Agricultural product processing quality traceability monitoring method and system based on edge nodes

Abstract

The invention provides an agricultural product processing quality traceability monitoring method and system based on edge nodes, and the method comprises the steps: constructing an analytic hierarchy process model through employing the multi-source data of time-space alignment, decomposing an agricultural product processing quality evaluation index into a plurality of sub-indexes, forming an index hierarchical structure, and employing a fuzzy comprehensive evaluation method to obtain a traceability result of the agricultural product processing quality. The importance weight and the satisfaction degree of each sub-index are comprehensively considered, the comprehensive score of each scheme is calculated, and the processing quality of the agricultural products is judged; after the quality problem is processed, key control points and influence factors in the agricultural product processing process are analyzed through data mining, a quality evaluation model and a rule base are dynamically updated by adopting an incremental learning algorithm, and continuous optimization of the model is realized by continuously receiving new data samples and updating model parameters; and the accuracy of agricultural product processing quality evaluation is continuously improved.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to an edge node-based agricultural product processing quality traceability monitoring method and system. Background Art

[0002] In the agricultural product processing quality traceability monitoring system based on edge nodes, there are technical difficulties in multi-source heterogeneous data fusion and decision-making. The system needs to process and analyze data from various sensors, such as images, temperature, humidity, gas concentration, etc. These data differ in format, sampling frequency, data volume, etc. How to achieve efficient and real-time data fusion and decision-making under the limited computing resources and storage capacity of edge nodes is a key technical issue. Specifically, the monitoring of key quality indicators in the agricultural product processing process needs to take into account multiple factors, and a single type of data may not fully reflect the quality status. For example, in the fruit and vegetable processing process, it is not enough to judge the appearance quality of agricultural products only by image recognition. It is also necessary to combine temperature and humidity data to analyze the impact of the storage environment on quality. However, different types of data may not be completely synchronized in time and space. How to align and correlate data requires the design of appropriate data fusion algorithms. At the same time, the quality assessment of agricultural product processing involves multiple indicators. How to weigh the importance of each indicator and build a comprehensive quality assessment model is also a complex decision-making problem. In addition, when performing data mining and pattern recognition at the edge node, how to choose a suitable machine learning algorithm under limited computing resources and balance computing efficiency and analysis accuracy is also a technical difficulty that needs to be considered. Summary of the invention

[0003] The present invention provides a method for monitoring the quality of agricultural product processing traceability based on edge nodes, which mainly includes:

[0004] Acquire multi-source heterogeneous sensor data, establish a unified data model in advance according to the differences in format, sampling frequency and quantity of the multi-source heterogeneous sensor data, map the multi-source heterogeneous sensor data to the unified data model, use data interpolation and downsampling methods to align the time axis of the multi-source heterogeneous sensor data, eliminate the sampling frequency differences, and obtain a time-synchronized multi-source data set; based on the time-synchronized multi-source data set, construct a spatiotemporal data cube model, map the multi-dimensional data including images, temperature and humidity to a unified spatiotemporal coordinate system, and on the basis of the spatiotemporal data cube model, Through data interpolation and aggregation methods, the coarse-grained data is refined and the fine-grained data is summarized, so that the spatiotemporal granularity of data from different sources tends to be consistent; using the multi-source data aligned in time and space, a hierarchical analysis method model is constructed to decompose the agricultural product processing quality evaluation index into several sub-indicators to form an indicator hierarchy structure; using the fuzzy comprehensive evaluation method, the importance weight and satisfaction of each sub-indicator are comprehensively considered, and the comprehensive score of each scheme is calculated to judge the agricultural product processing quality; according to the evaluation results of agricultural product processing quality, the feature analysis of agricultural product processing quality evaluation data is carried out to screen out the key features with greater influence and reduce the data Dimension, using decision tree, support vector machine machine learning algorithms, while ensuring the accuracy of analysis, reduce computing resource consumption, and realize data mining and pattern recognition in resource-constrained environments; based on the analysis results of the machine learning algorithm, using the rule engine, compare the quality assessment indicators with the preset threshold, automatically determine whether there are quality defects in the agricultural product processing process, if there are quality defects in the agricultural product processing process, trigger the early warning mechanism, through data visualization, display the problem links and causes, assist in the rapid location and disposal of quality problems; after the quality problem is handled, through data mining, analyze the key control points and influencing factors in the agricultural product processing process, using incremental learning algorithms, dynamically update the quality assessment model and rule base, by continuously receiving new data samples, update model parameters, realize continuous optimization of the model, and continuously improve the accuracy of agricultural product processing quality assessment; record the quality assessment results, optimization plan and processing process in the blockchain to form an unalterable quality traceability chain, use cryptographic methods to encrypt and protect sensitive data, ensure data security and privacy, and automatically determine and trace the responsibility for agricultural product quality through the smart contract mechanism, so as to achieve full traceability, verifiability and accountability of agricultural product quality and enhance consumer trust.

[0005] The present invention provides an agricultural product processing quality traceability monitoring system based on edge nodes, which mainly includes: a data preprocessing module for format unification and time synchronization of multi-source heterogeneous sensor data; a spatiotemporal data modeling module for building a spatiotemporal data cube model and aligning data granularity; a quality assessment module for building a hierarchical analysis method model and performing fuzzy comprehensive evaluation; a data mining module for feature analysis, dimensionality reduction and machine learning algorithm application; a quality monitoring module for automatic judgment, early warning and problem visualization of quality defects; and a continuous optimization module for model updating, quality tracing and responsibility determination. The technical solution provided by the embodiment of the present invention may include the following beneficial effects:

[0006] The present invention discloses an edge node-based traceability monitoring method for agricultural product processing quality. The method constructs a unified data model and performs spatiotemporal alignment to form a spatiotemporal data cube based on the differences in multi-source heterogeneous sensor data. On this basis, the hierarchical analysis and fuzzy comprehensive evaluation methods are used to evaluate the processing quality of agricultural products. Through feature analysis and machine learning algorithms, data mining and pattern recognition are realized in a resource-constrained environment. Combined with the rule engine, quality defects are automatically judged and early warnings are triggered. Incremental learning is used to continuously optimize the evaluation model, and the results are recorded in the blockchain to form a traceable quality chain. The present invention realizes intelligent quality monitoring and evaluation of the entire process of agricultural product processing, improves the accuracy and traceability of the evaluation, and effectively ensures the quality and safety of agricultural products. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 The present invention is a flow chart of the agricultural product processing quality traceability monitoring method based on edge nodes.

[0008] Figure 2 It is a schematic diagram of the agricultural product processing quality traceability monitoring method and system based on edge nodes of the present invention.

[0009] Figure 3 It is another schematic diagram of the agricultural product processing quality traceability monitoring method and system based on edge nodes of the present invention.

[0010] Figure 4 It is a structural schematic diagram of an agricultural product processing quality traceability monitoring system based on edge nodes of the present invention. DETAILED DESCRIPTION

[0011] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0012] like Figure 1-4 In this embodiment, an agricultural product processing quality traceability monitoring system based on edge nodes may specifically include:

[0013] Step S101, acquiring multi-source heterogeneous sensor data, and pre-establishing a unified data model for differences in format, sampling frequency, and quantity of the multi-source heterogeneous sensor data; mapping the multi-source heterogeneous sensor data to the unified data model, and using data interpolation and downsampling methods to time-align the multi-source heterogeneous sensor data, eliminate sampling frequency differences, and obtain a time-synchronized multi-source data set.

[0014] The raw data of multi-source heterogeneous sensors are obtained, and according to a pre-established unified data model, the raw data is mapped to the model to obtain standardized multi-source heterogeneous sensor data; for the standardized multi-source heterogeneous sensor data, the timestamp information of each data source is extracted, and the data segments that need to be aligned on the time axis are determined by comparing the timestamps of different data sources; for the data segments that need to be aligned, the linear interpolation method is used to interpolate the data sources with fewer timestamps on the time axis to align them with the data sources with more timestamps; after interpolation, according to the preset sampling frequency threshold, it is determined whether downsampling processing is required; if necessary, the average downsampling method is used to downsample the interpolated data to make its sampling frequency consistent with the threshold; the multi-source heterogeneous sensor data after interpolation and downsampling are processed according to The unified time axis is used to sort the multi-source data set to obtain a time-synchronized multi-source data set; based on the time-synchronized multi-source data set, the wavelet transform method is used to extract the time-frequency domain features of each data source, and the feature vector of the multi-source heterogeneous sensor data is constructed; feature selection is performed on the extracted feature vector to remove redundant and irrelevant features and reduce the feature dimension; then the data set is divided into a training set and a test set using the K-fold cross-validation method; the processed feature vector is input into a pre-trained support vector machine model, and the model hyperparameters are optimized through grid search to achieve classification and prediction of multi-source heterogeneous sensor data; the test set is used to evaluate the performance of the model, and the accuracy, precision, recall and F1 value evaluation indicators are calculated; if the model performance meets the requirements, it is applied to the actual multi-source heterogeneous sensor data analysis and prediction tasks.

[0015] Exemplarily, the raw data of multi-source heterogeneous sensors are first obtained, such as the temperature sensor collects data once per second, the humidity sensor collects data once every 10 seconds, and the light sensor collects data once every minute. According to the pre-established unified data model, the raw data is mapped to the model, the temperature data is mapped to a floating point number in the range of (-50, 100), the humidity data is mapped to an integer in the range of (0, 100), and the light data is mapped to an integer in the range of (0, 10000), so as to obtain standardized multi-source heterogeneous sensor data. For the standardized multi-source heterogeneous sensor data, the timestamp information of each data source is extracted, and the data segments that need to be aligned on the time axis are determined by comparing the timestamps of different data sources, such as the data segments of temperature data and humidity data between 09:00:00 and 09:10:00 need to be aligned. For the data segments that need to be aligned, the linear interpolation method is used to interpolate the temperature data based on the timestamp of the humidity data so that it is aligned with the humidity data on the time axis. The sampling frequency of the interpolated temperature data is once every 10 seconds. According to the preset sampling frequency threshold of once per minute, it is judged that the interpolated temperature data and humidity data need to be downsampled. The average downsampling method is used to reduce the sampling frequency of temperature data and humidity data to once per minute, so that it is consistent with the sampling frequency of light data. The temperature data, humidity data and light data after interpolation and downsampling are sorted according to a unified time axis to obtain a time-synchronized multi-source data set. According to the time-synchronized multi-source data set, the time-frequency domain features of each data source are extracted using wavelet transform, such as the mean of the wavelet coefficient of temperature data is 35, the variance is 0.8, the mean of the wavelet coefficient of humidity data is 42, the variance is 12, and the mean of the wavelet coefficient of light data is 56, the variance is 21, and the feature vector of multi-source heterogeneous sensor data is constructed. The maximum information coefficient method is used to select the extracted feature vectors, remove redundant and irrelevant features, select the 5 most representative features, and reduce the feature dimension. Then, the data set is divided into a training set and a test set by a 5-fold cross validation method, with the training set accounting for 80% and the test set accounting for 20%. The processed feature vector is input into the pre-trained support vector machine (SVM) model, and the model hyperparameters are optimized through grid search. The penalty coefficient C ranges from [1, 1, 10], the kernel function selects the radial basis function (RBF), and the gamma parameter of the RBF kernel function ranges from [0.1, 1, 1] to achieve classification and prediction of multi-source heterogeneous sensor data. The performance of the model is evaluated using the test set, and the calculation accuracy is 92%, the precision is 97%, the recall rate is 94%, and the F1 value is 91%. The model performance meets the requirements and is applied to actual multi-source heterogeneous sensor data analysis and prediction tasks, such as predicting the environmental comfort level within the next hour based on temperature, humidity, and light data.

[0016] Step S102, based on the time-synchronized multi-source data set, construct a spatiotemporal data cube model, and map the multi-dimensional data including images, temperature and humidity into a unified spatiotemporal coordinate system; on the basis of the spatiotemporal data cube model, through data interpolation and aggregation methods, the coarse-grained data is refined and the fine-grained data is summarized, so that the spatiotemporal granularity of data from different sources tends to be consistent.

[0017] Based on the time-synchronized multi-source heterogeneous data, image and temperature and humidity multidimensional data are obtained, and a spatiotemporal data cube model is constructed; for the spatiotemporal data cube model, the multidimensional data including image data and temperature and humidity data are mapped to a unified spatiotemporal coordinate system; the mapped multidimensional spatiotemporal data is obtained, and according to a preset data granularity threshold, it is determined whether the data granularity meets the requirements.

[0018] Among them, the preset data granularity threshold is set according to the specific application scenario and requirements. For example, in weather forecasting, the granularity threshold of temperature and humidity data can be set to once per hour. The method for judging whether the data granularity meets the requirements is to compare the time interval of the data with the preset granularity threshold. If the time interval is greater than the threshold, it is considered to be coarse-grained data, otherwise it is considered to be fine-grained data. If the data granularity is coarse-grained data, the coarse-grained data is refined by linear interpolation method to obtain fine-grained data. For example, the interp function of the NumPy library can be used for linear interpolation. If the data granularity is fine-grained data, the fine-grained data is summarized by data aggregation method to obtain coarse-grained data. For example, the resample function of the Pandas library can be used for data aggregation and downsampling of fine-grained data. According to the processed spatiotemporal data, the data granularity in the spatiotemporal data cube model is updated to make the spatiotemporal granularity of data from different sources consistent. The method for updating the data granularity is to remap the processed data into the spatiotemporal data cube model to replace the original data. Based on the consistent spatiotemporal granularity data, subsequent data analysis and application can be performed. For example, the Matplotlib library can be used to visualize spatiotemporal data and generate line graphs, scatter graphs, and other charts; the Scikit-learn library can be used for spatiotemporal data mining, such as cluster analysis and anomaly detection.

[0019] Exemplarily, first obtain multi-dimensional heterogeneous data such as image, temperature and humidity, such as image data collected once every minute, temperature data collected once every 10 minutes, and humidity data collected once every 30 minutes. According to the pre-built spatiotemporal data cube model, the image data is mapped to the three-dimensional space coordinate system, and the temperature and humidity data are mapped to the time axis to form a unified spatiotemporal coordinate system. For the mapped multi-dimensional spatiotemporal data, the preset data granularity threshold is set to once every 5 minutes. By comparing the time interval of the data with the threshold, it is judged that the image data is fine-grained data, and the temperature data and humidity data are coarse-grained data. For temperature data and humidity data, a linear interpolation method is used to interpolate the data at a time interval of 5 minutes to keep it consistent with the granularity of the image data. For example, for temperature data, two data points are inserted between the 10th minute and the 20th minute, and the temperature value is the average of the temperature values ​​at the 10th minute and the 20th minute. For image data, a data aggregation method is used to downsample the data at a time interval of 30 minutes to keep it consistent with the granularity of the humidity data. For example, the image data from the 0th minute to the 29th minute is aggregated into one data point, and the image features are taken as the average of 30 images. The processed image data, temperature data, and humidity data are remapped into the spatiotemporal data cube model, and the data granularity of the model is updated to reach a spatiotemporal granularity of once every 5 minutes. Based on the consistent spatiotemporal granularity data, the K-means clustering algorithm is used to perform cluster analysis on the data. The number of clusters is set to 3 and the number of iterations is set to 10 times. Three cluster centers are obtained, representing different environmental states. The clustering results are visualized and a scatter plot is generated. Points of different colors represent different clusters, which intuitively show the spatiotemporal distribution and aggregation characteristics of multi-source heterogeneous data, providing data support for subsequent environmental monitoring and early warning.

[0020] Step S103, using the time-space aligned multi-source data, constructing a hierarchical analysis method model, decomposing the agricultural product processing quality evaluation index into several sub-indicators to form an index hierarchy structure; using a fuzzy comprehensive evaluation method, comprehensively considering the importance weight and satisfaction of each sub-indicator, calculating the comprehensive score of each scheme, and judging the agricultural product processing quality.

[0021] Obtain the data of various parameters in the agricultural product processing process, including raw material quality, processing temperature, processing time and equipment parameters, and establish an agricultural product processing database; for the agricultural product processing quality evaluation indicators, use the hierarchical analysis method to decompose the quality evaluation indicators into multiple sub-indicators including sensory indicators, physical and chemical indicators and safety indicators, and construct a hierarchical structure model of agricultural product processing quality evaluation indicators; select industry experts, use the Delphi method, conduct two rounds of scoring on the importance of each sub-indicator, use the 1-9 scaling method, take the average of the scores of each expert, and determine the weight vector of each sub-indicator. The weight coefficient ranges from 0 to 1, and the sum of the weight coefficients is 1; for each sub-indicator, combine the national and industry standards, construct a membership function, map the measured value of each sub-indicator to the interval [0, 1], and obtain the membership value of each sub-indicator; according to the weight vector and membership value of each sub-indicator, use the weighted average method to calculate the comprehensive evaluation score of agricultural product processing quality; the formula is: comprehensive evaluation score = ∑ (sub-indicator weight × sub-indicator membership value); the comprehensive evaluation score of agricultural product processing quality is compared with the preset compliance score. If the comprehensive evaluation score is greater than or equal to the qualified threshold, the processing quality of the batch of agricultural products is judged to be qualified; otherwise, it is judged to be unqualified and the batch is marked as a quality defective batch; for the quality defective batch, the processing parameter data subset corresponding to the batch is extracted, and the Apriori association rule algorithm is used to mine the association rules between quality defects and processing parameters, and identify the key processing parameters that cause quality defects; according to the results of the quality defect cause analysis, the orthogonal experimental design method is adopted, and the key processing parameters are used as experimental factors to optimize the agricultural product processing process parameters and formulate a quality improvement plan; in the next batch of agricultural product processing production, the optimized agricultural product processing process parameters are set to the control program of the processing equipment to guide the actual production, and the agricultural product processing database is continuously collected and updated; by real-time monitoring of agricultural product processing process parameters, the real-time data is input into the established quality assessment model, and the processing quality is predicted online. When the predicted quality is lower than the preset threshold, timely warning and adjustment of processing parameters are given to achieve online control and continuous improvement of agricultural product processing quality.

[0022] For example, in the agricultural product processing process, raw material quality parameters (such as raw material moisture content, pH value, etc.), processing temperature, processing time and equipment parameters (such as equipment speed, power, etc.) are collected in real time through online sensors. Data is collected every 5 minutes and the collected data is stored in the MySQL database. For the agricultural product processing quality assessment, a three-level indicator system including sensory indicators (color, smell, etc.), physical and chemical indicators (water content, protein content, etc.) and safety indicators (heavy metal content, pesticide residue content, etc.) is constructed through the hierarchical analysis method. Five industry experts were invited to use the 1-9 scale method to score each indicator twice. After taking the average, the AHP hierarchical analysis software was used to calculate the weight of each indicator. The weight of the sensory indicator is 2, the weight of the physical and chemical indicator is 5, and the weight of the safety indicator is 3. For each sub-indicator, the membership function was constructed with reference to the national standard GB / T 1234-2022. For example, the membership function of moisture content is: when the moisture content is ≤10%, the membership is 1; when the moisture content is ≥20%, the membership is 0; when the moisture content is between 10% and 20%, the membership decreases linearly. According to the weighted average formula, the comprehensive score of agricultural product processing quality is calculated. If the comprehensive score is ≥85, the quality of the batch of products is judged to be qualified, otherwise it is judged to be unqualified. The Apriori algorithm was used to perform association analysis on the unqualified batch data, and the support threshold was set to 0.5 and the confidence threshold was set to 8. It was found that the processing temperature was too high and the processing time was too long. The main causes of quality defects were found. The orthogonal experimental design method was adopted, with processing temperature and processing time as factors, and each factor was set to 3 levels. 9 groups of experiments were designed by L9 (34) orthogonal table to optimize the processing parameters, and the optimal processing temperature was determined to be 120℃ and the processing time was 30 minutes. The optimization parameters are set in the PLC control program of the processing equipment, and the processing data is collected and analyzed in real time. When the comprehensive score is lower than 85, the system automatically alarms and adjusts the processing parameters to achieve online quality control.

[0023] Step S104, based on the agricultural product processing quality evaluation results, feature analysis is performed on the agricultural product processing quality assessment data to screen out key features with greater influence and reduce data dimensions; decision tree and support vector machine machine learning algorithms are used to reduce computing resource consumption while ensuring analysis accuracy, thereby realizing data mining and pattern recognition in a resource-constrained environment.

[0024] The evaluation data is obtained based on the evaluation results of agricultural product processing quality, and the data is stored in the relational database MySQL; the evaluation data in MySQL is preprocessed, the data is read using Python's Pandas library, and the noise data and outliers are removed using Pandas' data cleaning function to obtain the cleaned data; the Relief feature selection algorithm is used to analyze the feature importance of the preprocessed data, and the Python Scikit-learn library is used to calculate the importance weight of each feature, and the top 10 key features with the highest weight are selected; based on the selected key features, the Pandas data selection function is used to select the corresponding feature columns from the original data set to construct a new low-dimensional data set; if the number of data set samples is greater than 1000, the decision tree algorithm is used to establish the quality assessment model; if the number of data set samples is less than or equal to 1000, the support vector machine algorithm is used to establish the quality assessment model; the Scikit-learn library is used to complete model training and testing; the established quality assessment model is encapsulated into a RESTAPI, the production parameters collected by the sensors of the agricultural product processing equipment are received in real time, the processing quality is evaluated in real time, and the evaluation results are returned to the production control system; according to the quality evaluation results, the production control system dynamically adjusts the key process parameters that affect the processing quality, and realizes automatic adjustment of the parameters through the PID control algorithm to ensure that the agricultural product processing quality is stable within the target range.

[0025] Exemplarily, in the process of agricultural product processing quality assessment, the assessment data is first stored in a MySQL database. Then, the data in MySQL is read using Python's Pandas library, and the dropna() function of Pandas is used to remove missing values, and the quantile() function is used to calculate the upper and lower quartiles. The data that exceeds the upper and lower quartiles by 5 times is regarded as an outlier and removed to obtain a cleaned data set. Then, the Relief feature selection algorithm is used, and the ReliefF class of the Scikit-learn library is used to calculate the importance weight of each feature, such as color, smell, moisture content, and other 30 features, and the top 10 key features with the highest weight are selected, such as color 85, smell 76, moisture content 69, etc. Then, the loc[] function of Pandas is used to select these 10 feature columns from the original data set to construct a new low-dimensional data set. Since the number of data set samples is 2000, which is greater than 1000, the CART decision tree algorithm is used to establish a quality assessment model. Using the DecisionTreeClassifier class of the Scikit-learn library, the maximum depth of the decision tree is set to 5, the minimum number of leaf node samples is set to 10, and a 10-fold cross validation is performed on the data set, and the model accuracy is 92%. Finally, the decision tree model is encapsulated into a RESTAPI, the interface code is written using the Flask framework, and deployed on the cloud server. The agricultural product processing equipment collects production parameters every 5 minutes, transmits the parameters to the RESTAPI through an HTTP POST request, and the API calls the decision tree model to perform real-time quality evaluation on the parameters, and returns the evaluation results Good or Bad to the production control system. The control system dynamically adjusts key process parameters such as processing temperature and pressure according to the evaluation results. Using the PID control algorithm, the temperature target value is set to 120℃, the pressure target value is set to 5MPa, and the PID parameters Kp=5, Ki=8, Kd=2, to achieve automatic adjustment of parameters, ensure that the quality of agricultural product processing is stable within the target range, and improve the product qualification rate.

[0026] Step S105, based on the analysis results of the machine learning algorithm, a rule engine is used to compare the quality assessment indicators with the preset thresholds to automatically determine whether there are quality defects in the agricultural product processing process; if there are quality defects in the agricultural product processing process, an early warning mechanism is triggered, and the problem links and causes are displayed through data visualization, to assist in the rapid location and handling of quality problems.

[0027] A1. Obtain the quality assessment index data in the agricultural product processing process, and use the support vector machine or random forest algorithm to analyze each index to obtain the quality assessment result; A2. According to the pre-established quality assessment rules, compare the quality assessment result with the preset threshold to determine whether there are quality defects in the agricultural product processing process; A3. If the judgment result shows that there are quality defects, the early warning mechanism is triggered, and the processing links with quality problems and the causes of quality defects are displayed in the form of charts and reports through data visualization technology; A4. For the displayed quality problem links and causes, the Apriori algorithm is used to analyze the historical quality problem data to mine the key factors causing the quality defects; A5. According to the mined key factors, the corresponding disposal measures are matched from the pre-built quality problem disposal knowledge base based on case reasoning, and the disposal measures are presented through the data visualization interface; A6. After executing the quality problem disposal measures, the quality assessment index data in the agricultural product processing process is continuously obtained. If the quality improvement effect does not meet expectations, the early warning mechanism is triggered again according to the latest quality data, and steps A3-A5 are repeated until the agricultural product processing quality fully meets the preset standards.

[0028] For example, in the process of agricultural product processing, 10 quality assessment index data such as temperature, humidity, and pH value are collected in real time through online sensors every 5 minutes. For each indicator, Python's Scikit-learn library is used to analyze the support vector machine algorithm, set the RBF kernel function, the penalty coefficient C=10, gamma=1, and perform 10-fold cross validation on 2000 sample data to obtain a quality assessment model with an average accuracy of 95%. The collected indicator data is input into the evaluation model to obtain the quality assessment results, and compared with the preset thresholds, such as the temperature threshold of 20℃~30℃, the humidity threshold of 60%~80%, and the pH threshold of 5~5. If it exceeds the threshold range, it is judged as a quality defect. When a quality defect occurs, the early warning mechanism is triggered, and the Matplotlib library is used to draw a line chart to show the changing trend of key indicators such as temperature and humidity. The Seaborn library is used to draw a heat map to analyze the correlation between the indicators and locate the key factors leading to quality defects, such as excessive temperature causing water loss in agricultural products. Then, the Apriori association rule mining algorithm was used to analyze 5,000 historical quality problem data, setting the minimum support to 0.5 and the minimum confidence to 8, and obtaining association rules, such as "temperature is too high, humidity is too low → agricultural products are shriveled", etc., to extract key quality problem factors. According to the key factors, the corresponding disposal measures are matched from the quality problem disposal knowledge base based on case reasoning, such as "temperature is too high → reduce equipment power, increase ventilation and cooling", etc. The PyQt library is used to develop a graphical interface to present the disposal measures in the form of a list. After the disposal measures are implemented, quality data is collected every 5 minutes, and the same support vector machine algorithm as before is used for evaluation. The Plotly library is used to draw a dynamic line chart of the quality score over time to show the progress of quality improvement. If the quality score continues to be lower than 90 points, the early warning mechanism will be triggered again, and the above quality problem analysis, positioning and disposal process will be repeated. Through closed-loop feedback control, the quality of agricultural product processing will be continuously optimized, and finally the quality will be stable and meet the standards.

[0029] Step S106, after the quality problem is handled, the key control points and influencing factors in the agricultural product processing process are analyzed through data mining; the incremental learning algorithm is used to dynamically update the quality assessment model and rule base, and the model parameters are updated by continuously receiving new data samples to achieve continuous optimization of the model and continuously improve the accuracy of agricultural product processing quality assessment.

[0030] Based on the data collected during the processing of agricultural products, the data is cleaned, denoised and standardized, and the principal component analysis feature extraction method is used to extract key feature parameters; the decision tree algorithm is used to build an agricultural product processing quality assessment model, and a quality assessment rule base is established through training on historical data; new data samples are continuously collected during the processing of agricultural products, and the new data are preprocessed and feature extracted, and the extracted feature parameters are input into the quality assessment model for quality prediction; the model prediction results are compared with the actual quality measurement values, and the model prediction error is calculated; based on the calculated model prediction error, the random forest algorithm is used to optimize the quality assessment model, and the prediction accuracy of the model is improved by dynamically updating the node threshold and branch structure of the decision tree; in the model optimization process In the process, if the prediction error is greater than the preset threshold, the rule base update mechanism is triggered, and the quality assessment rules in the rule base are corrected and supplemented according to the newly collected data samples; the update and optimization process of the quality assessment model and the rule base is continuously iterated to enable it to adapt to the dynamic changes of raw materials and process factors in the agricultural product processing process, and improve the accuracy and real-time performance of quality assessment; the optimized quality assessment model and rule base are integrated into the agricultural product processing process control system, and the processing parameters of key control points are collected and monitored in real time; when the monitoring parameters deviate from the optimal parameter range given by the quality assessment model, the system automatically generates early warning information, and according to the preset control strategy, sends parameter adjustment instructions to the actuator, and automatically adjusts the process parameters of the processing equipment to ensure that the quality of agricultural product processing is stable and controllable.

[0031] For example, in the agricultural product processing process, the collected data is first cleaned to remove noise data such as missing values ​​and outliers, and the minimum-maximum standardization method is used to normalize the data. Then, principal component analysis is used to extract the key features of the data, the cumulative variance contribution rate threshold is set to 90%, and the first 5 principal components are selected as input features. The quality assessment model is constructed using the C5 decision tree algorithm, 3,000 historical data are trained, the optimal partition attribute is selected by the information gain ratio, and the pruning strategy is used to avoid overfitting, and a quality assessment rule base with an average accuracy of 93% is obtained. In the agricultural product processing process, data is collected every 10 minutes, and after preprocessing and feature extraction of the new data, it is input into the quality assessment model for prediction. The prediction results are compared with the actual quality measurement values, and the model prediction performance is evaluated using the mean square error (MSE). When the MSE is greater than 0.5, the model optimization mechanism is triggered. The random forest algorithm is used to integrate and optimize the decision tree model. The number of decision trees is set to 100. Multiple decision tree sub-models are constructed through self-service sampling and random selection of attributes. The prediction results of each sub-model are integrated through the voting mechanism to improve the accuracy of quality assessment. At the same time, when the prediction error is greater than 1 for 5 consecutive times, the rule base update mechanism is triggered. The decision tree model is retrained through the C5 algorithm using the latest 1000 data samples to update the quality assessment rules. The optimized quality assessment model and rule base are integrated into the agricultural product processing control system to collect and monitor the temperature, humidity, pressure and other parameters of the key control points in real time. When the monitoring parameters deviate from the optimal range (such as the temperature range of 20℃~30℃), the system automatically generates early warning information and sends parameter adjustment instructions to the PLC controller according to the fuzzy control strategy. By adjusting the valve opening, speed and other operations, the processing parameters are maintained within the optimal range to ensure the stable and controllable quality of agricultural product processing.

[0032] Step S107, the quality assessment results, optimization schemes and processing procedures are recorded in the blockchain to form an unalterable quality traceability chain; cryptographic methods are used to encrypt and protect sensitive data to ensure data security and privacy; through the smart contract mechanism, the responsibility for agricultural product quality is automatically determined and traced, so that the quality of agricultural products can be fully traceable, verifiable and accountable, thereby enhancing consumer trust.

[0033] Acquire agricultural product quality assessment result data, record it on the chain through blockchain technology, and use the tamper-proof nature of blockchain to build an agricultural product quality traceability chain; use the RSA asymmetric encryption algorithm to encrypt sensitive information in agricultural product quality data to ensure the security of data transmission and storage in the blockchain network and protect data privacy from being leaked; based on blockchain smart contracts, write agricultural product quality responsibility determination rules into the contract code.

[0034] For example, if the product quality index exceeds the range specified by the national standard, it is determined to be the producer's responsibility; if the temperature and humidity during transportation exceed the specified range and cause quality problems, it is determined to be the logistics responsibility, etc. The traceability rules are written into the contract code, such as associating the production link data with the product batch number, and the logistics link data with the waybill number, so as to realize the automatic judgment and traceability of agricultural product quality problems. When a quality dispute occurs, the responsible party is automatically verified by calling the contract. The data of each link of agricultural products from planting, harvesting, processing, packaging, transportation to sales are collected and recorded on the blockchain quality traceability chain through IoT devices, including origin information, processing information, logistics information, sales information, etc., so that the quality of agricultural products can be traced, verified and accountable throughout the process. By scanning the QR code on the packaging of agricultural products, consumers can view the quality information of agricultural products in various links such as production, processing, and logistics. Through blockchain technology, a trust mechanism for agricultural product quality is established, and information such as quality assessment results is made public and transparent, subject to supervision from all sectors of society, to enhance consumers' trust in the quality of agricultural products and promote the improvement of agricultural product brand value. The alliance chain model is adopted to build a blockchain network for agricultural product quality, with the agricultural authorities as regulators, agricultural product production enterprises, distributors, logistics companies, etc. as data providers, and third-party quality inspection agencies as data auditors. The regulator is responsible for formulating agricultural product quality standards and management systems, the data provider is responsible for uploading quality data at all links, and the data auditor is responsible for verifying the authenticity of the data. All parties access quality data according to their permissions, which not only ensures information sharing but also protects the privacy and security of sensitive data. By deploying random forest and support vector machine algorithms on blockchain network nodes, agricultural product quality data can be intelligently analyzed, quality problems can be warned, decision support can be provided for agricultural product production optimization and quality improvement, and the intelligent level of agricultural product quality management can be improved. Based on historical quality data, the random forest model is trained to predict the quality trend of agricultural products; through the support vector machine algorithm, the key factors affecting the quality of agricultural products are analyzed, the weak links in quality control are found, and optimization suggestions are provided for agricultural product production.

[0035] For example, in the construction of the agricultural product quality traceability chain, the RSA asymmetric encryption algorithm is first used to encrypt the agricultural product quality data. A 2048-bit key pair is used, the public key is used for data encryption, and the private key is held by the data owner for data decryption. The encrypted data generates a digital summary through the SHA-256 algorithm, which is packaged and uploaded to the blockchain network together with the original data. The blockchain node verifies the digital summary to ensure that the data has not been tampered with during transmission and storage. The smart contract sets the rules for determining the responsibility of agricultural product quality, such as the temperature exceeding 0-10℃ is determined as logistics responsibility, and the pesticide residue exceeding 0.1mg / kg is determined as producer responsibility. In terms of traceability rules, the production link data is associated with a unique product batch number such as "P20220101001", and the logistics link data is associated with the waybill number such as "T20220101001". The data of each link of agricultural products is collected and uploaded to the blockchain through IoT devices, such as temperature and humidity sensors that collect data every 10 minutes and upload them to the blockchain. Consumers can scan the QR code of the product, such as "Q20220101001", to view information such as the product's production date, origin, and logistics temperature and humidity change curve. The blockchain network adopts the PBFT consensus algorithm and consists of 5 nodes. The agricultural department, agricultural enterprises, logistics, distributors, and third-party quality inspection agencies each control a node to ensure the credibility of the data on the chain. Random forest and SVM algorithm models are deployed in the nodes and trained based on historical quality data. When the random forest model predicts that the probability of excessive pesticide residues is greater than 5%, an early warning prompt is triggered; the SVM model is used to analyze the key factors affecting pesticide residues, and it is found that the amount of pesticide used is the primary factor, providing farmers with scientific drug use recommendations. Through the application of blockchain technology, a credible and traceable agricultural product quality and safety system has been established, providing strong guarantees for the improvement of agricultural product quality and brand value.

[0036] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations. In addition, the various different embodiments of the present invention can also be combined arbitrarily, as long as they do not violate the concept of the present invention, they should also be regarded as the contents disclosed by the present invention.

Claims

1. The agricultural product processing quality traceability monitoring method based on edge nodes is characterized by: The method comprises: Acquire multi-source heterogeneous sensor data, establish a unified data model in advance according to the differences in format, sampling frequency and quantity of the multi-source heterogeneous sensor data, map the multi-source heterogeneous sensor data to the unified data model, use data interpolation and downsampling methods to align the time axis of the multi-source heterogeneous sensor data, eliminate sampling frequency differences, and obtain a time-synchronized multi-source data set; Based on the time-synchronized multi-source data set, a spatiotemporal data cube model is constructed to map the multi-dimensional data including images, temperature and humidity into a unified spatiotemporal coordinate system. On the basis of the spatiotemporal data cube model, the coarse-grained data is refined and the fine-grained data is summarized through data interpolation and aggregation methods, so that the spatiotemporal granularity of data from different sources tends to be consistent; By using time-space aligned multi-source data, a hierarchical analysis method model was constructed to decompose the agricultural product processing quality assessment indicators into several sub-indicators to form an indicator hierarchy. The fuzzy comprehensive evaluation method was used to comprehensively consider the importance weight and satisfaction of each sub-indicator, calculate the comprehensive score of each scheme, and judge the agricultural product processing quality. According to the results of agricultural product processing quality evaluation, feature analysis is performed on agricultural product processing quality assessment data to screen out key features, reduce data dimensions, and use decision tree and support vector machine machine learning algorithms to reduce computing resource consumption while ensuring analysis accuracy, thus achieving data mining and pattern recognition in resource-constrained environments; Based on the analysis results of the machine learning algorithm, the rule engine is used to compare the quality assessment indicators with the preset thresholds to automatically determine whether there are quality defects in the agricultural product processing process. If there are quality defects in the agricultural product processing process, the early warning mechanism is triggered, and the problem links and causes are displayed through data visualization, assisting in the rapid location and disposal of quality problems; After the quality problems are dealt with, the key control points and influencing factors in the agricultural product processing process are analyzed through data mining, and the quality assessment model and rule base are dynamically updated by using incremental learning algorithms. By continuously receiving new data samples and updating model parameters, the model is continuously optimized, and the accuracy of agricultural product processing quality assessment is continuously improved; The quality assessment results, optimization plans and processing processes are recorded in the blockchain to form an unalterable quality traceability chain. Cryptography methods are used to encrypt and protect sensitive data to ensure data security and privacy. Through the smart contract mechanism, the quality responsibility of agricultural products can be automatically determined and traced.

2. The method according to claim 1, characterized in that Acquire multi-source heterogeneous sensor data, establish a unified data model in advance according to the differences in format, sampling frequency and quantity of the multi-source heterogeneous sensor data; map the multi-source heterogeneous sensor data to the unified data model, use data interpolation and downsampling methods to align the time axis of the multi-source heterogeneous sensor data, eliminate sampling frequency differences, and obtain a time-synchronized multi-source data set, including: Obtain the raw data of multi-source heterogeneous sensors, and map the raw data to the pre-established unified data model to obtain standardized multi-source heterogeneous sensor data; For the standardized multi-source heterogeneous sensor data, extract the timestamp information of each data source, and determine the data segments that need to be time-aligned by comparing the timestamps of different data sources; For the data segments that need to be aligned, a linear interpolation method is used to interpolate the data source with fewer timestamps on the time axis to align it with the data source with more timestamps; After interpolation, determine whether downsampling is required based on the preset sampling frequency threshold; If necessary, the interpolated data is downsampled using the average downsampling method so that its sampling frequency is consistent with the sampling frequency threshold; The multi-source heterogeneous sensor data after interpolation and downsampling are sorted according to a unified time axis to obtain a time-synchronized multi-source data set; Based on the time-synchronized multi-source data sets, the wavelet transform method is used to extract the time-frequency domain features of each data source and construct the feature vector of multi-source heterogeneous sensor data; Perform feature selection on the extracted feature vectors to remove redundant and irrelevant features and reduce feature dimensions; Then the K-fold cross validation method is used to divide the data set into training set and test set; The processed feature vectors are input into the pre-trained support vector machine model, and the model hyperparameters are optimized through grid search to achieve classification and prediction of multi-source heterogeneous sensor data; Use the test set to evaluate the performance of the support vector machine model and calculate the accuracy, precision, recall and F1 value evaluation indicators; If the performance of the support vector machine model meets the requirements, it will be applied to actual multi-source heterogeneous sensor data analysis and prediction tasks.

3. The method according to claim 1, characterized in that Based on the time-synchronized multi-source data set, a spatiotemporal data cube model is constructed to map the multi-dimensional data including images, temperature and humidity into a unified spatiotemporal coordinate system; based on the spatiotemporal data cube model, the coarse-grained data is refined and the fine-grained data is summarized through data interpolation and aggregation methods, so that the spatiotemporal granularity of data from different sources tends to be consistent, including: Based on time-synchronized multi-source heterogeneous data, we obtain image and temperature and humidity multi-dimensional data and build a spatiotemporal data cube model; For the spatiotemporal data cube model, multidimensional data including image data and temperature and humidity data are mapped into a unified spatiotemporal coordinate system; Obtain the mapped multidimensional spatiotemporal data and determine whether the data granularity meets the requirements based on the preset data granularity threshold.

4. The method according to claim 1, characterized in that: By using the time-space aligned multi-source data, a hierarchical analysis method model is constructed to decompose the agricultural product processing quality evaluation index into several sub-indicators to form an index hierarchy structure; the fuzzy comprehensive evaluation method is adopted to comprehensively consider the importance weight and satisfaction of each sub-indicator, calculate the comprehensive score of each scheme, and judge the agricultural product processing quality, including: Obtain data on various parameters during agricultural product processing, including raw material quality, processing temperature, processing time and equipment parameters, and establish an agricultural product processing database; Aiming at the agricultural product processing quality assessment indicators, the hierarchical analysis method is used to decompose the quality assessment indicators into multiple sub-indicators including sensory indicators, physical and chemical indicators and safety indicators, and to construct a hierarchical structure model of agricultural product processing quality assessment indicators; Select experts in the industry and use the Delphi method to conduct two rounds of scoring on the importance of each sub-indicator. Use the 1-9 scale method, take the average of the scores of each expert, and determine the weight vector of each sub-indicator. The weight coefficient ranges from 0 to 1, and the sum of the weight coefficients is 1; For each sub-indicator, in combination with national and industry standards, a membership function is constructed to map the measured value of each sub-indicator to the interval [0, 1] to obtain the membership value of each sub-indicator; According to the weight vector and membership value of each sub-indicator, the weighted average method is used to calculate the comprehensive evaluation score of agricultural product processing quality; The formula is: comprehensive evaluation score = ∑ (sub-indicator weight × sub-indicator membership value); Compare the comprehensive evaluation score of agricultural product processing quality with the preset qualified threshold value. If the comprehensive evaluation score is greater than or equal to the preset qualified threshold value, the processing quality of the batch of agricultural products is determined to be qualified. Otherwise, it is judged as unqualified and the batch is marked as a quality defective batch; For batches with quality defects, extract the processing parameter data subset corresponding to the batch, use the Apriori association rule algorithm to mine the association rules between quality defects and processing parameters, and identify the key processing parameters that cause quality defects; According to the results of the quality defect cause analysis, the orthogonal experimental design method is used to optimize the agricultural product processing parameters and formulate a quality improvement plan with key processing parameters as experimental factors; In the next batch of agricultural product processing production, the optimized agricultural product processing process parameters will be set into the control program of the processing equipment to guide the actual production, while continuously collecting and updating the agricultural product processing database; By real-time monitoring of agricultural product processing parameters, the real-time data is input into the established quality assessment model to predict the processing quality online. When the predicted quality is lower than the preset threshold, timely warning and adjustment of processing parameters are carried out to achieve online control and continuous improvement of agricultural product processing quality.

5. The method according to claim 1, characterized in that: According to the results of agricultural product processing quality evaluation, feature analysis is performed on agricultural product processing quality assessment data to screen out key features and reduce data dimensions; decision tree and support vector machine machine learning algorithms are used to reduce computing resource consumption while ensuring analysis accuracy, and to achieve data mining and pattern recognition in resource-constrained environments, including: Obtain assessment data based on the agricultural product processing quality assessment results and store the data in the relational database MySQL; Preprocess the evaluation data in MySQL, use Python's Pandas library to read the data, and use Pandas' data cleaning function to remove noise data and outliers to obtain cleaned data; The Relief feature selection algorithm is used to analyze the feature importance of the preprocessed data. The Python Scikit-learn library is used to calculate the importance weight of each feature and select the top 10 key features with the highest weight. Based on the selected key features, use Pandas’ data selection function to select the corresponding feature columns from the original data set to construct a new low-dimensional data set. If the number of data set samples is greater than 1000, the decision tree algorithm is used to establish the quality assessment model; If the number of data set samples is less than or equal to 1000, the support vector machine algorithm is used to establish the quality assessment model; Use the Scikit-learn library to complete model training and testing; The established quality assessment model is encapsulated into a REST API to receive production parameters collected by sensors of agricultural product processing equipment in real time, conduct real-time assessment of processing quality, and return the assessment results to the production control system; According to the quality assessment results, the production control system dynamically adjusts the key process parameters that affect the processing quality, and realizes automatic adjustment of the parameters through the PID control algorithm to ensure that the processing quality of agricultural products remains stable within the target range.

6. The method according to claim 1, characterized in that Based on the analysis results of the machine learning algorithm, the rule engine is used to compare the quality assessment indicators with the preset thresholds to automatically determine whether there are quality defects in the agricultural product processing process; if there are quality defects in the agricultural product processing process, the early warning mechanism is triggered, and the problem links and causes are displayed through data visualization to assist in the rapid location and disposal of quality problems, including: A1. Obtain the quality assessment index data during the agricultural product processing process, and use the support vector machine or random forest algorithm to analyze each index to obtain the quality assessment results; A2. According to the pre-established quality assessment rules, the quality assessment results are compared with the preset threshold value to determine whether there are quality defects in the agricultural product processing process; A3. If the judgment result shows that there are quality defects, the early warning mechanism will be triggered. Through data visualization technology, the processing links with quality problems and the causes of quality defects will be displayed in the form of charts and reports; A4. Based on the quality problem links and causes displayed, the Apriori algorithm is used to analyze historical quality problem data and dig out the key factors that lead to the quality defects; A5. According to the mined key factors, the corresponding disposal measures are matched from the pre-built quality problem disposal knowledge base based on case reasoning, and the disposal measures are presented through the data visualization interface; A6. After implementing quality problem handling measures, continue to obtain quality assessment indicator data during the agricultural product processing process. If the quality improvement effect does not meet expectations, trigger the early warning mechanism again based on the latest quality data and repeat steps A3-A5 until the agricultural product processing quality fully meets the preset standards.

7. The method according to claim 1, characterized in that After the quality issues are resolved, the key control points and influencing factors in the agricultural product processing process are analyzed through data mining; the incremental learning algorithm is used to dynamically update the quality assessment model and rule base, and the model parameters are updated by continuously receiving new data samples to achieve continuous optimization of the model and continuously improve the accuracy of agricultural product processing quality assessment, including: According to the data collected during the processing of agricultural products, the data are cleaned, denoised and standardized, and the principal component analysis feature extraction method is used to extract key feature parameters; The decision tree algorithm is used to build a quality assessment model for agricultural product processing, and a quality assessment rule base is established through training on historical data; Continuously collect new data samples during the agricultural product processing, pre-process and extract features from the new data, and input the extracted feature parameters into the quality assessment model for quality prediction; Compare the model prediction results with the actual quality measurements and calculate the model prediction error; According to the calculated model prediction error, the random forest algorithm is used to optimize the quality assessment model, and the prediction accuracy of the model is improved by dynamically updating the node threshold and branch structure of the decision tree; During the model optimization process, if the prediction error is greater than the preset threshold, the rule base update mechanism is triggered to revise and supplement the quality assessment rules in the rule base based on the newly collected data samples; Continuously iterate the updating and optimization process of the quality assessment model and rule base to enable it to adapt to the dynamic changes of raw materials and process factors in the processing of agricultural products, and improve the accuracy and real-time performance of quality assessment; Integrate the optimized quality assessment model and rule base into the agricultural product processing control system to collect and monitor the processing parameters of key control points in real time; When the monitoring parameters deviate from the optimal parameter range given by the quality assessment model, the system automatically generates early warning information and sends parameter adjustment instructions to the actuator according to the preset control strategy to automatically adjust the process parameters of the processing equipment to ensure that the quality of agricultural product processing is stable and controllable.

8. The method according to claim 1, characterized in that The quality assessment results, optimization schemes and processing processes are recorded in the blockchain to form an unalterable quality traceability chain; cryptographic methods are used to encrypt and protect sensitive data to ensure data security and privacy; through the smart contract mechanism, the responsibility for agricultural product quality is automatically determined and traced, so that the quality of agricultural products can be fully traceable, verifiable and accountable, and consumer trust can be enhanced, including: Obtain agricultural product quality assessment result data, record it on the blockchain through blockchain technology, and use the tamper-proof nature of blockchain to build an agricultural product quality traceability chain; For sensitive information in agricultural product quality data, the RSA asymmetric encryption algorithm is used for encryption processing to ensure the security of data transmission and storage in the blockchain network and protect data privacy from being leaked; Based on blockchain smart contracts, the rules for determining agricultural product quality responsibility are written into the contract code.

9. An agricultural product processing quality traceability monitoring system based on edge nodes, characterized in that: The system comprises: Data preprocessing module, used for format unification and time synchronization of multi-source heterogeneous sensor data; Spatiotemporal data modeling module, used to build spatiotemporal data cube models and align data granularity; Quality assessment module, used to build AHP model and conduct fuzzy comprehensive evaluation; Data mining module for feature analysis, dimensionality reduction, and application of machine learning algorithms; Quality monitoring module, used for automatic judgment, early warning and problem visualization of quality defects; Continuous optimization module for model updating, quality traceability and responsibility determination.

Images