Granary grain mildew early warning method, system, equipment and storage medium

By obtaining grain category and storage environment data in the granary, optimizing the sensor array layout, and identifying the VOCs marks during grain mold, we achieve efficient and accurate early warning of grain mold, and solving the long detection cycle and low accuracy in the existing technology. The problem ensures food security.

CN118657625BActive Publication Date: 2024-11-05BEIJING LIANGAN TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411142561.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-20
Publication Date
2024-11-05
Estimated Expiration
2044-08-20

AI Technical Summary

Technical Problem

In the existing technology, the mold detection cycle of grain grains in granaries is long and the accuracy is not high. It is impossible to warn mold in time, which poses safety hazards.

Method used

By obtaining grain category and storage environment data in the granary, we determine the typical VOCs marks of mold release, optimize the sensor array layout, obtain real-time response signals, perform feature extraction and identification, and combine preset fingerprint library and concentration estimation model to determine grain mold warnings. level.

Benefits of technology

It improves the accuracy and timeliness of mold warnings, provides targeted anti-mold measures to ensure food security.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118657625B_ABST
    Figure CN118657625B_ABST
Patent Text Reader

Abstract

The present application provides a method, system, device and storage medium for warning of mildew of grain in a granary, and relates to the technical field of agricultural product quality analysis. The method determines the main mold categories and typical VOCs markers released by grain mildew by obtaining the categories of grain stored in the granary and the storage environment data. Then, the real-time response signal data of each sensor is obtained, pre-processed to obtain the target real-time response signal, feature extraction is performed to obtain the feature vector, and the classification result is input into the preset recognition model. The classification result is matched with the preset VOCs fingerprint library to determine the currently detected VOCs category and the corresponding concentration of each category. The concentration corresponding to each VOCs category is input into the preset VOCs concentration estimation model to obtain the absolute concentration of each VOCs category, and the warning level of mildew of grain in the granary is determined based on the absolute concentration of each VOCs category and the preset threshold. This method improves the accuracy and timeliness of mildew warning.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of agricultural product quality analysis, and specifically to a method, system, equipment and storage medium for early warning of mildew of grain in a granary. Background Art

[0002] As the main food source for human survival, the safe storage of grain has always been a top priority in the agricultural field. During the storage process, due to the inevitable mold contamination, grains often become moldy and deteriorate, and even produce toxic substances, which bring serious harm to human health. Therefore, it is of great practical significance to establish an efficient online monitoring and early warning system for grain mildew in granaries.

[0003] Existing technologies usually use regular manual sampling and testing to prevent and determine whether grain in the granary has mold. Specifically, a small amount of grain samples are taken from different locations in the granary on a regular basis, and then the type and quantity of mold in the samples are detected through laboratory analysis such as the drying loss method and colony counting method, and the mold risk level is assessed accordingly. However, this method has the problems of long detection cycle, low detection accuracy and failure to provide timely mold warning. Summary of the invention

[0004] The present application provides a method, system, equipment and storage medium for early warning of mildew of grain in a granary, so as to improve the accuracy and timeliness of early warning of mildew.

[0005] In a first aspect, the present application provides a method for early warning of mildew of grain in a granary, the method comprising:

[0006] Obtaining the type of grain stored in the granary and the grain storage environment data, and determining the typical VOCs markers released by mold when the grain is moldy according to the type of the stored grain and the grain storage environment data;

[0007] Obtaining a storage structure schematic diagram inside the granary, and determining an optimal sensor array layout based on the storage structure schematic diagram;

[0008] Arranging sensors based on the optimal sensor array layout, acquiring real-time response signal data of each sensor, and preprocessing the real-time response signal data to obtain a target real-time response signal;

[0009] Extracting features of the target real-time response signal to obtain a feature vector, and inputting the feature vector into a preset recognition model to obtain the VOCs category corresponding to the target real-time response signal;

[0010] Determine the VOCs category according to the VOCs category corresponding to the target real-time response signal, the feature vector and a preset VOCs fingerprint library, and determine the concentration corresponding to each VOCs category according to the VOCs category and the target real-time response signal;

[0011] The concentration corresponding to each of the VOCs categories is input into a preset VOCs concentration estimation model to obtain the absolute concentration corresponding to each VOCs category, and the mildew warning level of grain in the granary is determined based on the absolute concentration corresponding to each of the VOCs categories and the preset concentration threshold.

[0012] By adopting the above technical scheme, by obtaining the type of grain stored in the granary and the grain storage environment data, combined with the grain variety and storage conditions, the type of VOCs that may be released when the grain is moldy is predicted, providing targeted indicators for subsequent VOCs detection. On this basis, by analyzing the internal storage structure of the granary, using three-dimensional modeling and optimization algorithms, the optimal layout scheme of the sensor array is obtained, ensuring the full spatial coverage of VOCs collection and the representativeness of monitoring data, laying the foundation for improving the accuracy of mold warning. When actually collecting VOCs data, high-quality target response signals are obtained by preprocessing the real-time response signals of the sensors such as denoising and filtering, reducing noise interference and providing effective input for subsequent data analysis. Then, the feature extraction algorithm is used to quantize the features of the target signal, and the pre-trained recognition model is used to classify it, accurately identifying the compound category to which the VOCs belong, and realizing the qualitative identification of VOCs. On the basis of qualitative analysis, the correlation between the feature vector and the preset VOCs fingerprint library was calculated, the specific category of VOCs was determined by pattern matching method, and the real-time concentration values ​​of various VOCs were converted according to the category-specific response-concentration curve, thus realizing the quantitative analysis of VOCs. Finally, the VOCs concentration estimation model was used to calibrate the various concentration values ​​to obtain more accurate absolute concentrations. By comparing with the preset threshold, the degree of mildew was determined and the corresponding warning level was given, providing a quantitative basis for grain depot managers to take anti-mildew measures in a timely manner.

[0013] In a second aspect of the present application, a method system for early warning of mildew of grain in a granary is provided, comprising:

[0014] A data acquisition module is used to obtain the category of grain stored in the granary and the grain storage environment data, and determine the main mold category and typical VOCs markers released by the mold when the grain is moldy according to the category of the stored grain and the grain storage environment data;

[0015] A layout optimization module, used to obtain a storage structure diagram inside the granary and determine an optimal sensor array layout based on the storage structure diagram;

[0016] A preprocessing module, used to arrange sensors based on the optimal sensor array layout, obtain real-time response signal data of each sensor, and preprocess the real-time response signal data to obtain a target real-time response signal;

[0017] A classification module, for extracting features of the target real-time response signal to obtain a feature vector, and inputting the feature vector into a preset recognition model to obtain the VOCs category corresponding to the target real-time response signal;

[0018] A matching module, for determining the VOCs category according to the VOCs category corresponding to the target real-time response signal, the feature vector and a preset VOCs fingerprint library, and determining the concentration corresponding to each VOCs category according to the VOCs category and the target real-time response signal;

[0019] The early warning module is used to input the concentration corresponding to each of the VOCs categories into a preset VOCs concentration estimation model to obtain the absolute concentration corresponding to each VOCs category, and determine the mildew early warning level of grain in the granary based on the absolute concentration corresponding to each of the VOCs categories and a preset concentration threshold.

[0020] In a third aspect of the present application, a computer storage medium is provided, wherein the computer storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor and executing the above method steps.

[0021] In the fourth aspect of the present application, an electronic device is provided, including a processor, a memory, a user interface and a network interface, the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the above method.

[0022] In summary, one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0023] 1. This application obtains the type of grain stored in the granary and the grain storage environment data, combines the grain variety and storage conditions, predicts the type of VOCs that may be released when the grain becomes moldy, and provides targeted indicators for subsequent VOCs detection. On this basis, by analyzing the internal storage structure of the granary, using three-dimensional modeling and optimization algorithms, the optimal layout of the sensor array is obtained, ensuring the full spatial coverage of VOCs collection and the representativeness of monitoring data, laying the foundation for improving the accuracy of mold warning.

[0024] 2. When determining the typical VOCs markers when grains are moldy, this application first determines the main mold categories when grains are moldy under these conditions based on the categories of stored grains and grain storage environment data by consulting relevant literature. This step uses existing scientific research accumulation and practical experience to screen out the dominant strains that are most likely to cause problems from a large number of grain storage molds, providing important clues for the subsequent determination of VOCs markers.

[0025] 3. This application makes full use of the prior knowledge and data accumulation of the fingerprint library, realizes the rapid qualitative analysis of VOCs through feature matching, avoids the blind search for unknown substances, and greatly improves the pertinence and efficiency of identification. On the basis of identifying the VOCs category, by querying the response-concentration calibration curve corresponding to the preset VOCs category and substituting the amplitude of the real-time response signal, the real-time concentration value of the VOCs can be converted. The calibration curve is a quantitative relationship between the response signal and the concentration obtained by fitting a large number of tests on VOCs samples of known concentration. The calibration curve can be used to achieve rapid quantitative analysis of VOCs concentration, avoiding complex chemical analysis steps and greatly simplifying the detection process. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A schematic diagram of a process for early warning of mildew of grain in a granary provided in an embodiment of the present application;

[0027] Figure 2 An architectural diagram of a grain mildew early warning system for a granary provided in an embodiment of the present application;

[0028] Figure 3 A schematic diagram of the structure of an electronic device provided in this application. DETAILED DESCRIPTION

[0029] In order to enable technicians in this field to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments.

[0030] In the description of the embodiments of the present application, words such as "for example" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "for example" or "for example" is intended to present related concepts in a specific way.

[0031] In the description of the embodiments of the present application, the meaning of the term "multiple" refers to two or more. For example, multiple systems refer to two or more systems, and multiple screen terminals refer to two or more screen terminals. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. The terms "include", "comprise", "have" and their variations all mean "including but not limited to", unless otherwise specifically emphasized.

[0032] In order to facilitate understanding of the method and system provided by the embodiments of the present application, before introducing the embodiments of the present application, the background of the embodiments of the present application is first introduced.

[0033] Grain is the main source of food for human survival, and ensuring safe grain storage has always been a top priority in the agricultural field. However, during the storage process in granaries, grain will inevitably be contaminated by mold, causing it to mold and deteriorate, and even produce toxic and harmful substances, posing a serious threat to human health. Therefore, establishing an efficient online monitoring and early warning system for grain mold in granaries is crucial to maintaining food security and protecting people's lives and health.

[0034] After the background introduction of the above content, those skilled in the art can understand the problems existing in the prior art. The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0035] Based on the above background technology, please refer to Figure 1 , Figure 1 A flowchart of a method for early warning of mildew of grain in a granary provided in an embodiment of the present application. The system can be implemented by a computer program or can be run as an independent tool application. Specifically, in an embodiment of the present application, the method can be applied on a server, but can also be applied on electronic devices such as a server. A method for early warning of mildew of grain in a granary includes the following steps:

[0036] S101, obtaining the type of grain stored in the granary and the grain storage environment data, and determining the typical VOCs marker released by mold when the grain is moldy based on the type of stored grain and the grain storage environment data.

[0037] Specifically, during the grain storage process, affected by the storage environment conditions, the grain is prone to quality problems such as mildew and insect infestation. Moldy grain not only reduces its nutritional value, but may also produce toxic and harmful substances, endangering the safety of humans and animals. However, before the grain becomes obviously moldy, it is difficult to observe the quality changes with the naked eye. Therefore, it is necessary to establish an effective grain quality monitoring and early warning method. Studies have found that when many grains become moldy, the metabolism of mold will release some characteristic volatile organic compounds (VOCs), which can be used as indicators of early mildew of grains. However, different types of grains have different types and concentrations of VOCs released when they become moldy, and external environmental conditions such as temperature and humidity will also affect the production of VOCs.

[0038] Based on the above embodiment, as an optional embodiment, the main mold category and the typical VOCs marker released by the mold when the grain is moldy are determined according to the category of the stored grain and the grain storage environment data, including:

[0039] S201 The category of stored grain and the grain storage environment data are queried in literature to determine the main mold category when the grain is moldy.

[0040] Specifically, when establishing a grain mold early warning model, we first need to identify the dominant mold groups that are most likely to cause grain mold under specific grain types and storage environments. This is because different types of grains have different physical structures and nutritional components, and are susceptible to different types of mold. In addition, external environmental conditions such as temperature and humidity can also affect the growth and reproduction of mold. Only by clarifying the main mold categories that cause grain mold under specific circumstances can we further determine the corresponding VOCs markers and lay the foundation for subsequent grain mold early warnings.

[0041] S202, querying a preset mold category-VOCs fingerprint database according to the main mold category when the grain is moldy, and determining the typical VOCs markers released by the mold.

[0042] Specifically, after determining the main mold categories that cause grain mildew, it is necessary to further clarify which typical VOCs substances these molds will release during their growth and metabolism, which is crucial for the subsequent early warning of grain mildew. By detecting characteristic VOCs components and their concentrations in the grain storage environment, mold growth can be discovered in time and the risk of grain mildew can be determined. However, the VOCs components released by different types of molds vary greatly. Only by selecting highly targeted markers can the specificity and reliability of mildew warning be improved. In order to efficiently screen typical VOCs markers released by molds, this scheme introduces a preset mold category-VOCs fingerprint database. This database systematically organizes the VOCs release characteristics of common grain mildew fungi by collecting published research data. By performing GC-MS analysis on the culture media of different types of molds, the composition and relative content of VOCs released by molds during their growth can be obtained. These data are brought together to form a VOCs fingerprint of molds. Generally, VOCs components with higher release and more stable release during the vigorous growth period of molds are selected as characteristic markers of the strain. Using this database, it is possible to quickly match mold categories with their characteristic VOCs markers. Once the main mold categories that may infect a batch of grain are identified, the corresponding VOCs marker types can be determined by simply searching the fingerprint data of the fungus in the database. For example, if it is determined that wheat is mainly infected with Aspergillus fungi, the VOCs fingerprint of Aspergillus can be retrieved from the database, and 1-octanol and 3-octanol are found to be its typical release products, which can be used as VOCs markers for wheat mildew.

[0043] S102, obtaining a schematic diagram of a storage structure inside a granary, and determining an optimal sensor array layout based on the schematic diagram of the storage structure.

[0044] Specifically, in order to obtain the optimal sensor array layout scheme, it is first necessary to obtain the storage structure diagram inside the granary. By consulting the granary building plan or performing three-dimensional modeling, the structural characteristics of the grain pile, such as the stacking form, partition layout, and ventilation channel location, can be intuitively reflected. This information provides a spatial reference for the formulation of the sensor point layout scheme. On the basis of clarifying the internal structure of the granary, it is necessary to further determine the optimal sensor array layout. Generally, the layered and partitioned point layout method is adopted, and the sensor collection nodes are arranged at key locations by comprehensively considering factors such as grain pile area division, sampling representativeness, and monitoring blind spots. For example, the grain pile is evenly divided into several small areas, and 1-2 monitoring points are set in each small area; sensors are arranged on the surface, middle, and lower layers of the grain pile to monitor the changes in grain temperature and moisture at different depths; the density of points is increased in areas where the environment is prone to drastic changes, such as ventilation ducts; and the number of points is appropriately reduced in areas where it is not easy to mold, such as the edge of the grain pile and dead corners. Through reasonable planning, while meeting the monitoring needs, sensor redundancy is reduced and costs are saved.

[0045] Based on the above embodiment, as an optional embodiment, determining the optimal sensor array layout based on the storage structure schematic diagram includes:

[0046] S301, establishing a three-dimensional solid model of a granary according to the storage structure schematic diagram, and constructing a gas diffusion model according to the three-dimensional solid model of the granary.

[0047] Specifically, first of all, it is necessary to restore the three-dimensional solid structure model of the granary based on the given storage structure diagram with the help of three-dimensional modeling software. The model needs to highly restore the size, shape, grain pile distribution, etc. of the storage area, and finely depict the structures and components that affect the airflow, such as the air inlet, air outlet, and mixer. After the three-dimensional solid model is established, it is imported into professional computational fluid dynamics (CFD) simulation software to construct a gas diffusion model. The essence of the gas diffusion model is to grid the internal space of the granary, and on this basis, to discretely solve the control equations (such as NavierStokes equations, energy equations, etc.) to simulate the diffusion law of the gas in it. When constructing the model, it is necessary to reasonably set the model boundary conditions and initial conditions, including the operating parameters of the ventilation system, the temperature and humidity of the grain pile, etc., so that the model is as close to the actual operating environment of the granary as possible. At the same time, it is also necessary to determine the gas physical properties of VOCs in the model according to the metabolic characteristics of molds. Through CFD calculation, the concentration distribution cloud map of VOCs gas inside the granary at different time and space positions can be obtained, which intuitively shows its flow field and diffusion characteristics. Combined with the concentration distribution, it is possible to determine which areas have a higher risk of mold occurrence, and sensors should be placed in these key areas first to improve detection sensitivity.

[0048] S302, in the gas diffusion model, virtually deploy sensor arrays of different categories, numbers and positions, and iterate the array layout of the sensor array through an optimization algorithm. When the number of iterations reaches a preset number, the optimal sensor array layout is obtained.

[0049] Specifically, this scheme proposes a virtual optimization strategy for sensor layout based on gas diffusion model. First, by adding sensor array module to the diffusion model, the response characteristics of sensors of different categories, quantities and positions to gas concentration can be simulated. By setting the spatial coordinates of sensors such as temperature, humidity and VOCs concentration, the mapping relationship between gas diffusion field and sensor measurement points is established. Several typical mildew scenarios are given in the model to simulate the release and diffusion process of VOCs gas and record the data sequence of each sensor node. Then, the optimization algorithm is introduced into the model, with the sensitivity and reliability of mildew warning as the optimization goal. Under the constraint of a certain total number of sensors, the optimal sensor combination layout is determined through iterative search. Generally, algorithms based on intelligent optimization are adopted, such as genetic algorithm or particle swarm algorithm. The algorithm randomly generates a series of possible layout schemes, calculates their performance indicators (such as coverage, repetition rate, missed detection rate, etc.), eliminates the layout combinations with poor performance, retains and optimizes the layout combinations with good performance, and iterates repeatedly until the preset convergence conditions (such as the number of iterations or the optimal value threshold, etc.) are met, and an approximate optimal solution can be obtained.

[0050] S103, arranging sensors based on the optimal sensor array layout, acquiring real-time response signal data of each sensor, and preprocessing the real-time response signal data to obtain a target real-time response signal.

[0051] Specifically, after obtaining the optimal sensor array layout through model optimization in the early stage, it is necessary to actually install sensors inside the granary according to the layout plan, and build a real-time data acquisition and transmission system to provide continuous monitoring data support for grain mildew warning. The arrangement of the sensor array is the physical basis for realizing online monitoring of grain quality, and the real-time response signal data generated by the sensor is the core basis for mildew warning analysis. According to the optimal sensor layout plan, monitoring nodes such as temperature and humidity sensors and gas sensors are installed in different areas and layers of the grain pile. Through the wireless communication module, each node forms a distributed sensor network to realize the automatic collection and remote transmission of monitoring data. The system sets a reasonable data sampling frequency, such as collecting data every 10-30 minutes, which can not only reflect the dynamic changes of environmental parameters, but also avoid data redundancy. The measurement range and accuracy of the sensor must meet the requirements of grain storage monitoring, such as the temperature measurement range of -20℃100%, the accuracy of ±3%RH, etc.

[0052] Based on the above embodiment, as an optional embodiment, the preprocessing of the real-time response signal data to obtain the target real-time response signal includes:

[0053] S401, performing baseline calibration on the real-time response signal data to obtain preliminary response signal data;

[0054] Specifically, after obtaining the real-time response signal data from the sensor, these raw data need to be preprocessed, first of all, baseline calibration is performed to obtain preliminary response signal data. This step is necessary because during the actual operation of the sensor, its response baseline will drift over time. If it is not calibrated in time, the accuracy of subsequent concentration calculations will be affected.

[0055] S402, filtering the preliminary response signal data to obtain a target real-time response signal.

[0056] Specifically, after completing the baseline calibration to obtain the preliminary response signal data, it is necessary to filter it to obtain the target real-time response signal. The function of filtering is to remove the noise interference in the response signal and improve the signal-to-noise ratio of the signal. This step is necessary because in the actual detection process, the sensor response signal is often affected by a variety of internal and external factors and generates various noises. Such as the thermal noise of the circuit itself, the interference noise of the surrounding electromagnetic field, the drift noise caused by the fluctuation of environmental parameters, etc. These noises will mask and reduce the quality of the real VOCs response signal. If the noisy response signal is directly processed without filtering, it is easy to cause a large deviation in the calculation of the VOCs concentration, affecting the sensitivity and accuracy of the detection. Therefore, it is necessary to filter and optimize the preliminary response signal to obtain ideal, noise-free target response signal data. There are many options for specific filtering processing methods, and the appropriate filtering algorithm can be selected according to the specific type and characteristics of the noise. Commonly used ones include wavelet transform denoising filtering, Kalman filtering, adaptive filtering, etc.

[0057] S104, extracting features of the target real-time response signal to obtain a feature vector, and inputting the feature vector into a preset recognition model to obtain a VOCs category corresponding to the target real-time response signal;

[0058] Specifically, the target real-time response signal obtained through the above steps has preliminarily removed noise and interference, and retained useful information reflecting the quality status of grain. However, due to the large number of sensor types and quantities and the high data dimension, the difference in response signals under different states is not intuitive, which is not convenient for the identification and early warning of moldy states. Therefore, it is necessary to further extract features from the target signal, highlight the differences in sensor signals under different moldy states, and establish a mapping relationship between features and moldy states. Common feature extraction methods include time domain statistical features, frequency domain features, time-frequency domain features, etc. For example, in the time domain, the mean, variance, peak and other statistical quantities of the signal are extracted to reflect the energy and fluctuation characteristics of the signal; in the frequency domain, the spectrum, cepstrum, power spectrum and other features of the signal are extracted to reflect the frequency distribution law of the signal; in the time-frequency domain, the wavelet coefficients and Hilbert transform of the signal are extracted to comprehensively characterize the time-frequency evolution behavior of the signal. In addition, signal decomposition and dimensionality reduction methods, such as empirical mode decomposition and principal component analysis, can also be used to further condense features and reduce information redundancy. Through feature extraction, high-dimensional sensor signals are converted into low-dimensional feature vectors. While the data size is greatly compressed, the correlation between the features and the mold state is significantly improved. Since the grain mold process is usually accompanied by dynamic changes in VOCs gas concentration, the characteristics of the VOCs response signal are of great significance for the classification of mold grades and early warning. The characteristic parameters of the VOCs signal are combined into a feature vector to facilitate quantitative analysis and pattern recognition.

[0059] Based on the above implementation, as an optional embodiment, the step of extracting features from the target real-time response signal to obtain a feature vector includes:

[0060] The target real-time response signal is divided into N data slices according to the continuous response time sequence, and a feature vector of the response value in each data slice is extracted.

[0061] Specifically, after obtaining the target real-time response signal, in order to improve the subsequent classification and recognition accuracy, it is necessary to extract features from the response data to obtain a feature vector that can well characterize the VOCs response pattern. The specific approach is to divide the target real-time response signal into N data slices according to the continuous response time series, and extract the vector describing its response characteristics from each data slice.

[0062] This slicing feature extraction method is adopted for the following two reasons:

[0063] First, the response process of VOCs gas to the sensor array is a dynamic time-varying process. Its response value shows a certain trend of change and transition characteristics in the time dimension, rather than a simple static value. Dividing the entire time series into multiple slices can well retain and reflect the dynamic information of this time-varying response pattern. Secondly, after the response signal is divided into multiple slices, the statistical response characteristics on each slice can be extracted to form a feature vector. Compared with the overall response vector, this sliced ​​feature vector can more finely describe the response details of VOCs in different time periods, thereby providing more criteria for classification and identification. The specific feature extraction steps are: first, according to the response time length and sampling frequency, the entire response time series is divided into N slices at equal intervals, and each slice contains M sampling points. Then, for each slice, the statistics of its M response values ​​are calculated as feature values, such as mean, standard deviation, peak value, energy, etc., to form an M-dimensional feature vector. In this way, a one-dimensional response sequence with a length of N*M is converted into N M-dimensional slice feature vectors. These feature vectors can reflect the response intensity, change rate, transition process and other detailed information of VOCs in different time periods. Through slicing feature extraction, not only the dynamic pattern of the time-varying response process is excavated, but also the feature dimension used for subsequent classification is greatly increased, enriching the judgment criteria, thereby significantly improving the recognition and resolution capabilities of complex VOCs mixed gases. This method is simple and efficient, and has been widely used in VOCs intelligent recognition and classification based on response time series.

[0064] S105, determining the VOCs category according to the VOCs category corresponding to the target real-time response signal, the feature vector and a preset VOCs fingerprint library, and determining the concentration corresponding to each VOCs category according to the VOCs category and the target real-time response signal.

[0065] Specifically, after the VOCs category corresponding to the target response signal was identified through the machine learning model, the mold stage and risk level of the sample gas were preliminarily determined. However, since each category may contain VOCs of various sources and chemical properties, with different toxicity and hazard levels, the category information alone is not enough to fully assess the mold degree and quality risk of the sample, and further identification of the specific category and concentration of VOCs is required.

[0066] To this end, the system pre-builds a VOCs fingerprint library based on chemical analysis and sensor array response. For dozens of VOCs commonly found in grain storage environments, the multi-sensor array response signals of pure samples are collected, characteristic parameters are extracted, and a mapping relationship between gas types and response patterns is established to form a complete fingerprint library. At the same time, through the analysis of VOCs release data of a large number of moldy grain samples, the corresponding rules between different types of VOCs and moldy states are mastered and stored in the fingerprint library. This library can be regarded as a standard "portrait" of various VOCs, and the characteristic patterns of each type of VOCs are formed through a large number of example learning.

[0067] After identifying the VOCs category to which the sample belongs, the system automatically retrieves the fingerprint features of various VOCs under the category, compares them one by one with the real-time feature vector of the sample, and calculates the feature similarity. Generally, the similarity between vectors is measured by metrics such as Euclidean distance and Mahalanobis distance. The smaller the distance, the higher the matching degree. Considering that the sample signal may be affected by environmental interference and noise pollution, the fingerprint features need to be adaptively scaled, translated, and transformed during the matching process to improve robustness. Finally, the VOCs corresponding to the fingerprint with the highest matching degree with the sample feature vector are judged as the specific category of the sample. After determining the specific category of VOCs, the concentration value of each VOCs needs to be further calculated. Since different types of sensors have different sensitivities and selectivities to VOCs, the multi-sensor data fusion method is used to improve the reliability of concentration estimation. First, the concentration range of VOCs is roughly estimated based on the comparison of the response signal intensity of the sample with the response intensity of the standard concentration sample in the fingerprint library. Then, several sensors that are most sensitive to this type of VOCs are selected, and the response intensity is converted into concentration value by querying the fitting curve. The estimated results of multiple sensors are weighted averaged to obtain the final concentration value of this type of VOCs.

[0068] The VOCs category and concentration information obtained through fingerprint library matching and data fusion measurement can provide a key basis for the accurate determination of the mildew state of grain. Different types of fungi release significantly different VOCs components during their growth process, and some VOCs are characteristic markers of grain deterioration. For example, octanol produced by Aspergillus ochraceus and nonanal produced by toxin-producing Penicillium are sensitive indicators of severe mildew of grain. Combined with the trend of VOCs concentration changes, the growth stage and reproduction rate of mold can also be determined, and early signals of accelerated mildew can be captured in a timely manner.

[0069] Based on the above embodiment, as an optional embodiment, according to the VOCs category corresponding to the target real-time response signal, the feature vector and a preset VOCs fingerprint library, the VOCs category is determined, and according to the VOCs category and the target real-time response signal, the concentration corresponding to each VOCs category is determined, including:

[0070] S501, determining the category characteristics of the corresponding category in the preset VOCs fingerprint library according to the VOCs category corresponding to the target real-time response signal;

[0071] Specifically, during the storage process, grain is easily affected by multiple factors such as temperature and humidity, grain conditions, and insect pests, and quality deterioration such as mildew and insect infestation occurs, producing a large amount of harmful VOCs gases. These VOCs not only directly contaminate grains, but also induce respiratory diseases and even lead to acute poisoning. Therefore, timely detection of characteristic VOCs in the granary environment is crucial to prevent mildew and control grain quality.

[0072] Traditional granary VOCs detection mainly relies on artificial olfaction and chemical analysis. However, human sense of smell is easily interfered by subjective factors, and its sensitivity and resolution are limited, making it difficult to detect early signs of mold. Conventional analysis methods such as gas chromatography require complex sample pretreatment and professional operation, with a long detection cycle, making it difficult to achieve real-time monitoring of granaries.

[0073] This solution uses electronic nose technology to achieve rapid qualitative and quantitative analysis of VOCs in granaries through pattern recognition methods. The basic idea is to introduce the mixed VOCs gas in the granary environment into the sensitive element array, collect its response signal and extract the characteristic vector, and then perform pattern matching with the pre-established standard VOCs fingerprint library to identify the types of main VOCs in the sample; further, according to the response-concentration calibration curve of each VOCs, its concentration in the sample can be converted. Since VOCs from different sources and causes vary greatly in chemical properties, the identification process adopts a classification approach. First, the corresponding subset of the fingerprint library is screened out according to the VOCs category of the sample, and then fine category matching is carried out in the sub-library, which can effectively reduce the amount of calculation and improve the efficiency and accuracy of identification.

[0074] In specific implementation, the system collects sufficient VOCs standard samples for typical scenarios such as mildew and insect infestation in granaries, and performs component identification and data labeling. The sample data is mined using principal component analysis, dimensionality reduction projection and other methods, and high-level semantic information is extracted based on its source, composition, response characteristics, etc., and the samples are divided into several VOCs categories, such as fungal metabolites, insect infestation markers, etc. The samples within each category have a high degree of similarity in chemical properties and sensor responses, while there are obvious differences between different categories.

[0075] For each VOCs category, the system selects representative standard samples so that their concentration levels and composition structures can cover the main variation space of the category. By passing the standard samples one by one into the sensitive element array and collecting their steady-state response signals, the "fingerprint" of the category can be obtained. The so-called fingerprint is actually a typical multidimensional response pattern, which is composed of characteristic response values ​​of multiple sensitive elements to the VOCs of this category. The fingerprints of each standard sample are summarized to obtain the category feature set of the category. The category features are stored in the form of a matrix, with each column corresponding to a sensitive material and each row corresponding to a standard sample. The number of rows reflects the sample capacity of the category, and the number of columns reflects the number of sensitive elements required to characterize the category.

[0076] In actual detection, when the sensitive element array responds to the VOCs mixed gas in the granary, the system collects the dynamic response curve of each sensitive element in real time, and extracts its steady-state value, response time and other characteristic parameters to form the real-time feature vector of the target sample. According to the correspondence between the VOCs category and the sensor response in the prior knowledge base, it can be preliminarily determined which VOCs category the sample belongs to, and the category features of the category in the fingerprint library can be retrieved as a template.

[0077] Next, the system searches for the standard fingerprint that is most similar to the target sample feature vector in the category feature set of the category. The specific pattern recognition algorithms used can be correlation coefficient matching, Euclidean distance calculation, support vector machine, etc. The core of the algorithm is to measure the similarity between two multidimensional vectors. The higher the score, the closer the two are in mathematical form. The recognition result is the VOCs category corresponding to the standard fingerprint that best matches the target response signal, reflecting the qualitative information of the main VOCs components in the sample.

[0078] In order to further understand the concentration level of each VOCs component, the system also needs to quantitatively analyze the category recognition results. The basis of quantitative analysis is the response-concentration calibration curve, which establishes a quantitative relationship between the sensor response value of a certain VOCs category and its concentration. The system has completed the construction of calibration curves for each category during the fingerprint library establishment stage. Once the VOCs category of the sample is obtained, the calibration curve of the corresponding category can be called up, substituted into the measured response value of the sample, and the estimated concentration of each VOCs component can be obtained through interpolation, extrapolation and other methods, thus achieving complete quantitative analysis of mixed VOCs gas.

[0079] S502, calculating the correlation coefficient between the feature vector and the category feature of the corresponding category in the preset VOCs fingerprint library, and obtaining the VOCs category based on the correlation coefficient;

[0080] Specifically, in the previous step, the system has preliminarily determined the VOCs category to which the granary VOCs sample belongs based on its response signal, and extracted the category feature set in the fingerprint library of the category. The category features are stored in the form of a multidimensional matrix, with each row corresponding to the fingerprint of a standard sample.

[0081] In order to find the fingerprint that is most similar to the target sample in the category feature set, a similarity measurement index needs to be defined. Commonly used measurement indexes include Euclidean distance, Mahalanobis distance, correlation coefficient, etc. Among them, the correlation coefficient can overcome the influence of dimension and scale differences, measure the consistency of the change trends of two vectors, and is an ideal choice for characterizing the similarity of fingerprints.

[0082] In the specific implementation, the feature vector of the sample is recorded as A, and each fingerprint spectrum of the category feature set is recorded as B1, B2,..., Bn. First, the Pearson correlation coefficient between A and each Bi is calculated, and the formula is:

[0083] r(A, Bi) = Cov(A, Bi) / [D(A)*D(Bi)]

[0084] Among them, Cov(A, Bi) is the covariance of A and Bi, D(A) and D(Bi) are the standard deviations of A and Bi respectively. The Pearson correlation coefficient actually measures the consistency of the changing trends of two vectors starting from their mean. When r is a positive value, it indicates that A and Bi change in the same direction in each dimension, and their response patterns are similar in morphology; when r is a negative value, it means that A and Bi have opposite changing trends and large differences in response patterns; when r is 0, it means that there is no linear correlation between A and Bi. The larger the absolute value of the correlation coefficient, the closer the morphology of the two fingerprints is.

[0085] Through pairwise calculation, the correlation coefficient between the sample feature vector A and each standard fingerprint Bi in the category feature set can be obtained, based on which the similarity between the sample and each category of VOCs can be judged. In order to avoid setting subjective thresholds, the system uses the maximum correlation coefficient method to determine the category of the sample. That is, the VOCs category corresponding to the standard fingerprint spectrum with the highest correlation with A (largest correlation coefficient) is taken as the identification result.

[0086] Assuming that the correlation coefficient between A and Bk is the largest, it can be determined that the main component in the sample belongs to the kth category of VOCs. This means that the typical components and characteristic odors of this category of VOCs appear in the atmosphere of the granary. The degree of pollution and the level of harm can refer to the standard samples and pollution events of this category. If this category of VOCs is a fungal metabolite, such as 8-carbon alcohol, 2-octanone, etc., it indicates that the grain has been moldy to a certain extent. The system can then issue an early warning signal to prompt the management personnel to check in time and take countermeasures such as ventilation and fumigation. The correlation coefficient matching method makes full use of the multidimensional morphological characteristics of the fingerprint spectrum, and reveals the intrinsic connection between the sample and the standard category through the coordinated changes in the response values ​​of each sensitive element. It avoids the comparison of the response values ​​of the fingerprint spectrum one by one, but focuses on the grasp of the overall trend. It has strong robustness and can adapt to the fluctuations of sensitive elements and measurement errors. Traditional pattern recognition methods such as the nearest neighbor algorithm pay too much attention to local differences and are easily affected by the failure of individual sensitive elements.

[0087] S503, determining the concentration corresponding to each VOCs category according to the preset response-concentration calibration curve corresponding to the VOCs category and the target real-time response signal.

[0088] Specifically, through the first two steps, the system has determined the VOCs category of the granary VOCs sample based on the real-time response signal, and obtained a qualitative identification result. However, only the type information is not enough to fully evaluate the pollution level of the granary VOCs, and further concentration data of each VOCs is required. In the construction stage of the fingerprint library, the system has carried out response-concentration calibration experiments for each category of VOCs. In the specific implementation, according to the detection limit and pollution threshold of different categories of VOCs, a series of standard sample gases with known concentrations are prepared, covering several orders of magnitude such as 0.1-1000ppm. The standard gas is passed into the sensor array in turn, and the steady-state response value of the sensitive element at each concentration level is collected to obtain a response curve similar to an "S" shape. Mathematical fitting of the response curve can establish a functional relationship between the sensor response value and the VOCs concentration. Commonly used fitting models include polynomial regression, power function regression, exponential regression, etc. The selection criteria are that the coefficient of determination R^2 is large enough and the model complexity is moderate. When R^2 is greater than 0.99, it means that the fitting function can explain more than 99% of the concentration-related variation in the sensor response value and can be used for subsequent concentration estimation. The model should be as simple as possible, with as few parameters as possible to avoid overfitting.

[0089] Taking polynomial regression as an example, its fitting function form is:

[0090] y=a0+a1x+a2x^2+...+an*x^n

[0091] Among them, y is the sensor response value, x is the VOCs concentration, and a0 to an are fitting coefficients. The polynomial order n is generally not more than 3 to avoid the model being too complicated. The function curve is approximately linear in the low concentration range and gradually tends to saturation in the medium and high concentration range.

[0092] Based on the above embodiment, as an optional embodiment, the concentration corresponding to each VOCs category is determined according to the preset response-concentration calibration curve corresponding to the VOCs category and the target real-time response signal, including:

[0093] Data extraction is performed on the target real-time response signal to obtain a response signal corresponding to the target real-time response signal, and the response signal is substituted into the response-concentration calibration curve corresponding to the preset VOCs category to obtain the concentration corresponding to each VOCs category.

[0094] Specifically, through fingerprint library matching, we have determined the specific category of VOCs corresponding to the target real-time response signal. However, to evaluate the degree of harm and pollution level of different types of VOCs, their concentration values ​​need to be further determined. Due to the different interaction mechanisms between VOCs of different types and concentrations and the sensor array, there are significant differences in the intensity and dynamic characteristics of the response signal. Therefore, it is necessary to establish a quantitative relationship between sensor response and VOCs concentration, that is, the response-concentration calibration curve. Before the system is activated, a series of standard gas samples with different concentration gradients are collected for the main VOCs commonly found in the grain depot environment, usually covering the ppb to ppm level. The standard gas samples are passed through the sensor array, the response signals of each sensor are collected, and the characteristic parameters such as the amplitude, integrated area, and dynamic response time of the response signal are extracted. After obtaining sufficient data points through repeated experiments, the concentration is used as the independent variable and the response signal characteristics are used as the dependent variable. The experimental data is fitted with a polynomial function or a piecewise function to obtain the response-concentration calibration curve.

[0095] For different types of VOCs, due to differences in their chemical structures and sensor sensitivity, the function form and parameter values ​​of the response-concentration curve are not the same. Therefore, it is necessary to establish calibration curves for each type of VOCs to form a complete curve library to prepare for subsequent concentration estimation. In the curve fitting process, the simplest function form should be selected as much as possible to avoid overfitting; at the same time, a high goodness of fit and a wide applicable concentration range should be guaranteed. The optimized calibration curve can minimize the nonlinearity and hysteresis effects of the sensor response and improve the accuracy of concentration measurement. When the VOCs category contained in the unknown grain depot environmental sample is identified, the system automatically retrieves the response-concentration calibration curve of the corresponding category. But before that, the real-time response signal of the sample needs to be preprocessed and data extracted. First, smoothing, outlier removal and other methods are used to remove high-frequency noise and accidental interference in the response signal to improve the signal-to-noise ratio. Then, according to the different VOCs categories, the characteristic parameters with the strongest correlation with their concentration are selected for extraction. Taking alcohol VOCs as an example, the response signal of C2-C8 alcohol to SnO2-based metal oxide sensor has good linearity, so the maximum amplitude of the response signal can be directly extracted as a feature; while the concentration of ester VOCs shows a good exponential relationship with the integral area of ​​the response signal, so it is appropriate to extract the definite integral value of the response curve as a feature. The purpose of data extraction is to selectively retain the signal components that are highly correlated with the measurement to be measured, remove the remaining information, and prepare for the input of the concentration inversion model.

[0096] Substitute the extracted response signal characteristics into the calibration curve equation of the corresponding VOCs category to calculate the real-time concentration value of this type of VOCs. For example, it is known that the calibration curve equation for ethanol is y=ax+b, where y is the response signal amplitude, x is the ethanol concentration, and a and b are fitting coefficients. Substitute the amplitude characteristic Ys of the sample signal into the equation, Xs=(Ys-b) / a, and you can get the real-time concentration estimate of ethanol. Similarly, for all VOCs categories detected in the sample, their concentration levels are estimated by the above method.

[0097] See also Figure 2 , Figure 2 The present invention provides an architecture diagram of a grain mildew early warning system for a granary, which may include:

[0098] The data acquisition module 1 is used to obtain the type of grain stored in the granary and the grain storage environment data, and determine the typical VOCs markers released by mold when the grain is moldy according to the type of stored grain and the grain storage environment data;

[0099] The layout optimization module 2 is used to obtain a storage structure diagram inside the granary and determine an optimal sensor array layout based on the storage structure diagram;

[0100] A preprocessing module 3 is used to arrange sensors based on the optimal sensor array layout, obtain real-time response signal data of each sensor, and preprocess the real-time response signal data to obtain a target real-time response signal;

[0101] Classification module 4, used for extracting features of the target real-time response signal to obtain a feature vector, and inputting the feature vector into a preset recognition model to obtain the VOCs category corresponding to the target real-time response signal;

[0102] A matching module 5 is used to determine the VOCs category according to the VOCs category corresponding to the target real-time response signal, the feature vector and a preset VOCs fingerprint library, and determine the concentration corresponding to each VOCs category according to the VOCs category and the target real-time response signal;

[0103] The early warning module 6 is used to input the concentration corresponding to each of the VOCs categories into the preset VOCs concentration estimation model to obtain the absolute concentration corresponding to each VOCs category, and determine the mildew warning level of grain in the granary based on the absolute concentration corresponding to each of the VOCs categories and the preset concentration threshold.

[0104] Please refer to Figure 3 The application also discloses an electronic device. Figure 3 The electronic device 300 may include: at least one processor 301 , at least one network interface 304 , a user interface 303 , a memory 305 , and at least one communication bus 302 .

[0105] The communication bus 302 is used to realize the connection and communication between these components.

[0106] The user interface 303 may include a display screen (Display) and a camera (Camera). Optionally, the user interface 303 may also include a standard wired interface and a wireless interface.

[0107] The network interface 304 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).

[0108] Among them, the processor 301 may include one or more processing cores. The processor 301 uses various interfaces and lines to connect various parts in the entire server, and executes various functions of the server and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 305, and calling data stored in the memory 305. Optionally, the processor 301 can be implemented in at least one hardware form of digital signal processing (Digital Signal Processing, DSP), field programmable gate array (Field~Programmable Gate Array, FPGA), and programmable logic array (Programmable Logic Array, PLA). The processor 301 can integrate one or a combination of a central processing unit (Central Processing Unit, CPU), a graphics processing unit (Graphics Processing Unit, GPU) and a modem. Among them, the CPU mainly processes the operating system, user interface and application programs; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 301, and it can be implemented separately through a chip.

[0109] The memory 305 may include a random access memory (RAM) or a read-only memory (Read~Only Memory). The memory 305 includes a non-transitory computer-readable medium. The memory 305 may be used to store instructions, programs, codes, code sets or instruction sets. The memory 305 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data involved in the above-mentioned various method embodiments, etc. The memory 305 may optionally be at least one storage system located away from the aforementioned processor 301. Refer to Figure 3 The memory 305 as a computer storage medium may include an operating system, a network communication module, a user interface module, and an application program of a method for early warning of grain mildew in a granary.

[0110] exist Figure 3In the electronic device 300 shown, the user interface 303 is mainly used to provide an input interface for the user and obtain the data input by the user; and the processor 301 can be used to call the application program storing the road assessment method in the memory 305. When executed by one or more processors 301, the electronic device 300 executes one or more methods in the above-mentioned embodiments. It should be noted that for the aforementioned method embodiments, for the sake of simple description, they are all expressed as a series of action combinations, but those skilled in the art should know that the present application is not limited to the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required for the present application. In the above embodiments, the description of each embodiment has its own emphasis. For the parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0111] In the several implementations provided in this application, it should be understood that the disclosed system can be implemented in other ways. For example, the system embodiments described above are only schematic, such as the division of units, which is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some service interfaces, and the indirect coupling or communication connection of the system or unit can be electrical or other forms.

[0112] The present application also provides a computer storage medium that can store multiple instructions, which are suitable for being loaded and executed by a processor as described above. Figure 1 The road assessment method of the embodiment shown in the figure can be found in the specific implementation process. Figure 1 The specific description of the illustrated embodiment will not be repeated here.

[0113] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0114] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a memory and includes N instructions for a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned memory includes: various media that can store program codes, such as USB flash drives, mobile hard drives, magnetic disks or optical disks.

[0115] The above are only exemplary embodiments of the present disclosure and cannot be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. After considering the disclosure of the specification and the truth of practice, those skilled in the art will easily think of other embodiments of the present disclosure.

[0116] This application is intended to cover any variation, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary technical means in the art not described in the present disclosure. The description and examples are to be regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A method for early warning of grain mildew in a granary, characterized in that: The method comprises: The category of grain stored in the granary and the grain storage environment data are obtained, and according to the category of the stored grain and the grain storage environment data, typical VOCs markers released by mold when the grain is moldy are determined; the method of determining the typical VOCs markers released by mold when the grain is moldy according to the category of the stored grain and the grain storage environment data comprises: searching literature for the category of the stored grain and the grain storage environment data to determine the main mold category when the grain is moldy; searching a preset mold category VOCs fingerprint database according to the main mold category when the grain is moldy, and determining the typical VOCs markers released by the mold; A storage structure diagram inside a granary is obtained, and an optimal sensor array layout is determined based on the storage structure diagram; the determining of the optimal sensor array layout based on the storage structure diagram includes: establishing a three-dimensional solid model of the granary according to the storage structure diagram, and constructing a gas diffusion model according to the three-dimensional solid model of the granary; in the gas diffusion model, sensor arrays of different categories, quantities and positions are virtually deployed, and the array layout of the sensor array is optimized and iterated through an optimization algorithm, and when the number of iterations reaches a preset number, the optimal sensor array layout is obtained; Arranging sensors based on the optimal sensor array layout, acquiring real-time response signal data of each sensor, and preprocessing the real-time response signal data to obtain a target real-time response signal; Extracting features of the target real-time response signal to obtain a feature vector, and inputting the feature vector into a preset recognition model to obtain the VOCs category corresponding to the target real-time response signal; Determine the VOCs category according to the VOCs category corresponding to the target real-time response signal, the feature vector and a preset VOCs fingerprint library, and determine the concentration corresponding to each VOCs category according to the VOCs category and the target real-time response signal; The concentration corresponding to each of the VOCs categories is input into a preset VOCs concentration estimation model to obtain the absolute concentration corresponding to each VOCs category, and the mildew warning level of grain in the granary is determined based on the absolute concentration corresponding to each of the VOCs categories and the preset concentration threshold.

2. The method according to claim 1, characterized in that The preprocessing of the real-time response signal data to obtain a target real-time response signal includes: Performing baseline calibration on the real-time response signal data to obtain preliminary response signal data; The preliminary response signal data is filtered to obtain a target real-time response signal.

3. The method according to claim 1, characterized in that: The step of extracting features from the target real-time response signal to obtain a feature vector includes: The target real-time response signal is divided into N data slices according to a continuous response time sequence, and a feature vector of a response value in each of the data slices is extracted.

4. The method according to claim 1, characterized in that: The VOCs category is determined according to the VOCs category corresponding to the target real-time response signal, the feature vector and a preset VOCs fingerprint library, and the VOCs category and the VOCs fingerprint library are determined according to the VOCs category and the The target real-time response signal determines the concentration corresponding to each VOCs category, including: Determine the category characteristics of the corresponding category in the preset VOCs fingerprint library according to the VOCs category corresponding to the target real-time response signal; Calculating the correlation coefficient between the feature vector and the category feature of the corresponding category in the preset VOCs fingerprint library, and obtaining the VOCs category based on the correlation coefficient; The concentration corresponding to each VOCs category is determined according to the preset VOCs category corresponding response concentration calibration curve and the target real-time response signal.

5. The method according to claim 4, characterized in that The step of determining the concentration corresponding to each VOCs category according to the preset VOCs category corresponding response concentration calibration curve and the target real-time response signal includes: Data extraction is performed on the target real-time response signal to obtain a response signal corresponding to the target real-time response signal, and the response signal is substituted into the response concentration calibration curve corresponding to the preset VOCs category to obtain the concentration corresponding to each VOCs category.

6. A grain mildew early warning system in a granary, characterized in that: The system comprises: The data acquisition module is used to obtain the category of grain stored in the granary and the grain storage environment data, and determine the main mold category and the typical VOCs markers released by the mold when the grain is moldy according to the category of the stored grain and the grain storage environment data; the typical VOCs markers released by the mold when the grain is moldy according to the category of the stored grain and the grain storage environment data include: querying the literature to determine the main mold category when the grain is moldy according to the category of the stored grain and the grain storage environment data; querying the preset mold category VOCs fingerprint database according to the main mold category when the grain is moldy, and determining the typical VOCs markers released by the mold; A layout optimization module is used to obtain a storage structure schematic diagram inside a granary, and determine an optimal sensor array layout based on the storage structure schematic diagram; the determination of the optimal sensor array layout based on the storage structure schematic diagram includes: establishing a three-dimensional solid model of the granary according to the storage structure schematic diagram, and constructing a gas diffusion model according to the three-dimensional solid model of the granary; in the gas diffusion model, virtually deploying sensor arrays of different categories, quantities and positions, optimizing and iterating the array layout of the sensor array through an optimization algorithm, and obtaining the optimal sensor array layout when the number of iterations reaches a preset number; A preprocessing module, used to arrange sensors based on the optimal sensor array layout, obtain real-time response signal data of each sensor, and preprocess the real-time response signal data to obtain a target real-time response signal; A classification module, for extracting features of the target real-time response signal to obtain a feature vector, and inputting the feature vector into a preset recognition model to obtain the VOCs category corresponding to the target real-time response signal; A matching module, for determining the VOCs category according to the VOCs category corresponding to the target real-time response signal, the feature vector and a preset VOCs fingerprint library, and determining the concentration corresponding to each VOCs category according to the VOCs category and the target real-time response signal; The early warning module is used to input the concentration corresponding to each of the VOCs categories into a preset VOCs concentration estimation model to obtain the absolute concentration corresponding to each VOCs category, and determine the mildew early warning level of grain in the granary based on the absolute concentration corresponding to each of the VOCs categories and a preset concentration threshold.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor and executing the method according to any one of claims 1 to 5.

8. An electronic device, characterized in that: It includes a processor, a memory and a transceiver, the memory is used to store instructions, the transceiver is used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the method according to any one of claims 1 to 5.