preset slice duration, containing synchronized spectral and temperature features. Within the short-time data slice, there is a near-infrared spectral subsequence corresponding to the spectral features and a temperature subsequence corresponding to the temperature features.

4. The grain condition detection method for grain quality and safety as described in claim 3, characterized in that, The specific process for constructing the multidimensional feature vector is as follows: For each short-time data slice, spectral feature parameters are extracted to characterize the dynamic changes in the chemical state of grain. The spectral feature parameters include the first derivative curve of the near-infrared spectral subsequence within a preset band range, and the standard deviation of the corresponding near-infrared spectral subsequence. From the same short-time data slice, temperature characteristic parameters for characterizing the thermodynamic process of grain are extracted. The temperature characteristic parameters include the slope of the linear fitting for characterizing the average temperature change rate within the slice time interval, and the standard deviation of the corresponding temperature subsequence. After normalizing the spectral and temperature characteristic parameters, they are spliced ​​together in a preset fixed order to obtain a multidimensional feature vector that characterizes the synergy between changes in grain chemical properties and thermodynamic state, and a preliminary grain risk assessment is then performed.

5. The grain condition detection method for grain quality and safety as described in claim 4, characterized in that, The preliminary grain risk assessment process is as follows: The acquired multidimensional feature vectors are input into a preset sampling state-probability distribution network structure, which outputs the risk probability of the target grain sample corresponding to the risk category and performs early risk differentiation. If there is a target grain sample whose risk probability exceeds the preset first risk probability threshold, the corresponding target grain sample is determined to have an early risk, and the corresponding risk disaster category label and the corresponding risk probability are output. If the risk probability of all target grain samples does not exceed the preset first risk probability threshold, but the risk probability of a certain target grain sample exceeds the preset second risk probability threshold, it is determined to be a risk to be verified, and the corresponding risk disaster category label and the corresponding risk probability are output. If the risk probability of all target grain samples does not exceed the preset second risk probability threshold, then the target grain samples of the short-time data slice are determined to have no early risk and are marked as risk-free. After the initial risk classification is completed, the grain is stored in categories, and the value of the first risk probability threshold is greater than the value of the second risk probability threshold.

6. The grain condition detection method for grain quality and safety as described in claim 1, characterized in that, The risk association analysis is performed as follows: The accumulation depth of the grain pile in the current grain warehouse is obtained at each monitoring point, and the accumulation depth represents the vertical length of the grain pile in the current grain warehouse. If the accumulation depth is less than the preset first accumulation depth, then the area within the grain warehouse to which the corresponding monitoring and collection point belongs is recorded as the first longitudinal monitoring area. If the accumulation depth is between the preset first accumulation depth and the preset second accumulation depth, then the area within the grain warehouse to which the corresponding monitoring and collection point belongs is recorded as the second longitudinal monitoring area. If the accumulation depth is greater than the preset second accumulation depth, then the area within the grain warehouse to which the corresponding monitoring and collection point belongs is recorded as the third longitudinal monitoring area. Obtain the statistical characteristic values ​​of the predicted risk indicators of all monitoring and collection points in each longitudinal monitoring area, and use them as the regional risk assessment values ​​for the longitudinal monitoring area. If the regional risk assessment value is greater than the preset regional risk index threshold, it indicates that the corresponding longitudinal monitoring area is unqualified, and the grain warehouse environment will be regulated. Conversely, if the corresponding longitudinal monitoring area is qualified, a command indicating no abnormal behavior will be sent.

7. The grain condition detection method for grain quality and safety as described in claim 6, characterized in that, When the duration for which the regional risk assessment value of a certain longitudinal monitoring area exceeds the preset regional risk indicator threshold is not less than the preset duration, the specific process of grain warehouse environmental control is as follows: When the regional risk assessment value of the first longitudinal monitoring area is not up to standard, the current ventilation humidity adjustment value of the grain warehouse is calculated based on the deviation of the regional risk assessment value of the area and combined with the pre-constructed dew point tracking and control relationship. The humidity of the grain warehouse environment is then controlled according to the adjustment value. When the regional risk assessment value of the second longitudinal monitoring area is not qualified, the actual accumulation depth of the monitoring collection point in the area is input into the preset gas adjustment mapping relationship to solve for the nitrogen concentration adjustment value in the grain warehouse. Then, the nitrogen supply is adjusted according to the adjustment value to inhibit grain respiration. When the regional risk assessment value of the third longitudinal monitoring area is not up to standard, the target heat dissipation speed of the directional ventilation duct is calculated and determined by combining the actual accumulation depth of the area with the risk assessment deviation, and then the operating parameters of the grain warehouse directional ventilation duct are adjusted according to the heat dissipation speed.

8. The grain condition detection method for grain quality and safety as described in claim 6, characterized in that, When the duration for which the risk assessment values ​​of non-single longitudinal monitoring areas all exceed the preset regional risk indicator threshold is not less than the preset duration, the specific process of grain warehouse environmental control is as follows: The regional risk assessment values ​​of each longitudinal monitoring area currently acquired are input into the pre-built grain warehouse environment control model, outputting the global adjustment parameters of the grain warehouse, and issuing a grain warehouse risk status prompt on the designated platform. Within a preset monitoring time window, regional risk assessment values ​​are collected in real time and the rate of risk change is calculated. If the rate of risk change is greater than the corresponding preset value, the current global adjustment parameters are maintained. Conversely, if the situation does not improve, the designated personnel will be immediately alerted to inspect the grain storage environment to confirm the effectiveness of the control measures and identify any potential hazards.

9. A grain condition monitoring system for grain quality and safety, comprising the following modules: The target sampling and acquisition module is used to perform multi-target three-dimensional sampling of the grain pile inside the grain transport vehicle during the grain transport vehicle's entry into the warehouse to obtain target sampling points, and to collect corresponding multi-dimensional sensor data at each target sampling point. The preliminary grain analysis and judgment module is used to construct a multidimensional feature vector to characterize the early state of grain at the target sampling point by performing joint feature extraction and fusion analysis on the multidimensional sensor data, and to perform preliminary grain risk judgment and early differentiation of risk sources based on the multidimensional feature vector. The grain condition prediction analysis module, based on the early differentiation of risk sources, classifies and stores the target grain samples corresponding to each target sampling point. At the same time, during the grain storage process, it collects gas and temperature data of the grain warehouse in real time, performs grain condition prediction analysis, and generates a dynamic grain condition risk structure. The risk correlation analysis module is used to perform risk correlation analysis based on the dynamic grain condition risk structure, generate abnormal operation early warning signals, detect abnormal behavior during grain storage, and regulate the grain storage environment.

Images