Fresh corn circulation mildew-proof and quality-guaranteeing method based on intelligent monitoring

Abstract

The invention discloses a fresh corn circulation mildew-proof and quality-guaranteeing method based on intelligent monitoring, and particularly relates to the technical field of agricultural product circulation management.The method comprises the steps that quality data are collected through image, gas, humidity, temperature and microorganism multi-source sensing means, and a quality degradation trend curve and a risk grade score are generated in combination with a time sequence modeling algorithm; identification parameters are automatically adjusted in a high-humidity or oscillation environment, so that early-stage mildew identification accuracy is improved; a high-risk batch is identified through multi-dimensional cross analysis, and a circulation suspension response is triggered; the whole-process data is uploaded to a management platform, so that a traceable quality record is formed, and whole-process dynamic monitoring and closed-loop control are realized; according to the method, recessive mildew is recognized through multi-modal perception, the shelf life is predicted and evaluated through trend, the recognition stability is dynamically adjusted and improved, a chain type data tracing mechanism is constructed to achieve model self-evolution, the quality early warning, risk control and continuous optimization capacity in the whole fresh corn circulation process is improved, and it is ensured that the quality guaranteeing and mildew preventing effects are accurate and reliable.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural product circulation management. More specifically, the present invention relates to a method for preventing mildew and ensuring the quality of fresh corn circulation based on intelligent monitoring. Background Art

[0002] Fresh corn is a type of high-moisture and high-activity agricultural product with both fresh food attributes and nutritional value. During the transportation, storage, and distribution processes from harvesting to terminal sales, it is extremely prone to quality degradation and mildew problems. Especially in the high-temperature and high-humidity environment in summer, it is extremely easy to induce the rapid reproduction of molds such as Fusarium graminearum, resulting in phenomena such as corruption, peculiar smell, and color fading, seriously affecting the commodity quality and food safety.

[0003] The existing technologies for preventing mildew and ensuring the quality of fresh corn circulation mainly rely on a combination of cold chain transportation, conventional antibacterial treatments (such as ozone, ultraviolet, packaging modification, etc.) and manual sampling to ensure quality. However, there are generally the following technical bottlenecks: First, the detection means are single, mainly relying on manual observation and visible appearance features, making it difficult to identify the risk of early "hidden mildew"; second, there is a lack of the ability to model and predict the trend of product quality changes, often only being able to respond passively after problems occur and unable to intervene in advance; third, the quality judgment depends on manual experience, is greatly affected by subjective factors, and the response is not timely; fourth, the monitoring links are scattered, not forming a complete closed loop, and lacking a data-driven dynamic quality management mechanism.

[0004] Especially in the context of the rapid development of intelligent agricultural product circulation and digital supply chains, the traditional methods can no longer meet the intelligent requirements of high-frequency turnover, stable quality, and controllable quality preservation of fresh corn. Therefore, the present invention proposes a method for preventing mildew and ensuring the quality of fresh corn circulation based on intelligent monitoring to meet the needs of closed-loop management of product quality during the circulation process. Summary of the Invention

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for preventing mildew and ensuring the quality of fresh corn circulation based on intelligent monitoring, comprising the following steps:

[0007] By configuring an image acquisition module, a gas detection module, a humidity and temperature sensing module, and a rapid microbial detection device, multi-source quality data of fresh corn are synchronously collected in different circulation links, including color, epidermal texture, gas volatile concentration, temperature and humidity changes, and pathogen metabolic indicators;

[0008] Using a quality trend prediction algorithm based on time series modeling, the above multi-dimensional data are serially trained to construct a quality degradation trend curve, automatically judge the degree of quality deterioration and the future shelf life cycle, and assign a risk level score;

[0009] When high humidity fluctuations or shock events occur in the storage and transportation environment, the color recognition sensitivity, gas warning threshold, and microbial detection frequency are adjusted in real time to achieve the adaptive reconstruction of the parameter recognition model, so as to improve the recognition accuracy in the initial stage of mildew;

[0010] Perform multi-dimensional cross-analysis on the multi-modal data of each batch of fresh corn. When the risk level predicted by the recognition model exceeds the standard warning value, a circulation suspension processing signal is triggered, and potentially hidden mildew batches are preferentially removed to prevent products with normal appearance but internal deterioration from entering the sales process;

[0011] Upload the data collected at each stage, prediction results, and processing records to the centralized management platform through the data interface, and automatically archive them to form a quality traceability chain.

[0012] In a preferred embodiment, the image acquisition module includes a high-resolution image sensor and a near-infrared reflection unit. The reflected images in the preset wavelength bands are used to extract the weak color difference on the surface of corn kernels and the change signals in the local wet spot area. When there is no obvious abnormality detected by normal visible light, potential mildew signs are determined through spectral enhanced feature recognition; the image data undergoes color space conversion and region segmentation operations in the preprocessing stage, converting the image from the red-green-blue color space to the lightness-saturation-hue space, and performing dynamic difference calculation by combining the edge gradient change and the mean value of color difference distribution, so as to improve the recognition probability of early mildew micro-clusters.

[0013] In a preferred embodiment, the gas detection module is constructed based on the principle of an electronic nose, and volatile organic compounds in the corn storage environment are identified through multi-channel metal oxide gas sensitive elements. The identification includes aldehydes, ketones, and low molecular weight fatty acid components in the metabolites of molds. By constructing characteristic response vectors under different odor intensity combinations, odor pattern recognition is achieved; at the same time, the "sensitivity coefficient of deteriorated odor deviation" is set. When the value of this coefficient is higher than the warning threshold of the gas perception sensitive response, the weight adjustment of the mildew process in the prediction model is triggered, and a high-risk tendency is given to the recent data segment to achieve the early locking of the mildew path. The "sensitivity coefficient of deteriorated odor deviation" is calculated based on any one of the three discrimination methods of Mahalanobis distance, Manhattan distance, and Shannon entropy.

[0014] In a preferred embodiment, the rapid microbial detection device adopts a non-destructive detection mechanism based on impedance method or bioluminescence method to capture the metabolic activity indexes of living colonies existing on the surface of corn kernels or inside the packaging. Before the detection data enters the prediction model, high-frequency sample fluctuation difference fitting is first performed. When the difference value exceeds the microbial metabolism trend deviation threshold, it enters the critical sample backtracking segment for historical curve comparison. This comparison adopts one of the two strategies of moving average fitting and perturbation trend regression to determine whether it is a significant deviation event. When the data within three consecutive cycles are all in the deviation interval, the prediction model raises the warning level for this batch of data, directly adjusting it from a low-risk level to a medium-high risk interval.

[0015] In a preferred embodiment, the quality trend prediction algorithm adopts a multi-level time series modeling strategy, combines short-term trend window and periodic trend window for dual evolution analysis, and introduces a causal weight optimization mechanism in the core part of the model. This mechanism adaptively and dynamically adjusts the data weight ratio participating in modeling based on the causal relationship strength of different data channels.

[0016] In a preferred embodiment, when the prediction result generates the shelf life cycle, the risk level is divided into five levels, namely stable state, fluctuation warning, initial risk, medium risk, and high-risk spoilage state. Each level corresponds to different warning responses and disposal strategies.

[0017] In a preferred embodiment, the adaptive reconstruction process of the parameter identification model includes three parts: a sensitivity dynamic adjustment module, an identification error feedback module, and an environmental response weight mapping mechanism. After detecting the trigger of a high-humidity environment, it automatically reduces the "subtle gray-scale speckle variation response threshold" in the color identification model and simultaneously increases the discrimination factor for odor intensity change. When the environmental fluctuation does not return to the normal state within three consecutive cycles, the identification error feedback module is activated to adjust the current model prediction boundary based on the true judgment result of the previous cycle.

[0018] In a preferred embodiment, data cross-analysis adopts a mechanism based on vector similarity matching and time series difference fusion. Feature vectors are constructed for data channels such as images, gases, temperature and humidity respectively, and an omni-directional data cross-section map for each batch of fresh corn within a time period is formed. If it is detected that the similarity index between the maps decreases and the predicted risk score approaches the quality anomaly trigger threshold, this batch will be automatically marked as a "key monitoring object". Such batches will trigger a secondary quality inspection mechanism and data refresh training before entering the warehousing or transportation process to strengthen the model's accurate identification ability for boundary samples.

[0019] In a preferred embodiment, the centralized management platform graphically displays the post - harvest processing, packaging forms, transportation routes, and quality change trends of different batches of fresh - eating corn based on a visual chain structure. Each time a prediction cycle is processed, the current input data status and output prediction results are recorded and matched with the actual quality feedback. If the prediction deviation amplitude is greater than the prediction accuracy stability threshold for two consecutive cycles, the platform automatically marks that the current model needs to be retrained and selects the latest batch of stable samples as the training basis.

[0020] Technical effects and advantages of the present invention:

[0021] By introducing multi - modal perception means, the present invention significantly improves the recognition ability of the "hidden mildew" state of fresh - eating corn, overcoming the technical bottleneck that traditional monitoring methods rely on appearance features and are prone to missing early decay judgments. Image recognition combined with near - infrared band enhancement, gas - sensing recognition of mold metabolites, and non - destructive discrimination of active signals by rapid microbial detection enables the present invention to capture early quality risks through internal parameter changes when there are no obvious external abnormalities, realizing pre - warning and ensuring food circulation safety.

[0022] By constructing a quality trend prediction model based on time series and combining a risk - level scoring and dynamic parameter adjustment mechanism, the present invention realizes an intelligent upgrade from "current judgment" to "trend prediction". The model can accurately estimate the shelf life according to the change trends of data from different channels, and dynamically adjust the recognition threshold and weight factors under abnormal environments such as high humidity and vibration, ensuring stable and accurate recognition ability under complex transportation conditions, thereby improving the reliability and adaptability of actual deployment.

[0023] The present invention realizes full - process closed - loop management of data during operation. The centralized platform forms a chain record of each batch of data, prediction results, and actual feedback, which is not only used for real - time display and warning but also supports model retraining and self - iterative optimization. When the recognition accuracy shows a downward trend, the platform automatically screens stable samples for model update to ensure that the prediction accuracy rate is still maintained under long - term use, and then forms an intelligent self - evolution ability of "perception - modeling - feedback - evolution", enabling the entire anti - mildew and quality - preservation strategy to have the technical advantage of continuous improvement. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] For the convenience of those skilled in the art to understand, the present invention will be further described below with reference to the accompanying drawings;

[0025] Figure 1 It is a schematic diagram of the method for preventing mildew and preserving the quality of fresh - eating corn in circulation based on intelligent monitoring in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0027] Referring to Figure 1 the following embodiments are obtained:

[0028] Embodiment 1: A method for preventing mildew and maintaining the quality of fresh corn during circulation based on intelligent monitoring, comprising the following steps:

[0029] By configuring an image acquisition module, a gas detection module, a humidity and temperature sensing module, and a rapid microbial detection device, multi-source quality data of fresh corn are synchronously collected in different circulation links, including color, epidermal texture, gas volatile concentration, temperature and humidity changes, and pathogen metabolic indicators; this step belongs to the perception stage of the present invention, and the purpose is to establish a perception network with multi-type acquisition capabilities to comprehensively obtain the quality status of fresh corn in each link during the whole process from post-harvest to sales. Specifically, it includes:

[0030] The image acquisition module: used to obtain information on the color, gloss, and texture changes of the corn epidermis;

[0031] The gas detection module: used to monitor the odoriferous volatile compounds produced by microbial metabolism in the storage environment;

[0032] The humidity and temperature sensing module: used to monitor the temperature and humidity fluctuations during the circulation process;

[0033] The rapid microbial detection device: used to identify the activity signals of potential pathogenic bacteria on the surface of the corn or inside the packaging; through the above configuration, a multi-dimensional data channel covering visual features, gas features, environmental parameters, and biological indicators can be established to provide data support for subsequent intelligent analysis.

[0034] Using a quality trend prediction algorithm based on time series modeling, the above multi-dimensional data is subjected to sequence training to construct a quality degradation trend curve, automatically judge the degree of quality deterioration and the future shelf life cycle, and assign a risk level score; this step is the intelligent judgment stage of the present invention, and the core lies in using the evolutionary relationship between historical data and real-time data to simulate and predict the quality change trend of fresh corn in the future for a period of time. The implementation methods include:

[0035] Train the data of multiple channels such as images, gases, temperature and humidity, and microorganisms in a time series manner; use a trend modeling algorithm to generate a quality degradation curve to reflect the evolution path of corn from the current state to the deteriorated state; deduce the remaining edible period by judging the current stage of the curve; at the same time, assign a five-level risk rating to the current state of the sample, corresponding to different quality states from stable to high-risk, so as to guide subsequent circulation management. The goal of this link is "early identification", rather than after-the-fact processing.

[0036] When high humidity fluctuations or shock events occur in the storage and transportation environment, adjust the color recognition sensitivity, gas warning threshold, and microorganism detection frequency in real time to achieve adaptive reconstruction of the parameter recognition model, so as to improve the recognition accuracy in the initial stage of mildew; the problem solved by this step is the problem that the recognition accuracy of traditional recognition decreases under abnormal environments (such as sudden increase in humidity, transportation vibration). For this reason, the present invention proposes a parameter dynamic adjustment mechanism, including:

[0037] When there is a sharp rise in humidity or continuous vibration during storage or transportation, monitor the environmental changes; on this basis, the model automatically reduces the sensitivity threshold for judging color spot changes in color recognition; at the same time, improve the response sensitivity to gas concentration changes to detect signs of active microorganisms caused by heat and humidity earlier; synchronously increase the sampling frequency of microorganism detection to increase the chance of capturing abnormal states. This step ensures that the model can still operate stably in a non-ideal environment and improves the recognition ability of early mildew characteristics.

[0038] Conduct multi-dimensional cross-analysis on the multi-modal data of each batch of fresh corn. When the risk level predicted by the recognition model exceeds the standard warning value, a circulation suspension processing signal is triggered, and potentially hidden mildew batches are preferentially eliminated to prevent products with normal appearance but internal deterioration from entering the sales process; this step is the intelligent response control stage and is a key link for the whole set of methods to provide intervention actions to the front end of the industrial chain. Its operation logic is as follows:

[0039] When each batch of corn is detected, multiple data sources are combined for cross-verification to improve the judgment accuracy; if the risk level is rated as medium to high level and exceeds the preset warning standard value; a "processing prompt" or "blocking signal" will be actively triggered, and this batch of products will be marked as "potentially unqualified"; the management end will suspend the further circulation of this batch in the logistics and recommend manual re-inspection or cold chain interruption; at the same time, this batch will be removed from the preset sales logic path to ensure that it does not enter the terminal market. Here, an actual "risk blocking mechanism" is formed, which is different from the traditional static method that relies on human experience judgment.

[0040] Upload the data collected in each stage, prediction results, and processing records to the centralized management platform through the data interface, and automatically archive them to form a quality traceability chain. This step is a closed-loop management link, aiming to establish a quality traceability record for the whole process of fresh corn from harvest to sales through continuous archiving and data accumulation. The key points are as follows:

[0041] The data generated in each link (such as sensing values, prediction results, risk scores, processing suggestions, etc.) are all recorded; the data forms a unified storage structure through the docking platform; the platform visually displays the circulation path and quality evolution trajectory of each batch of products; at the same time, it supports functions such as subsequent model optimization, quality liability backtracking, and intelligent assignment of product labels. Through this mechanism, the digitalization, standardization, and transparency of quality control are truly realized.

[0042] The image acquisition module includes a high-resolution image sensor and a near-infrared reflection unit. The reflected images in the preset bands are used to extract the weak color difference on the surface of corn kernels and the change signals in the local wet spot area. When there is no obvious abnormality detected by normal visible light, potential mildew signs are determined through spectral enhanced feature recognition; the image data undergoes color space conversion and region segmentation operations in the preprocessing stage, converting the image from the red-green-blue color space to the lightness-saturation-hue space, and performing dynamic difference calculation by combining edge gradient changes and the mean value of color difference distribution, thereby increasing the recognition probability of early mildew micro-clusters.

[0043] In this link, the high-resolution image sensor can select an industrial camera device with more than 12 million pixels, which can clearly image the details of the corn surface under different light source conditions. The near-infrared reflection unit, through the integrated band control module, enables image acquisition not only within the visible light range of the human eye, but also covers specific bands including the near-infrared band. The commonly used reflection center band range can be set between 720 and 900 nanometers. This reflection range can effectively improve the imaging ability of low-contrast structures such as water stains and initial appearance of hyphae.

[0044] The image data enters the preprocessing stage after being acquired. First, perform the color space conversion operation, converting the original image from the traditional red-green-blue color space to the lightness-saturation-hue space. This conversion can decouple the intensity and brightness of colors, facilitating subsequent quantitative analysis of "tiny color differences". This processing step can be achieved through the color space mapping function in open-source image processing tools, such as the common color space transformation algorithms in image processing libraries, or can also be replaced by a custom color mapping method with the same effect.

[0045] After the color space conversion is completed, the image enters the region segmentation processing stage, which is used to extract the grain unit region and the suspicious mildew region. For region segmentation, traditional algorithms based on edge detection, such as the Sobel operator, Canny operator, etc., can be used to extract the edge gradient change of the image; existing deep learning models, such as semantic segmentation networks based on convolutional structures, can also be applied to achieve intelligent segmentation of the grain contour. The present invention does not limit the algorithm structure for segmentation processing, and any technical method that can accurately extract the grain region can be used as an alternative.

[0046] In the segmented image region, the mean information of the edge gradient change and the color difference distribution is further calculated, and the pre-mildew characteristics that may exist in the image are extracted through a dynamic difference calculation method. The dynamic difference calculation can be realized by means of pixel block difference accumulation and local statistical variance comparison, or a sliding window filter based on local gray value change can be introduced for comparative analysis. By comparing the small differences in color, brightness, and texture in different regions, those image regions that have not yet formed visible mildew spots but have potential decay characteristics can be identified.

[0047] It should be noted that the feature recognition process used in the image acquisition module can correspond to the visual data channel in the quality trend prediction algorithm described in the present invention, that is, the processed image feature values are used as one of the inputs and transmitted to the model structure to participate in the subsequent shelf life prediction and risk level scoring.

[0048] In addition, the present invention does not limit which specific image processing algorithm or classification model is used, and any technical path that can achieve the function of enhancing and extracting early mildew image features can be used as an alternative, including but not limited to deep neural networks, support vector machines, fuzzy logic image classifiers, etc.

[0049] For further illustration, the following embodiments can be used as a reference in the present invention: In the acquisition stage, a red-green-blue and near-infrared dual-channel camera is configured to collect images of the corn epidermis. When there is no abnormality in the visible light band image, the change in the reflection distribution in the near-infrared image is used to compare the image difference map of the same batch, and it is found that there are regions with abnormally enhanced reflection intensity on the surface of some grains. After processing, it is determined as a region with abnormally increased local humidity. After combining with temperature and humidity data for confirmation, it can be marked as a potential mildew warning region, and finally this image region enters the model structure as a weight feature for recognition and judgment.

[0050] The gas detection module is constructed based on the principle of electronic nose, and uses multi-channel metal oxide gas sensitive elements to identify volatile organic compounds in the corn storage environment. The identification includes aldehydes, ketones and low molecular weight fatty acid components in mold metabolites. By constructing characteristic response vectors under different odor intensity combinations, odor pattern recognition is achieved. At the same time, the "sensitivity coefficient of deteriorated odor deviation" is set. When the coefficient value is higher than the warning threshold of gas perception sensitive response, the weight adjustment of the mildew process of the prediction model is triggered, and a high-risk tendency is given to the recent data segment to realize the early locking of the mildew path. The "sensitivity coefficient of deteriorated odor deviation" is calculated based on any one of the three discriminant methods: Mahalanobis distance, Manhattan distance, and Shannon entropy.

[0051] In terms of hardware composition, this module includes multiple independent gas sensitive elements. Each element is based on a metal oxide semiconductor structure. Through an oxidation-reduction reaction at a certain working temperature, a resistance change is generated to reflect the fluctuation of a specific gas concentration. Typical sensing elements can respond to mold metabolism-related components such as n-butyl aldehyde, hexanone, and acetic acid. Their response time is less than two seconds, and the recovery time is less than five seconds, which is suitable for dynamic real-time collection during the circulation process.

[0052] The original output signal of each sensing unit can be normalized and then constructed into a set of characteristic response vectors. This vector contains multiple dimensions, and each dimension corresponds to the output amplitude of a gas sensitive element. Through the combined structure of the vectors, a unique odor pattern map can be formed. For simplicity of implementation, in the specific implementation of the present invention, the first n principal components can be extracted by principal component analysis to construct a reduced-dimensional feature vector, but it is not limited to this method. Any algorithm that can meet the requirements of feature extraction and recognition accuracy, including but not limited to linear discriminant analysis, autoencoder compression network, and fuzzy clustering method, can be used for substitution.

[0053] The above-mentioned "sensitivity coefficient of deteriorated odor deviation" refers to the degree of similarity deviation of the gas response vector between multiple time periods, and is used to measure the drastic change of the sample's abnormality at the odor level. The present invention provides three implementable but not exclusive calculation methods:

[0054] Mahalanobis distance calculation method: It can be used to measure the multi-dimensional statistical difference between the current gas response vector and the historical standard sample. This method considers the covariance relationship of each dimension feature and is effective for non-independent features.

[0055] Manhattan distance calculation method: The overall change of the odor is measured by the sum of the absolute value differences of each dimension. The calculation complexity is low, and it is suitable for edge processing scenarios.

[0056] Shannon entropy difference comparison: By statistically analyzing the distribution probability of each gas component, calculating the total amount of its information entropy, and comparing the entropy difference between the current sample and the reference sample, it is used to detect the increase in pattern uncertainty.

[0057] The present invention does not limit that a certain type of method must be used to calculate the odor offset sensitivity coefficient. As long as the method used can objectively reflect the odor response difference and has stability and repeatability, it can be regarded as an equivalent technical means.

[0058] When the odor offset sensitivity coefficient exceeds the set early warning threshold for gas perception sensitive response, a subsequent processing mechanism is triggered. This threshold can be set by statistical derivation from historical mildew sample data, or can be optimized according to the false positive rate in the actual deployment environment. The specific value can be set as a risk threshold according to the percentile of the odor response offset rate in the sample set (for example, the 85th percentile).

[0059] When the above-mentioned odor offset value is determined to be in a high-risk area, the mildew process weight factor in the quality trend prediction model mentioned in the present invention will be dynamically adjusted, so that the model automatically increases the risk discrimination weight of the odor channel in the recent data analysis. This weight adjustment mechanism can be reflected in the input feature weighting layer of the model. For example, a multiplicative weight amplification factor is given to the output result of the gas channel in the data block of the past two hours, so that it has a higher decision-making influence in the internal calculation of the model.

[0060] The function of this mechanism is that even if there are no obvious abnormalities in the current image features or microbial indicators, based on the "early change of olfactory risk" signal, an "early tendency judgment" of the future trend can be realized, the mildew risk path can be locked in time, and the relevant batches can be included in the early warning range through the subsequent control mechanism.

[0061] In a specific example, the gas detection module includes gas sensing elements in five different response spectral bands, which respond to components such as acetaldehyde, butanone, and acetic acid respectively. During storage and transportation, data sampling is carried out every ten minutes, and a response vector set is constructed by comparing with historical samples in the past hour. The Mahalanobis distance algorithm is used to calculate the deviation between the current sample and the first five healthy samples, and the result is compared with the pre-set early warning threshold. Once the offset coefficient is higher than the threshold, a "high-risk" signal is immediately transmitted to the trend prediction model (that is, the time series judgment model constructed in the present invention), and the odor channel risk weighting sub-module is activated, so that the model pays more attention to the "olfactory abnormal trend" when judging the remaining shelf life, and thus the batch of corn is included in the high-risk pre-judgment list in advance for reference during circulation control.

[0062] The rapid microbial detection device adopts a non-destructive detection mechanism based on impedance method or bioluminescence method to capture the metabolic activity indicators of living colonies on the surface of corn kernels or inside the packaging. Before the detection data enters the prediction model, high-frequency sample fluctuation difference fitting is first performed. When the difference value exceeds the microbial metabolic trend deviation threshold, the critical sample backtracking section is entered for historical curve comparison. This comparison adopts one of the two strategies of moving average fitting and perturbation trend regression to determine whether it is a significant deviation event; when the data are in the deviation interval for three consecutive cycles, the prediction model increases the alert level of the batch of data, and directly adjusts it from low risk to medium and high risk interval.

[0063] Among them, the detection device based on the impedance method mainly indirectly determines the growth activity of microorganisms by measuring the change in the conductivity of the solution or surface medium between electrodes. Living colonies produce charge flow and ion concentration changes during metabolism, which causes the impedance value to shift by a specific amplitude. This method is suitable for direct contact with corn kernels or indirect detection through a tiny space probe embedded in the packaging environment. It has the advantages of no need to destroy the packaging structure and real-time detection.

[0064] Devices based on bioluminescence usually rely on the intensity of light signals released by specific enzyme systems (such as luciferase-luciferin system) involved in the fluorescence reaction of microorganisms during metabolism to determine their activity status. This device captures the intensity of fluorescence signals by setting up photoelectric probes and filter photosensitive components, and constructs a microbial activity map by combining parameters such as integration time and light intensity threshold.

[0065] The present invention is not limited to any specific implementation path of the impedance method or the bioluminescence method, both of which can be used to non-destructively and quickly obtain the activity status of microorganisms in the corn storage and transportation environment.

[0066] Before the detection data enters the prediction model, the high-frequency sample fluctuation difference fitting is first performed. When the difference value exceeds the microbial metabolic trend deviation threshold, the critical sample backtracking segment is entered for historical curve comparison. This comparison adopts one of the two strategies of moving average fitting and perturbation trend regression to determine whether it is a significant deviation event.

[0067] The above-mentioned "high-frequency sample fluctuation difference fitting" refers to the analysis of the change trend of the microbial activity data obtained by continuous detection at time intervals of minutes or hours. Its purpose is to filter out the data segments where the activity value changes dramatically in a short period of time. The fluctuation difference can be achieved by calculating the difference between the current data point and the mean value in the previous several cycles, or calculating the mutation point of the standard deviation sequence. The specific implementation methods are sliding window variance detection, incremental gradient analysis or differential fluctuation index method, etc.

[0068] When the above-mentioned fluctuation amplitude exceeds the set "threshold for deviation of microbial metabolism trend", it enters the critical sample analysis stage. In this stage, historical data of the current batch in the previous period will be retrieved for trend consistency comparison.

[0069] To improve the robustness of the judgment, the present invention provides two trend comparison strategies as optional implementation methods: Moving average fitting method: By performing weighted averaging on data points in the past several cycles, a stable trend line is constructed for deviation comparison with the current data; Perturbation trend regression method: Based on linear or non-linear fitting of slight perturbations in the time series, it is judged whether the current data change belongs to normal fluctuations within trend evolution or is an abnormal jump. The above two strategies can be used alone or in combination, or can be replaced by other algorithms with equivalent prediction and discrimination capabilities (such as adaptive curve regression, etc.).

[0070] When the data in the offset interval exists in three consecutive cycles, the prediction model raises the warning level for this batch of data, directly adjusting it from the low-risk level to the medium-high risk interval. The prediction model mentioned here is the quality trend prediction model constructed in the present invention, which predicts the evolution process of corn quality based on the time series modeling architecture and outputs a risk score.

[0071] The result of the microbial detection module serves as one of the model input channels. Its output value not only participates in the risk level scoring but also has the ability to trigger the internal risk weight switching mechanism of the model. Specifically:

[0072] When the detection results show a high-amplitude deviation in three consecutive cycles, it is defaulted that this batch is in an abnormal biological activity state;

[0073] The model will activate the risk weight adjustment sub-structure to increase the participation of the microbial data channel in the multi-channel fusion prediction calculation; at the same time, in the state classifier inside the model, this batch will be directly upgraded from the low-risk prediction label to the medium-risk or high-risk category to drive subsequent control actions, such as pausing shipments, starting refrigeration, etc. This mechanism ensures that the model no longer responds passively to visible abnormalities, but can complete early intervention in the mildew trend through this "hidden indicator" of microbial activity.

[0074] In a specific example, the detection device uses a micro impedance sensing head placed at the position of the air channel inside the corn packaging box, and collects the impedance change curve every fifteen minutes. The set threshold is that the change amplitude of the impedance value exceeds ten percentage points in two consecutive cycles. If it is within this offset range in three consecutive cycles, a trend mutation curve is obtained through moving average fitting, and in the subsequent quality trend prediction model, the risk weight of the microbial channel is multiplied by a doubling factor. At the same time, this batch is adjusted from the initial "stable state" to the "medium risk" state, and the processing suggestions of "strengthening the cold chain" and "sampling for re-inspection" are output.

[0075] The quality trend prediction algorithm adopts a multi-level time series modeling strategy, combines short-term trend windows and periodic trend windows for dual evolution analysis, and introduces a causal weight optimization mechanism in the core part of the model. This mechanism adaptively and dynamically adjusts the data weight ratio participating in modeling based on the causal relationship strength of different data channels.

[0076] In the present invention, the quality trend prediction algorithm corresponds to the above-mentioned "quality trend prediction model". Its core function is to dynamically model the quality evolution trend of fresh corn during storage, circulation, sales and other links, and output the estimated shelf life and risk level score to guide downstream risk control and circulation strategy decision-making.

[0077] The multi-level modeling strategy refers to simultaneously constructing trend evaluation windows on two time scales in the structure of this prediction algorithm:

[0078] Short-term trend window: It is used to capture the micro-fluctuation trend of the quality change of fresh corn in the recent few hours or dozens of minutes. This window is suitable for real-time early warning functions and mainly focuses on recent sudden quality changes;

[0079] Periodic trend window: It is used to evaluate the medium- and long-term evolution pattern of the quality of fresh corn under multiple acquisition cycles, such as the gradually declining trend caused by environmental fluctuations. This window can be set to a sliding evaluation cycle based on two days, three days or one week as a unit.

[0080] Both windows are input in the form of time series data, including multiple channels such as image feature parameters, gas volatile component concentrations, temperature and humidity records, and microbial metabolic activity data. Parallel modeling of the trends on two scales is carried out respectively, and fusion output is performed at the prediction end.

[0081] Specifically in the algorithm, the short-term window can adopt a time series processing structure based on a recurrent neural network or a sliding regression model, while the periodic window can be modeled based on exponential weighted average, long short-term memory network, or seasonal decomposition model, etc. The above methods are only one of the feasible paths. In the present invention, the causal weight optimization mechanism is used to solve the problem of the changing contribution degree of different types of data channels to the quality evolution trend, and improve the prediction flexibility and accuracy of the model in different environments or sample scenarios. This mechanism is embedded in the internal structure of the quality trend prediction model and is used to dynamically adjust the participation ratio of different input data. Its basic logic is:

[0082] By statistically analyzing the correlation degree between image features, gas indicators, microbial activities and quality scores; extracting the causal relationship strength index between each channel and the target variable, and converting this strength into the weight ratio participating in modeling; updating the weight configuration of each channel in real time according to the current batch or environmental state to enhance the response ability to the dominant factors.

[0083] The calculation of the causal relationship strength can be implemented using one of the following three types of methods:

[0084] Granger causality test method: To test whether a channel variable "precedes" the change in the performance degradation in a statistical sense;

[0085] Mutual information method: To evaluate the correlation by calculating the information sharing degree between the input variable and the target quality parameter;

[0086] Minimum description length principle: To measure the contribution degree of different channels to the model's explanatory ability.

[0087] The results of the double-layer trend modeling are fused at the model output end to form an integrated trend prediction result. This result includes:

[0088] A comprehensive quality degradation trend curve; a mark of the current quality trend segment; the expected salable period (i.e., the shelf life); a risk level score (divided into five levels from stable to high risk).

[0089] The prediction result is automatically transmitted to the "risk response control module" described in the present invention for determining whether to enter measures such as suspension of circulation processing, adjustment of packaging strategy, or pre-cooling start.

[0090] The quality degradation trend curve refers to a curve-type data structure generated by continuous learning of the multi-dimensional quality data of fresh corn collected through a time series modeling strategy in the present invention, and is used to dynamically describe the change trajectory of the quality state of fresh corn over time in each stage of storage, transportation, and circulation.

[0091] The construction process of the quality degradation trend curve is as follows:

[0092] Input data source: Continuous time series data from the image acquisition module, gas detection module, temperature and humidity sensing module, and microorganism detection module form a multi-channel input matrix.

[0093] Feature representation processing: The data of each channel can be pre-processed such as standardization, trend normalization, and principal component compression to extract its representative change path.

[0094] Time series modeling: Continuously train the data stream through a time series prediction model (such as a recursive neural network, long short-term memory model, moving weighted model, etc.) to form a mapping relationship between time and quality state.

[0095] Trend curve generation: After concatenating and smoothing the prediction outputs of the model at each time node, a continuous curve that varies on the time axis is obtained, which is called the quality degradation trend curve.

[0096] Key point extraction: Based on structural features such as the slope change, inflection point, and critical segments of the curve, state intervals such as the rapid quality decline period, stable period, and critical period can be identified.

[0097] This curve is essentially an "image of the time function of quality score", that is, by analyzing the time series changes of multi-channel data, a quality trend scoring function that decreases or fluctuates over time is derived.

[0098] Based on the quality degradation trend curve, the present invention designs the following two derived modules at the model output end:

[0099] Risk level scoring module: Compare the trend score corresponding to the current time point with the historical healthy sample distribution, and assign different levels such as "stable state", "initial risk", "medium risk", and "high risk" according to its position in the risk distribution interval.

[0100] Shelf life prediction module: Based on the slope trend between the current value and the termination value of the trend curve, calculate when it will drop to the inedible critical value (such as the score corresponding to the three-hundredth percentile of historical samples), so as to deduce the remaining available sales duration. For example, if the current trend score is eighty, the termination value is forty, and the average decline rate of the curve in the past three hours is five score units per hour, the shelf life can be deduced to be about eight hours as a reference for the available sales safety time limit.

[0101] In a specific example, the short-term trend window is set to six hours, the periodic trend window is a three-day sliding period, and the input data includes the change value of image spot aggregation, the integrated value of volatile gas concentration, the fluctuation range of air humidity, and the total light signal of colony activity. First, the data in the two windows are modeled using a recursive regression model and an exponentially weighted moving average model respectively.

[0102] Subsequently, the mutual information algorithm is called to calculate the correlation score between each channel and the shelf life label, and the modeling input weight configuration is adjusted accordingly. It is found that the causality of the gas data channel significantly increases in the current environment, so the model automatically increases the participation weight of this channel, making the gas factor have a greater impact on trend judgment. Finally, the remaining shelf life cycle of the current batch of products is predicted to be 24 hours, the risk level is "initial risk", and this result is transmitted to the management platform for the control decision-making module to call and execute.

[0103] In another specific example, the change rate of the image features, the gas concentration curve, the humidity fluctuation range, and the microbial metabolism index of fresh corn are collected, and continuous modeling is carried out with a 30-minute cycle. The constructed quality degradation trend curve shows that the score remains between 90 and 85 within the first six hours, and then starts to decline rapidly after being triggered by a high-humidity event, dropping to 65 at the tenth hour, and the trend significantly turns towards rapid deterioration. This is determined as the "critical downward section", the current risk level is adjusted to "medium risk", and it is predicted that if the current downward trend remains unchanged, it will fall into the unsafe score range (such as a score below 50) after six hours, thus obtaining a remaining shelf life of approximately six hours. This curve not only provides a basis for the internal warning mechanism but also is presented to the operation personnel through the platform's graphical interface, providing an auxiliary judgment basis for warehousing or shipping strategies.

[0104] While the prediction result generates the shelf life cycle, the risk level is divided into five levels, namely stable state, fluctuation warning, initial risk, medium risk, and high-risk decay state, and each level corresponds to different warning responses and disposal strategies.

[0105] The prediction result includes a quality degradation trend curve and the trend score value at the current time point. The current score value is compared with the existing score interval threshold to determine its belonging level in the five-level classification. This level is not only used for label identification but also directly triggers different downstream disposal logics.

[0106] Definition description of the five-level risk level: Definition of stable state: The current quality trend score is higher than the set stable threshold (for example, the score value is greater than 85, with a full score of 100), and the trend curve has no obvious downward trend; Characteristics: No peculiar smell, no obvious appearance abnormality, and normal microbial activity; Response: Allow normal circulation without additional intervention.

[0107] Definition of fluctuation warning: The score is in the range of 85 to 75, or there is a slight short-term decline; Characteristics: Obvious environmental fluctuations, but the quality is still within an acceptable range; Response: Record the warning and recommend strengthening the packaging seal or entering the short-chain sales channel.

[0108] Definition of initial risk: The score is in the range of 75 to 65, and the trend curve shows an accelerating downward inflection point; Characteristics: The microbial metabolism begins to rise, and there is still no obvious change in appearance; Response: Activate the warning prompt and recommend selling in advance or activating the pre-cooling mechanism.

[0109] Definition of medium risk: The score is between 65 and 55, and the downward rate of the trend curve is significant; Characteristics: The "sensitivity coefficient of deteriorated smell deviation" exceeds the warning value; Response: Suspend the shipment of this batch, transfer to the re-inspection process or activate the active sampling inspection mechanism.

[0110] Definition of high-risk putrefaction state: Score below 55, and the quality trend curve is close to the end value; Characteristics: Abnormal height of multi-channel signals, visible putrefaction or obvious peculiar smell; Response: Block the circulation permission of this batch, mark it as "unsellable", and it is recommended to destroy or recycle.

[0111] The score boundaries of each risk level can be set in the following ways: Empirical value setting: Based on the empirical distribution of historical quality data, take the mean percentile of stable batches to set the threshold; Clustering algorithm division: Use unsupervised clustering methods (such as k-means or Gaussian mixture model) to stratify and classify the score values; Supervised learning classification model: Such as support vector machine, random forest and other models, train the samples with existing labels to obtain the division interface. The present invention does not limit that a certain algorithm must be used for risk interval division. As long as a stable correspondence relationship between the score segments and the risk levels can be achieved, it can be regarded as a technically equivalent path.

[0112] The input of this risk division mechanism is the output content of the quality trend prediction model mentioned in the present invention, specifically the trend score value (i.e., the output value at the current moment of the quality scoring function) obtained by modeling based on multi-modal perception data in the model. The scoring generation module in the prediction model can directly map the regression prediction result to a score value, or can be obtained by normalizing the comprehensive features (such as the current value, curvature, slope) of the trend curve.

[0113] For example, the score value can be calculated in the following way (one of the implementable examples): Divide the point value of the trend curve at the current moment by the maximum value at the initial moment, multiply by one hundred to form a relative quality index; Use the slope change factor (trend fluctuation degree) as the weight coefficient to adjust the scoring result; The final score is mapped between zero and one hundred and enters the risk level classification module. Through the correspondence between this score and the five-level risk standard interval, automatic determination and level marking are realized, and each level corresponds to different response paths in the subsequent control logic.

[0114] In a specific example, a real-time quality trend prediction model is used to evaluate a batch of fresh corn. The current trend score value is 62, which is at the critical point between "initial risk" and "moderate risk". The slope of the trend curve has increased from 0.2 to 0.5 in the past three hours, indicating that the deterioration trend is accelerating. It is automatically determined that this batch enters the "moderate risk" interval, and the following processing logic is triggered: Issue a suspension of shipment instruction; Start a microbial sampling inspection plan and notify the operation platform to complete the operation.

[0115] The adaptive reconstruction process of the parameter recognition model includes three parts: a sensitivity dynamic adjustment module, an identification error feedback module, and an environmental response weight mapping mechanism. After detecting the trigger of a high-humidity environment, it automatically reduces the "subtle gray-scale speckle variation response threshold" in the color recognition model and simultaneously increases the discrimination factor for odor intensity changes. When the environmental fluctuations do not return to the normal state within three consecutive cycles, the identification error feedback module is activated to adjust the current model prediction boundary based on the true judgment result of the previous cycle.

[0116] This model is different from the quality trend prediction model. It does not directly output the shelf life or risk score, but its identification output result will be one of the important inputs of the quality trend prediction model, affecting the accuracy and stability of the latter's judgment result.

[0117] The sensitivity dynamic adjustment module is used to dynamically adjust the threshold in the recognition parameters according to the external environmental state. Environmental detection mechanism: Continuously collect data from the temperature and humidity sensing module. When it is detected that the current humidity value continuously exceeds the established "high-humidity threshold" for more than thirty minutes and the temperature synchronously rises by more than five degrees Celsius, it is determined as a "high-humidity fluctuation event".

[0118] Dynamic adjustment response content: After this event is triggered, the key discrimination value used to detect the subtle gray-scale changes in the image in the color recognition model will be adjusted, specifically including: automatically reducing the "subtle gray-scale speckle variation response threshold", for example, reducing the gray-scale difference judgment value from fifteen units to ten units; adjusting the sensitivity of the image noise filter to enhance the response ability to low-contrast areas; at the same time, increasing the weight of the discrimination factor for odor intensity changes, specifically by amplifying the fluctuation amplitude value of the current odor response vector to make it have a greater impact on the trend modeling output. The adjustment of the above "response threshold" and "discrimination factor" can be automatically completed by setting a function model. For example, in the embodiment of the present invention, a Sigmoid function form can be used to establish a threshold adjustment function, so that the response value adjustment is more sensitive near the critical point of environmental parameter changes. Alternative methods such as a linear coefficient increasing model and a stepwise step function can also be used to achieve the dynamic adjustment behavior.

[0119] The environmental response weight mapping mechanism is used to automatically adjust the weight ratio of multi-channel perception data in the recognition process according to the external environmental disturbance situation. Implementation path: Preset the association mapping model between each channel of data (image, gas, humidity, microorganism, etc.) and different environmental factors. For example, the response weight of the odor channel to the increase in humidity can be defined as a linear amplification curve, and the response of the image channel to vibration effects can be set as an inhibition function.

[0120] Response Logic: When entering a high-humidity or vibration-fluctuation state, the participation weights of different channels in the recognition module are automatically adjusted according to the mapping model. For example: In a high-humidity state, the recognition weight of the odor channel is increased by twenty percentage points; the recognition weight of the image edge is slightly suppressed to avoid misjudgment of image blurring caused by water vapor. This mapping mechanism can be implemented through weight matrix setting, Bayesian network modeling, or fuzzy logic control. The above methods can be used interchangeably and are not limited to a specific method.

[0121] Recognition Error Feedback Module and Prediction Boundary Adjustment Mechanism: When the environmental fluctuation does not return to the normal state within three consecutive cycles, the recognition error feedback module is activated, and the current model prediction boundary is adjusted based on the true judgment result of the previous cycle.

[0122] Trigger Condition Description: Define "consecutive abnormal cycles" as the situation where, in three or more cycles, the environmental parameters are still in the high-risk range, and there is a deviation between the judgment result of the recognition model and the true state such as manual sampling inspection and laboratory feedback.

[0123] Boundary Adjustment Logic: Calculate the error deviation value between the prediction result of the model in the previous cycle and the actual judgment label; if the deviation value is higher than the set error tolerance rate (such as one-tenth), it is determined as "recognition instability"; the model boundary adjustment includes: lowering the current recognition decision threshold, narrowing the classification boundary distance, and widening the abnormal definition trigger range.

[0124] For example, for the boundary classifier in the image recognition model, the original normal judgment boundary with a score of eighty can be adjusted to eighty-five to reduce the possibility of missed judgment. The adjustment method can adopt an adaptive boundary regression algorithm, a boundary buffer strategy, or introduce a positive and negative sample weight reallocation mechanism.

[0125] Relevance to the Quality Trend Prediction Model in the Present Invention: The change in the recognition boundary will affect the weights or credibility given to multi-channel data in trend modeling, thereby indirectly affecting the final shelf-life prediction and risk scoring results.

[0126] In a specific example, it is deployed in a long-distance transportation cold chain. During the journey, it encounters a continuous high-humidity environment, and the humidity rises from sixty-five to eighty-five and lasts for more than one hour. The image recognition model originally recognized mildew based on the change of gray spots, with a default threshold of fifteen, which is automatically adjusted to ten; the fluctuation sensitivity coefficient of odor recognition is doubled to the original value. The judgment accuracy drops below eighty-five within three consecutive cycles (each cycle is thirty minutes), and the error feedback mechanism is activated. The inspection sample is "initial mildew", but the model judges it as "stable", with an obvious deviation. Therefore, the current recognition boundary is adjusted from eighty-five to ninety to improve the judgment sensitivity. This adjustment immediately affects the result of the quality trend prediction model in the next cycle. The risk score of the model for the next batch of corn is increased from the original sixty-five to seventy, and the grade is changed to "initial risk", successfully triggering the pre-cooling measure.

[0127] Data cross-analysis adopts a mechanism that combines vector similarity matching and time-series differential fusion. Feature vectors are constructed for data channels such as images, gases, temperature, and humidity respectively, and an omnidirectional data cross-section map for each batch of fresh corn within a time period is formed. If the similarity index between the maps is detected to decrease and the predicted risk score approaches the quality anomaly trigger threshold, this batch will be automatically marked as a "key monitoring object". Such batches will trigger a secondary quality inspection mechanism and data refresh training before entering the warehousing or transportation process to strengthen the model's accurate recognition ability for boundary samples.

[0128] Construction of multi-channel feature vectors: For each batch of fresh corn within a given time period (such as continuous detection records within 24 hours), multi-dimensional features are collected from channels such as images, gases, temperature, and humidity. Examples are as follows:

[0129] Image channel: Such as color mean value, variance of gray-scale distribution, complexity of edge texture, etc.;

[0130] Gas channel: Such as odor response intensity, ratio between components;

[0131] Temperature and humidity channel: Such as standard deviation of temperature, humidity fluctuation gradient, etc.

[0132] The features of the above-mentioned channels are aggregated into a high-dimensional vector at the same time point, and each dimension corresponds to a processed physical or chemical index value.

[0133] Generation of cross-section map: The multi-dimensional vector sequences of consecutive time periods are spliced to construct an omnidirectional data cross-section map of this batch of products within this time period, which is essentially a two-dimensional matrix of time series-channel indicators, where the horizontal axis is time and the vertical axis is indicators extracted from different data channels. This map is the basis for subsequent similarity matching and anomaly recognition.

[0134] Vector similarity matching and time-series differential fusion mechanism: Vector similarity matching: The map of the current batch is matched with the map of historical healthy samples in the vector space, and the similarity calculation can be implemented by any of the following methods: Cosine similarity: Used to measure the angle between the direction of the current vector and the historical standard vector, the closer the value is to 1, the more similar; Euclidean distance: Used to measure the distance between the current sample and the historical sample in the feature space; Dynamic time warping method: Suitable for the overall similarity matching of time-series maps, considering the time offset problem. In the present invention, no specific algorithm is specified, and as long as the similarity measurement can be achieved, it can be substituted and implemented.

[0135] Temporal Difference Fusion: Based on vector comparison, further calculate the temporal difference change trend between the current spectral map and the spectral map of the past cycle, and extract factors such as the change rate, direction, and fluctuation intensity to determine whether there is a "specific deviation" in the current batch. The fusion mechanism comprehensively reflects the potential risk changes of product quality by superimposing the two-dimensional indicators of "spatial difference" and "time drift".

[0136] Determination Logic and Operational Response for "Key Monitoring Objects": When it is identified that the spectral similarity index is lower than the preset similarity threshold (such as 80%), and the risk score value output by the current quality trend prediction model is close to the quality anomaly trigger threshold (such as only two score differences from the moderate risk), then automatically mark the batch sample as a "key monitoring object". The key monitoring object will trigger the following measures: interrupt the original warehousing or transportation process; enter the secondary quality inspection mechanism for manual review and sample random inspection; re-enter the data of this batch into the quality trend prediction model of the present invention for data refreshing training to improve the model's determination ability in the boundary sample area and enhance the model's recognition accuracy for the "fuzzy interval".

[0137] Linkage Mechanism with the Quality Trend Prediction Model: The spectral map and the similarity calculation output do not directly generate a risk score, but are passed as auxiliary input factors into the quality trend prediction model defined in the present invention as the support basis for the following functions: adjusting the confidence parameter of the model input sample; starting the internal anomaly deviation discrimination logic of the model (such as dynamically adjusting the trend slope determination range); activating the retraining mechanism of the model to increase the number of boundary samples and improve the model's generalization ability. All batches marked as "key monitoring objects" will be fed back to the model input structure through the data interface, thus forming a complete logical chain of "perception → recognition → reflux training → improvement of recognition ability" in a closed loop.

[0138] In a specific example, during the process of a batch of fresh corn being transferred from normal temperature to refrigeration, the humidity changes violently. The image features and gas features collected form a multi-dimensional vector, and its cosine similarity with the standard sample is only 75%, which is lower than the threshold. At the same time, the risk score value output by the quality trend prediction model in the present invention is 66, only one point away from the "moderate risk" boundary. Immediately mark this batch as a key monitoring object: stop the transportation plan; arrange a special person for re-inspection; send the sample data of this batch into the training set to refresh the model boundary learning parameters and improve the subsequent recognition accuracy.

[0139] The centralized management platform graphically displays the post - harvest processing, packaging forms, transportation routes, and quality change trends of different batches of fresh - eating corn based on a visual chain structure. Each time a prediction cycle is processed, the current input data status and output prediction results are recorded and matched with the actual quality feedback. If the prediction deviation amplitude is greater than the prediction accuracy stability threshold for two consecutive cycles, the platform automatically marks that the current model needs to be retrained and selects the latest batch of stable samples as the training basis.

[0140] The visual chain structure refers to organizing the whole - process information from different batches of fresh - eating corn in a graphically - traceable data chain. Its structural logic corresponds one - to - one with the actual circulation process of fresh - eating corn and includes the following nodes:

[0141] Post - harvest processing information: such as the operation time and parameters of cleaning, sterilization, cooling, etc.;

[0142] Packaging form information: the type of packaging material used, barrier performance, encapsulation pressure, etc.;

[0143] Transportation route record: the logistics centers passed through, the transportation temperature and humidity trajectory, vibration records;

[0144] Quality change trend: the trend score curve, risk level determination result, and predicted shelf - life output generated by the quality trend prediction model defined in the present invention. The above - mentioned nodes are organized into a chain structure through timestamps, batch numbers, and spatial location logic. The chain is presented to the platform user through a graphical interface to achieve integrated data tracking, analysis, and management.

[0145] The processing logic of data record and feedback matching: Each time a prediction cycle is processed, the current input data status and output prediction results are recorded and matched with the actual quality feedback.

[0146] When each quality prediction behavior is executed, the platform will record the following content:

[0147] Input data status: including the current feature vectors of the image channel, gas channel, temperature and humidity channel, and microorganism channel;

[0148] Model output: the trend score value, predicted shelf - life cycle, and risk level generated by the quality trend prediction model described in the present invention;

[0149] Actual quality feedback: the true quality status label from terminal sampling inspection, consumer feedback, or manual sensory evaluation.

[0150] The matching behavior can be achieved through methods such as simple label comparison, score difference analysis, or error distribution evaluation. For example, when using the score difference method, the maximum allowable error (such as five percentage - point score units) between the predicted value and the measured value can be set to judge whether the current model is stably predicting.

[0151] Prediction accuracy judgment criteria and retraining trigger mechanism: If the prediction deviation amplitude is greater than the prediction accuracy stability threshold for two consecutive cycles, the platform automatically marks that the current model needs to be retrained.

[0152] Definition of prediction deviation: The prediction deviation is the numerical gap between the model output score value and the true score value. For example, if the model predicts that the remaining shelf life is 30 hours, but the actual product mildews after 20 hours, the deviation is 10 hours.

[0153] Explanation of the stability threshold: The stability threshold can be set according to the historical model performance. For example, take the standard deviation of the prediction errors of the past 100 batches multiplied by a proportionality coefficient, or set it as a fixed threshold (such as the error shall not exceed 10%).

[0154] Judgment of consecutive deviations: The judgment of two consecutive cycles means that in two consecutive prediction behaviors of the same batch of products, if the situation of exceeding the error threshold occurs, it is regarded as a decline in the short-term prediction stability of the model, triggering the self-learning mechanism.

[0155] Selection logic of retraining samples and model update: The platform automatically marks that the current model needs to be retrained and selects the latest batch of stable samples as the training base.

[0156] Sample selection method: Select batches with complete data and clear evaluation as the training basis from the latest samples in the current cycle that have no deviation, and the risk level is "stable state" or "fluctuation warning".

[0157] The training base may include: input feature vectors (including image, gas, environment and microbial features); quality trend score sequence; actual shelf life record or manual label.

[0158] Model update process: The platform sends the above training samples into the quality trend prediction model in the present invention and performs incremental training or complete retraining. The training methods may include:

[0159] Parameter iteration based on the sliding window method; fine-tuning the backend of the model using transfer learning methods; replacing the feature embedding layer structure to adapt to new packaging or processing conditions. The platform can set the training trigger frequency. For example, automatically detect the deviation status once a day. Once the cumulative number of marked batches reaches the threshold (such as three batches), the training behavior is forced to be triggered.

[0160] In a specific example, the batch number Z-123 was recorded with a score of 75 before transportation, the risk level was "initial risk", and the predicted shelf life was 48 hours. However, through on-site random inspection, it was found that mildew occurred at the 30th hour, and the prediction error reached 18 hours, exceeding the maximum error threshold of 10 hours set by the platform.

[0161] The deviation was recorded as an unstable event. Then, in the next cycle, another batch (number Z-124) was predicted again, with a predicted shelf life of 50 hours, but the actual measurement was only 28 hours, with a deviation of 22 hours. The identification deviation exceeded the standard twice in a row, and the current model's prediction ability was automatically determined to be reduced. The platform immediately performed the following operations: marking the current model as "needs optimization"; automatically selecting the data of stable batches Z-127 and Z-130 in the past day as the new training set; triggering model retraining, and recording the update log in the chain map for user audit.

[0162] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0163] The present invention focuses on the need for mildew prevention and preservation of fresh corn from harvest to circulation, and constructs an intelligent quality monitoring method with multimodal perception, trend modeling, intelligent judgment and dynamic control as the core. First, multi-dimensional feature input is constructed through image acquisition, gas detection, humidity and temperature perception and rapid identification of microorganisms to form a full data set that can be used for intelligent modeling; the system uses a multi-level time series modeling method to construct a quality degradation trend curve, predict the shelf life and output a risk level score, and at the same time, adaptively adjusts parameters such as image color recognition sensitivity and odor recognition threshold to improve the recognition accuracy of early "hidden mildew".

[0164] Based on the operation of the quality trend prediction model, the present invention introduces a data cross-analysis mechanism and abnormal sample recognition logic, uses vector similarity and time series difference algorithms to identify potential boundary samples, and triggers secondary quality inspection and data refresh training; at the same time, the system builds a five-level risk level decision-making system based on the comparison of prediction results and actual feedback, and automatically matches corresponding control strategies for different levels, including packaging replacement, circulation suspension, cold chain reinforcement and other operations, to achieve a transition from "passive monitoring" to "active intervention".

[0165] In order to achieve intelligent closed-loop management of the entire process, the present invention builds a visual chain traceability structure on the platform side to record the processing path and quality trend of each batch of corn. When the continuous prediction deviation exceeds the stability threshold, the platform automatically identifies the model degradation state, triggers model retraining and selects the latest stable sample to update the learning basis, realizing the self-evolution of the quality trend prediction model and the sustainable improvement of the long-term accuracy of the system.

[0166] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean 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 application.

[0167] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professionals can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of this application.

[0168] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above-described devices and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0169] As described above, the above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed by this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. A method for preventing mildew and maintaining the quality of fresh corn during circulation based on intelligent monitoring, characterized in that, It includes the following steps: By configuring an image acquisition module, a gas detection module, a humidity and temperature sensing module, and a rapid microbial detection device, multi-source quality data of fresh corn are synchronously collected in different circulation links, including color, epidermal texture, gas volatile concentration, temperature and humidity changes, and pathogen metabolic indicators; Using a quality trend prediction algorithm based on time series modeling, the above multi-dimensional data are serially trained to construct a quality degradation trend curve, automatically judge the degree of quality deterioration and the future shelf life cycle, and assign a risk level score; When high humidity fluctuations or shock events occur in the storage and transportation environment, the color recognition sensitivity, gas warning threshold, and microbial detection frequency are adjusted in real time to achieve adaptive reconstruction of the parameter recognition model, so as to improve the recognition accuracy in the initial stage of mildew; Perform multi-dimensional cross-analysis on the multi-modal data of each batch of fresh corn. When the risk level predicted by the recognition model exceeds the standard warning value, a circulation suspension processing signal is triggered, and potentially hidden mildew batches are preferentially removed to prevent products with normal appearance but internal deterioration from entering the sales process; Upload the data collected at each stage, prediction results, and processing records to a centralized management platform through a data interface, and automatically archive them to form a quality traceability chain.

2. The method for preventing mildew and maintaining the quality of fresh corn during circulation based on intelligent monitoring according to claim 1, characterized in that, The image acquisition module includes a high-resolution image sensor and a near-infrared reflection unit. The reflected images in the preset bands are used to extract the weak color difference on the surface of corn kernels and the change signals of local wet spot areas. When there is no obvious abnormality in the normal visible light detection, potential mildew signs are judged by spectral enhancement feature recognition; the image data undergoes color space conversion and region segmentation operations in the preprocessing stage, converting the image from the red-green-blue color space to the lightness-saturation-hue space, and performing dynamic difference calculation in combination with the edge gradient change and the mean value of color difference distribution, so as to improve the recognition probability of early mildew micro-clusters.

3. The method for preventing mildew and maintaining the quality of fresh corn during circulation based on intelligent monitoring according to claim 2, characterized in that, The gas detection module is constructed based on the principle of an electronic nose, and volatile organic compounds in the corn storage environment are identified through multi-channel metal oxide gas sensitive elements. The identification includes aldehydes, ketones, and low molecular weight fatty acid components in mold metabolites. By constructing a characteristic response vector under different odor intensity combinations, odor pattern recognition is achieved; at the same time, the sensitivity coefficient of the deteriorated odor shift is set. When the coefficient value is higher than the warning threshold of the gas sensing sensitive response, the weight adjustment of the mildew process in the prediction model is triggered, and a high risk tendency is given to the recent data segment to achieve early locking of the mildew path. The sensitivity coefficient of the deteriorated odor shift is calculated based on any one of the three discriminant methods of Mahalanobis distance, Manhattan distance, and Shannon entropy.

4. The method for preventing mildew and maintaining quality of fresh corn during circulation based on intelligent monitoring according to claim 3, characterized in that, The rapid microbial detection device adopts a non-destructive detection mechanism based on impedance method or bioluminescence method to capture the metabolic activity indicators of living colonies existing on the surface of corn kernels or inside the packaging. Before the detection data enters the prediction model, high-frequency sample fluctuation difference fitting is first performed. When the difference value exceeds the microbial metabolic trend shift threshold, it enters the critical sample backtracking segment for historical curve comparison. This comparison adopts one of the two strategies of moving average fitting and perturbation trend regression to judge whether it is a significant shift event; When the data within the offset interval occurs in three consecutive cycles, the prediction model raises the alert level for this batch of data, directly adjusting it from the low-risk level to the medium-high risk interval.

5. The method for preventing mildew and maintaining the quality of fresh corn during circulation based on intelligent monitoring according to claim 4, characterized in that, The quality trend prediction algorithm adopts a multi-level time series modeling strategy, combines short-term trend windows and periodic trend windows for dual evolutionary analysis, and introduces a causal weight optimization mechanism in the core part of the model. This mechanism adaptively and dynamically adjusts the data weight ratio participating in modeling based on the causal relationship strength of different data channels.

6. The method for preventing mildew and maintaining the quality of fresh corn during circulation based on intelligent monitoring according to claim 5, characterized in that, When generating the shelf life cycle of the prediction result, the risk level is divided into five levels, namely stable state, fluctuation warning, initial risk, medium risk, and high-risk decay state. Each level corresponds to different warning responses and disposal strategies.

7. The method for preventing mildew and maintaining quality during the circulation of fresh corn based on intelligent monitoring according to claim 6, wherein The adaptive reconstruction process of the parameter identification model refers to automatically reducing the mutation response threshold of fine gray-scale speckles in the color identification model and simultaneously increasing the discrimination factor for odor intensity changes after detecting the trigger of a high-humidity environment; when the environmental fluctuations do not return to the normal state within three consecutive cycles, the prediction boundary of the current model is adjusted based on the true determination result of the previous cycle.

8. The method for preventing mildew and maintaining quality during the circulation of fresh corn based on intelligent monitoring according to claim 7, characterized in that, Data cross-analysis adopts a mechanism based on vector similarity matching and time series difference fusion. Feature vectors are constructed for data channels such as images, gases, temperature, and humidity, and an omnidirectional data cross-section map for each batch of fresh corn within a time period is formed. If the similarity index between the maps is detected to decrease and the predicted risk score approaches the quality anomaly trigger threshold, this batch is automatically marked as a key monitoring object. Such batches will trigger a secondary quality inspection mechanism and data refresh training before entering the warehousing or transportation process to enhance the model's accurate identification ability for boundary samples.

9. The method for preventing mildew and maintaining quality during the circulation of fresh corn based on intelligent monitoring according to claim 8, characterized in that, The centralized management platform graphically displays the post-harvest processing, packaging form, transportation path, and quality change trend of different batches of fresh corn based on a visual chain structure. Each time a prediction cycle is processed, the current input data state and output prediction result are recorded and matched with the actual quality feedback. If the prediction deviation amplitude is greater than the prediction accuracy stability threshold in two consecutive cycles, the platform automatically marks that the current model needs to be retrained and selects the latest batch of stable samples as the training base.

Images