Mancozeb residue intelligent detection and analysis method and system

Abstract

The invention discloses an intelligent detection and analysis method and system for mancozeb residues, and relates to the technical field of agricultural product detection.The method comprises the steps that data interaction with a detection unit is conducted, the number and batch data of products to be detected are read, sample distribution is determined, and random sample collection under distribution constraint is executed; after the collected samples are subjected to multi-dimensional residue detection, a multi-dimensional data set with environmental factor identifiers is established; extracting features by using a depth feature extraction channel, establishing a depth feature set, and performing global-local feature alignment; and inputting the alignment result and the environment factor into a decision layer to carry out high-dimensional feature fusion, and generating a residual detection decision through reverse verification. The technical problems that a traditional detection method in the prior art depends on a single detection dimension and is poor in detection precision, efficiency and environmental adaptability are solved, and the technical effects of improving the detection precision and the working efficiency and enhancing the environmental adaptability through multi-dimensional residual detection and high-dimensional feature fusion are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural product detection, and in particular to an intelligent detection and analysis method and system for mancozeb residue. Background Art

[0002] Mancozeb is a broad-spectrum fungicide commonly used in crop disease prevention and control, but its residue problem has received increasing attention in food safety. Existing methods for detecting mancozeb residues mostly rely on a single detection technology, such as spectroscopy, chromatography or electrochemical sensors, which can usually only detect residues within a specific range and have limitations in accuracy, efficiency and environmental adaptability. These traditional methods are not only time-consuming and inefficient, but also difficult to handle complex samples or are affected by environmental changes, resulting in poor stability and accuracy of the test results. Summary of the invention

[0003] The present application provides a method and system for intelligent detection and analysis of mancozeb residues, which are used to solve the technical problems in the prior art that traditional detection methods rely on a single detection dimension and have poor detection accuracy, efficiency and environmental adaptability.

[0004] In a first aspect of the present application, a method for intelligent detection and analysis of mancozeb residues is provided, the method comprising: after the mancozeb residue intelligent detection and analysis system interacts with a detection unit, a product data set of a product to be tested is read, the product data set comprising product quantity data and product batch data; after determining the sample distribution according to the product data set, random sample collection under distribution constraints is performed, and after the analysis samples are subjected to multidimensional residue detection, a multidimensional data set is established, the multidimensional residue detection comprises spectral detection, chromatographic detection, and electrochemical sensor detection, and the multidimensional data set carries an environmental factor identifier; feature extraction of the multidimensional data set is performed using a deep feature extraction channel to establish a deep feature set; after global-local feature alignment of the deep feature set, the alignment result and the environmental factor identifier are input into a decision layer, and after high-dimensional feature fusion is performed using the decision layer, a residue detection decision is generated after reverse verification is performed according to the distribution constraint.

[0005] According to a second aspect of the present application, a system for intelligent detection and analysis of mancozeb residues is provided, the system comprising: a product data interaction module, the product data interaction module is used to read a product data set of a product to be tested after the intelligent detection and analysis system of mancozeb residues interacts with a detection unit, the product data set comprising product quantity data and product batch data; a multidimensional residue detection module, the multidimensional residue detection module is used to determine the sample distribution according to the product data set, perform random sample collection under distribution constraints, establish a multidimensional dataset after the analysis sample passes through multidimensional residue detection, the multidimensional residue detection comprises spectral detection, chromatographic detection, and electrochemical sensor detection, the multidimensional dataset carries an environmental factor identifier; a feature extraction module, the feature extraction module is used to extract features of the multidimensional dataset using a deep feature extraction channel, and establish a deep feature set; a residue detection decision module, the residue detection decision module is used to perform global-local feature alignment on the deep feature set, input the alignment result and the environmental factor identifier into a decision layer, perform high-dimensional feature fusion using the decision layer, and generate a residue detection decision after reverse verification according to the distribution constraint.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] The present application provides an intelligent detection and analysis method and system for mancozeb residues, which relate to the technical field of agricultural product detection. Product data is read through data interaction, sample distribution is determined and random collection is performed. After the collected samples are subjected to multidimensional residue detection, a multidimensional data set with environmental factor identification is established. Features are extracted using a deep feature extraction channel, global-local feature alignment is performed, and the features are input into a decision layer for high-dimensional feature fusion, and finally a residue detection decision is generated. The technical problem that traditional detection methods in the prior art rely on a single detection dimension and have poor detection accuracy, efficiency and environmental adaptability is solved, and the technical effect of improving detection accuracy and work efficiency and enhancing environmental adaptability is achieved through multidimensional residue detection and high-dimensional feature fusion. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0009] Figure 1 A schematic diagram of a flow chart of an intelligent detection and analysis method for mancozeb residue provided in an embodiment of the present application;

[0010] Figure 2A schematic diagram of the structure of an intelligent detection and analysis system for mancozeb residues provided in an embodiment of the present application;

[0011] Figure 3 A curve spectrum of mancozeb in pepper in an example of an intelligent detection and analysis method for mancozeb residue provided in an embodiment of the present application;

[0012] Figure 4 This is a curve spectrum of mancozeb in tomatoes in an example of an intelligent detection and analysis method for mancozeb residues provided in an embodiment of the present application.

[0013] Explanation of the reference numerals: product data interaction module 11, multi-dimensional residue detection module 12, feature extraction module 13, residue detection decision module 14. DETAILED DESCRIPTION

[0014] The present application provides a method and system for intelligent detection and analysis of mancozeb residues, which are used to solve the technical problems in the prior art that traditional detection methods rely on a single detection dimension and have poor detection accuracy, efficiency and environmental adaptability.

[0015] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0016] It should be noted that the terms "first", "second", etc. in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices.

[0017] Embodiment 1, as Figure 1 As shown, the present application provides a method for intelligent detection and analysis of mancozeb residues, the method comprising:

[0018] P10: After the intelligent detection and analysis system for mancozeb residues interacts with the detection unit, a product data set of the product to be tested is read, wherein the product data set includes product quantity data and product batch data.

[0019] Specifically, first of all, this intelligent detection and analysis system for mancozeb residues needs to interact with the detection unit to introduce the relevant data of the product to be tested into the detection system. Data interaction can be achieved through Internet of Things (IoT) technology, including real-time data transmission between sensors and smart devices, and sending product information to the background system. In addition, seamless connection with external data sources is achieved through API interfaces or cloud platforms to ensure that the basic information of the product to be tested can be quickly and accurately transmitted to the detection unit.

[0020] When the data is successfully transmitted, the product data set of the product to be tested can be read. This data set is the basis of the entire testing process and contains the core information of the sample to be tested, including two important data: product quantity data and product batch data. Among them, the product quantity data reflects the number of products to be tested, which can reflect the total amount of samples and provide guidance for subsequent operations when the samples are collected. For example, when the number of products to be tested is large, a stratified sampling strategy can be adopted based on the quantity information in the data set to ensure that the collected samples are statistically representative, thereby avoiding deviations, ensuring the universality and reliability of the test results, avoiding errors caused by too few or too many samples, and ensuring the accuracy of subsequent tests.

[0021] At the same time, the product batch data records the production batch information of the sample, including batch number, production date, supplier, etc. Different batches of products may be affected by different environmental factors during production, storage and transportation, resulting in differences in the concentration of mancozeb residues. The record of product batch data can not only be used to distinguish between multiple batches of samples, but also provide background information for subsequent data analysis. For example, a batch of samples may have a high concentration of mancozeb residues due to poor storage conditions. Based on this batch information, potential risks can be identified, so that corresponding adjustments can be made during the testing and analysis process.

[0022] Combined with these data, when determining the distribution of samples to be tested, the quantity and batch data can be effectively used as a reference to make appropriate sample allocations, ensuring that the samples collected subsequently are broadly representative of the overall product range.

[0023] P20: After determining the sample distribution according to the product data set, perform random sample collection under distribution constraints, and establish a multidimensional data set after the analysis samples pass multidimensional residue detection. The multidimensional residue detection includes spectral detection, chromatographic detection, and electrochemical sensor detection. The multidimensional data set is marked with environmental factors.

[0024] Optionally, first determine the distribution of samples to be tested based on the product data set read in the previous step. By analyzing the product quantity data and product batch data, samples are reasonably allocated so that the collected samples are representative of the entire product range, thereby ensuring the breadth and accuracy of subsequent test results. At this time, distribution constraints refer to specific rules or conditions that must be followed during sample collection. These conditions can be quantity distribution, batch coverage, or other factors that affect the test results. For example, if a batch has a large number of samples, an appropriate number of samples can be selected for testing based on the proportion.

[0025] Next, after the sample distribution is determined, random sample collection is performed. The core of this process is to ensure the representativeness of the sample by randomly selecting a certain number of samples. Random sampling can eliminate human bias, avoid selective errors, and make the test results statistically reliable. These samples will undergo multidimensional residue testing, including spectral testing, chromatographic testing, and electrochemical sensing testing.

[0026] Among them, spectral detection can analyze the composition of the sample by relying on its absorption, reflection or emission characteristics of light of a specific wavelength, and can quickly identify the residual amount of mancozeb without damage. Chromatographic detection uses the different migration speeds of the sample in the stationary phase and the mobile phase to separate the various components, thereby accurately measuring the concentration of the target substance. Electrochemical sensing detection monitors the characteristic current changes of mancozeb through electrochemical reactions, providing highly sensitive detection data. These three detection methods have their own characteristics and can obtain information on the residual mancozeb in the sample from different angles.

[0027] Exemplarily, the sample pretreatment process can be, first, the first step is to prepare the sample, accurately measure 10g of the homogenized vegetable and fruit sample, put it into a 50mL centrifuge tube, add salt bag A and 5mL water, and vortex for 1 minute. The spiked solution should be added in this step to calibrate the concentration of the target substance. The second step is to add salt bag B to the sample, mix it evenly and let it stand for 15 minutes. This step can help remove interfering substances in the sample. Then add 10mL of extract, vortex for 10 minutes, and let it stand for 10 minutes. (Preparation method of the extract: dissolve 2.44mL of dimethyl sulfate in 50mL of acetonitrile to obtain an extract. This extract helps to extract mancozeb from the sample) The third step is to freeze and centrifuge, put the sample in a refrigerator and freeze it for 10 minutes to help the solid particles settle. Add salt bag C and ceramic homogenizer and oscillate for 2 minutes to homogenize the sample. Then centrifuge at 8000rpm for 5 minutes to separate the supernatant. The fourth step is purification. Take 6 mL of the supernatant and put it into a new centrifuge tube (centrifuge tube D), vortex and purify for 2 minutes, and centrifuge for 5 minutes. Take 1 mL of the purified solution, blow it to near dryness with nitrogen, and then add 1 mL of ethyl acetate for re-dissolution. Finally, pass it through a 0.22 μm filter membrane to ensure that the solution is clear and free of particles, and use it directly for injection. The fifth step is standard curve. During the preparation of the standard curve, the steps are exactly the same as the sample processing, the only difference is that no sample is added.

[0028] Furthermore, the gas chromatography-tandem mass spectrometer (GCMS / MS) can be used to qualitatively and quantitatively analyze the residues of mancozeb. At the same time, the Agilent HP-5MS chromatographic column has a good separation effect and can efficiently separate the target substances in the sample. Program column temperature: 40℃ (maintained for 3min), 50℃ / min250℃ (maintained for 3min); injection port temperature: 280℃; injection volume: 1ul; injection mode: splitless injection; ion source temperature 230℃; transmission line temperature: 280℃. Quantitative ion pair: 78 / 63 (CE=10); qualitative ion pair: 62.9 / 45.1 (CE=10).

[0029] Through the spike recovery experiment, the mancozeb curve spectrum in pepper and the mancozeb curve spectrum in tomato were finally generated, such as Figure 3 , Figure 4 In addition, the test results are shown in the following table. When the spiked concentration is 50 ng / g, the determination results of pepper and tomato are close to the expected recovery rate, with pepper recovery rates of 98%, 96% and 94%, and tomato recovery rates of 104%, 102% and 96%, indicating that this method has good accuracy and reliability.

[0030]

[0031] Through these three detection methods, a multidimensional data set is generated, which contains the results of each sample in each detection dimension. These data not only include the detected mancozeb residue concentration, but also carry environmental factor labels, that is, environmental factors that may affect the results during the detection process, such as temperature, humidity, air pressure, etc. The recording of environmental factors can help the system calibrate the test results under different environmental conditions and improve the accuracy and robustness of the detection.

[0032] Furthermore, the analysis sample is subjected to multi-dimensional residue detection, and step P20 of the embodiment of the present application further includes:

[0033] P21: Establish a self-supervisory database for residue detection; P22: Perform active error identification of equipment drift during multi-dimensional residue detection based on the self-supervisory database, and establish an active error correction mechanism; P23: Compensate for detection results through the active error correction mechanism to complete multi-dimensional residue detection.

[0034] It should be understood that in order to ensure the accuracy of detection and the stability of processing equipment, a self-supervised database for residue detection can be established. This database contains a large amount of test data, including test results under different environmental factors and equipment conditions, as well as the corresponding actual residue concentrations. Specifically, the self-supervised database uses historical test data to continuously optimize the detection model and mine potential patterns and laws in the data through unsupervised learning. The database can automatically identify trend changes in the data, such as differences in residues in different batches and under different environmental conditions, and continuously improve the accuracy of subsequent tests by learning this information. In this way, the system can not only adapt to the characteristics of different samples, but also quickly adjust the detection strategy when faced with new samples or environmental conditions.

[0035] However, in the actual detection process, the drift of the equipment often affects the stability and accuracy of the data, especially the performance of precision instruments such as spectrometers and chromatographs may degrade after long-term use. In order to meet this challenge, an active error identification mechanism can be introduced to identify potential error sources by real-time monitoring of the working status of the equipment. For example, the system monitors the changes in the output signal of the instrument and compares it with the set standard value or historical data to automatically detect whether the equipment has drifted. If drift is found, the error correction mechanism is activated to identify the specific parameters of the drift (such as equipment calibration deviation, ambient temperature changes, etc.), and the errors in the detection process are corrected in real time.

[0036] Furthermore, the detection results are compensated through an active error correction mechanism to ensure the accuracy of the detection results. The working principle of this mechanism is to adjust the analysis results by calculating and compensating for the systematic errors generated during the detection process. For example, when the detection equipment drifts, the correction mechanism will correct the error based on the relationship between historical data and current detection data. This correction not only relies on the monitoring system of the device itself, but also can be combined with the accumulated learning experience in the self-supervised database to dynamically adjust and compensate the data.

[0037] The combination of the above steps ensures that the test results are highly accurate and reliable even when the equipment drifts or external factors change during the multi-dimensional residue detection process. Through self-supervised learning and active error correction mechanisms, the system can automatically optimize and adjust, improve the intelligence level of the detection process, reduce manual intervention, and ensure the long-term stability of the test results. This makes mancozeb residue detection more accurate and efficient in practical applications.

[0038] P30: Use the deep feature extraction channel to extract features of the multidimensional data set and establish a deep feature set.

[0039] Furthermore, the deep feature extraction channel is used to extract features of the multidimensional data set to establish a deep feature set. Step P30 of the embodiment of the present application further includes:

[0040] P31: A residual attention convolutional network is used to extract key peaks of spectral data in a multidimensional dataset and establish spectral features; P32: A gated recurrent unit is used to analyze the chromatographic peak shape and retention time of chromatographic data in a multidimensional dataset and extract chromatographic features; P33: A temporal convolutional network is used to analyze the long-term dependency of electrochemical sensor data in a multidimensional dataset and extract electrochemical features; P34: A deep feature set is established based on the spectral features, the chromatographic features, and the electrochemical features.

[0041] Optionally, a deep feature extraction channel is used to extract features from the multidimensional data set to build a deep feature set. The core of this process is to use deep learning methods to extract useful features from multiple dimensions such as spectral data, chromatographic data, electrochemical sensor data in the multidimensional data set, and fuse these features into a high-dimensional deep feature set to support subsequent analysis and decision-making.

[0042] To achieve this goal, the Residual Attention Convolutional Network (RACN) is first used to extract the key peaks of the spectral data in the multidimensional data set. Spectral data usually contains multiple peaks, each of which represents the absorption characteristics of a certain component in the sample. By using the residual attention convolutional network, the system can adaptively focus on the most important peaks to extract spectral features. The residual network can effectively avoid the problem of gradient vanishing, and the attention mechanism allows the network to automatically identify the key peaks that affect the residue of mancozeb during training. The advantage of this method is that it can extract high-quality features from complex spectral data, and it has strong robustness and can effectively deal with noise and interference.

[0043] Next, the gated recurrent unit (GRU) was used to analyze the chromatographic data in the multidimensional data. The key features of chromatographic data include chromatographic peak shape and retention time. The chromatographic peak shape reflects the separation of different components in the sample, while the retention time directly affects the migration speed of mancozeb in the chromatographic column. The gated recurrent unit is a recurrent neural network (RNN) model suitable for time series data. By capturing the time dependency in the chromatographic data, the GRU can accurately extract the characteristics of the chromatographic peak shape and retention time, and then obtain the chromatographic characteristics. The key to this step is that the GRU can handle the long-term dependencies in the chromatographic data, avoiding the gradient vanishing problem of the traditional RNN, and ensuring the accurate extraction of the chromatographic data features.

[0044] Furthermore, a temporal convolutional network (TCN) is used to perform long-term dependency analysis on electrochemical sensor data in multidimensional data. Electrochemical sensor data usually has obvious time series characteristics. By analyzing the temporal changes of electrochemical sensor signals through a temporal convolutional network, electrochemical characteristics can be extracted. Temporal convolutional networks are particularly suitable for processing long time series data. They can capture the deep laws in the signal within multiple time steps, and then identify potential influencing factors of mancozeb residues. Through this analysis, the system can more accurately capture the changes in electrochemical signals and further improve detection accuracy.

[0045] Finally, the spectral features, chromatographic features, and electrochemical features extracted in the above steps are combined and fused to establish the final deep feature set. These features represent the information of different detection dimensions in multidimensional data. After feature extraction by the deep learning model, a high-dimensional, comprehensive feature set is formed, which can provide complete data support for subsequent decision-making layer analysis. By fusing the features of different data sources, the deep feature set can more comprehensively reflect the residue situation of the sample to be tested, thereby improving the accuracy and robustness of detection.

[0046] P40: After globally and locally aligning the deep feature set, the alignment result and the environmental factor identifier are input into the decision layer. After high-dimensional feature fusion is performed using the decision layer, a residual detection decision is generated after reverse verification based on the distribution constraint.

[0047] Furthermore, the depth feature set is globally and locally aligned. Step P40 in the embodiment of the present application further includes:

[0048] P41: After performing feature normalization on the deep feature set, the information entropy of each modal feature after normalization is calculated, and information compensation is performed based on the information entropy to establish a global feature alignment result; P42: After performing feature point detection on the deep feature set, feature point matching between different modalities is performed through cross-modal attention to perform feature point alignment; P43: After performing local dynamic weight adjustment on the aligned feature points, a local feature alignment result is established.

[0049] It should be understood that the deep feature set is globally aligned to ensure that the features between different data modalities can be accurately matched, thereby improving the accuracy of the final detection result. The purpose of global-local feature alignment is to effectively fuse the information in multimodal data to provide high-quality input for the subsequent decision layer.

[0050] First, the deep feature set is normalized to eliminate the dimensional differences in different modal data. After normalization, the information entropy of each modal feature is calculated. Information entropy is used to measure the uncertainty and information content of a feature. Features with lower entropy values ​​indicate less information, while features with higher entropy values ​​contain more information. In order to ensure that the data of each modality can have a relatively balanced contribution when fused, information compensation is performed based on these information entropy values. Specifically, for low-entropy features, their weights are enhanced, while for high-entropy features, their weights are appropriately reduced to prevent certain high-entropy features from dominating the final decision due to excessive information. In this way, a global feature alignment result that has been compensated for information content is established, so that the information content between each modality is more balanced.

[0051] Next, feature point detection is performed on the deep feature set, and the accuracy of the feature alignment process is ensured by extracting key feature points in the spectral, chromatographic and electrochemical data. For example, in spectral data, peak detection is used to extract the peak position of the key wavelength; in chromatographic data, Gaussian fitting is used to determine the peak center position; for electrochemical data, dynamic time warping is used to align the reaction points to ensure that the key points in different modal data can be accurately docked. Feature point detection is the basis of the alignment process, which ensures that different modal data are comparable and consistent in the same coordinate system.

[0052] After feature point detection, the cross-modal attention mechanism is used to match features between modalities. By using spectral features as queries and chromatographic features as key-value pairs, the attention mechanism is used to calculate the relationship between the features of each modality, thereby achieving feature point alignment between modalities. This cross-modal attention mechanism can effectively capture the potential connections between different modalities and ensure that the features of all modalities are accurately matched in a unified space.

[0053] Furthermore, after cross-modal feature alignment, the aligned feature points are subjected to local dynamic weight adjustment. The purpose of local dynamic weight adjustment is to dynamically adjust the fusion weight of each feature point according to the similarity between feature points. Exemplarily, an adaptive feature weighting strategy can be adopted to calculate the similarity between feature points and adjust the weight during fusion according to the similarity. Through this dynamic weighting, the system can adjust the impact of each feature point on the final decision according to its importance, ensuring that the fusion result can more accurately reflect the mancozeb residue in the sample.

[0054] Finally, a deep feature set is established after global-local feature alignment. This feature set will be input into the decision layer to provide reliable input data for the decision layer and provide efficient support for the final residue detection decision.

[0055] Furthermore, the decision layer is used to perform high-dimensional feature fusion, and step P40 of the embodiment of the present application further includes:

[0056] P44: Call the fusion channel of the decision layer to perform cross-modal high-dimensional feature fusion, as follows:

[0057] ;in, Characterization The feature matrix of the layer, Characterization The feature matrix of the layer, is the activation function, Characterization The trainable weight matrix of the layer, Characterization Node The neighbor set of Characterization Node Any neighbor of Characterization Node The number of neighbors of Characterization Node No. layer feature matrix; P45: calculate the correlation data between the high-dimensional feature fusion result and the target label, perform mutual information loss minimization optimization according to the correlation data, and update the high-dimensional feature fusion result according to the optimization result.

[0058] Optionally, a decision layer is used to perform high-dimensional feature fusion, thereby ensuring that multimodal features can be integrated at a higher level and improving the detection accuracy of mancozeb residues.

[0059] First, the fusion channel of the decision layer is called to perform cross-modal high-dimensional feature fusion on features from different modalities. To achieve this process, a structure containing multiple feature matrices can be established to represent feature information at different levels. These feature matrices are used to describe the node features at each level and the relationship between them.

[0060] Among them, the feature matrix of each layer represents the feature representation of the input data of this layer, which can be regarded as the abstraction or transformation of the original data by this layer. It contains the feature information of each node. The trainable weight matrix represents the trainable weights between each node in each layer of the network. It is used to map the feature matrix of the previous layer to the feature matrix of the current layer, and is a key parameter in the deep learning model. In addition, when processing graph-structured data, the neighbor set of a node is used to describe the relationship between the current node and other nodes. The number of neighbors indicates how many directly adjacent nodes the node has. In cross-modal fusion, neighbor information is used to ensure that local information in the data is fully considered. Characterization Node , which represents other nodes that each node may be connected to. In this way, the relationship and information flow between nodes can be accurately modeled, ensuring the correct docking of cross-modal features.

[0061] The integration of these feature matrices and related information, through the action of the activation function, gives the feature matrix of each layer the ability to transform nonlinearly, making the feature fusion process more flexible and effective. The core goal of this process is to mine potential deep relationships in cross-modal data through accurate feature fusion, thereby improving the accuracy of the final detection results.

[0062] Next, the correlation data between the high-dimensional feature fusion result and the target label is calculated. Specifically, the relationship between the fusion result and the target label is measured by calculating the mutual information between the two. Mutual information is a statistic that measures the correlation between two random variables and can capture the hidden relationship in the data. Therefore, mutual information loss minimization optimization is performed based on the calculated mutual information data. The purpose is to optimize the high-dimensional feature fusion result by minimizing the loss function so that it is more in line with the target label.

[0063] During the optimization process, the parameters of the high-dimensional feature fusion results can be updated through the back-propagation algorithm, which can gradually improve the accuracy and robustness of the model. Finally, after the mutual information loss minimization optimization, a more accurate fusion result is generated, providing a more reliable basis for subsequent residue detection decisions.

[0064] Furthermore, step P44 of the embodiment of the present application also includes:

[0065] P44-1: Call the first channel of the decision layer to perform spectral-chromatographic mutual information calculation, perform feature fusion, and generate a first fusion focus result; P44-2: Call the second channel of the decision layer to perform spectral-electrochemical nonlinear relationship feature fusion, and generate a second fusion focus result; P44-3: Call the third channel of the decision layer to perform chromatographic-electrochemical high-dimensional feature mapping, and establish a third fusion focus result based on the high-dimensional feature mapping results; P44-4: Use the environmental factor identifier to compensate for the first fusion focus result, the second fusion focus result, and the third fusion focus result, and then establish a high-dimensional feature fusion result.

[0066] It should be understood that high-dimensional feature fusion across modalities can be achieved by calling multiple fusion channels of the decision layer. In order to ensure that information between different modalities can be combined more efficiently, different technical means can be used in this process to perform feature fusion on spectral, chromatographic and electrochemical data respectively.

[0067] First, the first channel of the decision layer is called to calculate the spectral-chromatographic mutual information. Mutual information calculation is used to measure the correlation between spectral data and chromatographic data, helping the system find the potential relationship between the two data types. By calculating the mutual information, it is possible to determine which features have a strong correlation between the two modalities, and these features are fused to generate the first fusion focus result. This fusion result reflects the complementarity and importance of spectral and chromatographic data in the detection of mancozeb residues, and can enhance the system's understanding of the joint effect of the two data types.

[0068] Next, the second channel of the decision layer is called to perform feature fusion on the nonlinear relationship between spectral data and electrochemical data. Since the characteristics of spectral and electrochemical data usually have a strong nonlinear relationship, the nonlinear feature fusion method can better capture the complex interaction between the two. Through this process, the second fusion focus result is generated, which reflects the nonlinear complementarity of spectral and electrochemical data in reflecting the residual characteristics of mancozeb.

[0069] Furthermore, the third channel of the decision layer is called to perform high-dimensional feature mapping between the chromatographic data and the electrochemical data. The core of this step is to map the chromatographic data and the electrochemical data into a common high-dimensional space through high-dimensional feature mapping, so that the system can effectively combine and compare the two in this space. This high-dimensional mapping not only retains the uniqueness of each data, but also enables their high-dimensional features to be fused under a unified framework to generate the third fusion focus result.

[0070] Finally, the environmental factor identifier is used to compensate for the above three fusion results. The environmental factor identifier represents the environmental change factors that may affect the detection results, such as temperature, humidity, etc. During the data fusion process, changes in environmental factors may affect the performance of certain features. Therefore, by compensating the fusion results, these external interferences can be eliminated to ensure the accuracy of the final results. For example, the corresponding environmental impact coefficient can be generated according to the degree of environmental impact, and the fusion results can be compensated and adjusted. After compensation, the system finally established a high-dimensional feature fusion result, which combines the advantages of spectral, chromatographic and electrochemical data, and effectively corrects external environmental factors.

[0071] Through these steps, not only can accurate feature fusion be performed between multiple modalities, but the reliability and accuracy of the fusion results can also be improved by introducing environmental factors, thereby providing high-quality input for the final residue detection decision.

[0072] Furthermore, the embodiment of the present application further includes step P50, and step P50 further includes:

[0073] P51: Obtain product source data of the product to be tested, conduct dynamic risk assessment based on the product source data and residue detection decision, and establish a residue diffusion simulation trajectory; P52: Provide abnormal residue warning based on the residue diffusion simulation trajectory.

[0074] In a possible embodiment of the present application, the intelligent decision-making ability of the mancozeb residue detection system can be further enhanced, and the ability to respond to potential residue risks can be improved through dynamic evaluation and early warning mechanisms.

[0075] First, obtain the product source data of the product to be tested. Product source data usually includes detailed information on the production, processing, storage, transportation and other links of the product. This information is very important for assessing the possible environmental impact of the product, the source of pollution during the treatment process, and the differences between different production batches. Combine these source data with the previous residue detection decision results to conduct a dynamic risk assessment. Dynamic risk assessment is to dynamically adjust the risk assessment value of the residue based on the real-time detected residue situation, historical data and product source characteristics. At this point, the system will calculate and establish a residue diffusion simulation trajectory, which can simulate how residues such as mancozeb spread and diffuse in different links, as well as their potential impact on the entire supply chain.

[0076] By establishing a simulated trajectory of residual diffusion, the system can predict the impact of different factors (such as raw material sources, production conditions, transportation routes, etc.) on product residues and provide data support for subsequent management measures. This trajectory not only reflects the current residue status of the product, but also helps trace possible sources of contamination, further providing effective guidance for product quality control and safety assurance.

[0077] Next, an abnormal residue warning is performed based on the aforementioned residue diffusion simulation trajectory. When the system identifies that the residue values ​​of certain links and batches are beyond the normal range, an abnormal warning can be automatically issued based on the residue diffusion simulation trajectory. This early warning mechanism can help manufacturers, regulatory authorities, etc. to detect problems in a timely manner and take necessary prevention and control measures, such as recalling products, adjusting production processes, and strengthening product testing. This not only improves product safety, but also prevents the spread of harmful substances and avoids potential harm to consumers and the environment.

[0078] Through this dynamic risk assessment and early warning mechanism, the spread trend of pesticide residues such as mancozeb can be predicted and controlled, the product safety status can be grasped in real time, and targeted measures can be taken to effectively intervene to ensure the compliance and safety of the product. At the same time, this method effectively combines residue detection with risk assessment and early warning systems, improving the intelligence and accuracy of the overall detection process.

[0079] In summary, the embodiments of the present application have at least the following technical effects:

[0080] This application interacts with the detection unit data, reads the quantity and batch data of the product to be tested, determines the sample distribution and performs random collection under distribution constraints. After the collected samples undergo multidimensional residue detection (spectroscopy, chromatography, electrochemical sensing), a multidimensional data set with environmental factor identification is established. Features are extracted using the deep feature extraction channel, a deep feature set is established, and global-local feature alignment is performed. The alignment results are combined with the environmental factor input decision layer for high-dimensional feature fusion, and finally a residue detection decision is generated through reverse verification.

[0081] The technical effect of improving detection accuracy and work efficiency while enhancing environmental adaptability is achieved through multi-dimensional residue detection and high-dimensional feature fusion.

[0082] Embodiment 2, based on the same inventive concept as the intelligent detection and analysis method for mancozeb residue in the above embodiment, Figure 2 As shown, the present application provides a mancozeb residue intelligent detection and analysis system, and the system and method embodiments in the present application embodiments are based on the same inventive concept. The system includes:

[0083] The product data interaction module 11 is used to read the product data set of the product to be tested after the mancozeb residue intelligent detection and analysis system interacts with the detection unit, and the product data set includes product quantity data and product batch data.

[0084] A multidimensional residue detection module 12 is used to determine the sample distribution according to the product data set, perform random sample collection under distribution constraints, and establish a multidimensional data set after the analysis samples pass the multidimensional residue detection. The multidimensional residue detection includes spectral detection, chromatographic detection, and electrochemical sensor detection. The multidimensional data set is marked with environmental factors.

[0085] The feature extraction module 13 is used to extract features of the multi-dimensional data set by using a deep feature extraction channel to establish a deep feature set.

[0086] The residual detection decision module 14 is used to perform global-local feature alignment on the deep feature set, input the alignment result and environmental factor identification into the decision layer, use the decision layer to perform high-dimensional feature fusion, and generate a residual detection decision after reverse verification based on the distribution constraint.

[0087] Furthermore, the multidimensional residue detection module 12 is further configured to perform the following steps:

[0088] A self-supervisory database for residue detection is established; active error identification of equipment drift in the multi-dimensional residue detection process is performed based on the self-supervisory database, and an active error correction mechanism is established; and detection result compensation is performed through the active error correction mechanism to complete multi-dimensional residue detection.

[0089] Furthermore, the feature extraction module 13 is also used to perform the following steps:

[0090] A residual attention convolutional network is used to extract key peaks of spectral data in a multidimensional data set and establish spectral features. A gated recurrent unit is used to analyze the chromatographic peak shape and retention time of chromatographic data in a multidimensional data set and extract chromatographic features. A temporal convolutional network is used to analyze the long-term dependency of electrochemical sensor data in a multidimensional data set and extract electrochemical features. A deep feature set is established based on the spectral features, the chromatographic features, and the electrochemical features.

[0091] Furthermore, the residue detection decision module 14 is further configured to perform the following steps:

[0092] After the deep feature set is subjected to feature normalization processing, the information entropy of each modal feature after normalization is calculated, and information compensation is performed based on the information entropy to establish a global feature alignment result; after feature point detection is performed on the deep feature set, feature point matching between different modalities is performed through cross-modal attention to perform feature point alignment; after local dynamic weight adjustment of the aligned feature points, a local feature alignment result is established.

[0093] Furthermore, the residue detection decision module 14 is further configured to perform the following steps:

[0094] The fusion channel of the decision layer is called to perform cross-modal high-dimensional feature fusion as follows:

[0095] ;in, Characterization The feature matrix of the layer, Characterization The feature matrix of the layer, is the activation function, Characterization The trainable weight matrix of the layer, Characterization Node The neighbor set of Characterization Node Any neighbor of Characterization Node The number of neighbors of Characterization Node No. layer; calculating the correlation data between the high-dimensional feature fusion result and the target label, performing mutual information loss minimization optimization according to the correlation data, and updating the high-dimensional feature fusion result according to the optimization result.

[0096] Furthermore, the residue detection decision module 14 is further configured to perform the following steps:

[0097] The first channel of the decision layer is called to perform spectral-chromatographic mutual information calculation, perform feature fusion, and generate a first fusion focus result; the second channel of the decision layer is called to perform spectral-electrochemical nonlinear relationship feature fusion, and generate a second fusion focus result; the third channel of the decision layer is called to perform chromatographic-electrochemical high-dimensional feature mapping, and a third fusion focus result is established according to the high-dimensional feature mapping result; the first fusion focus result, the second fusion focus result, and the third fusion focus result are compensated by using the environmental factor identifier to establish a high-dimensional feature fusion result.

[0098] Furthermore, the system also includes:

[0099] A dynamic risk assessment module, which is used to obtain the product source data of the product to be tested, perform dynamic risk assessment based on the product source data and the residue detection decision, and establish a residue diffusion simulation trajectory; an abnormal residue warning module, which is used to perform abnormal residue warning based on the residue diffusion simulation trajectory.

[0100] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0101] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0102] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A method for intelligent detection and analysis of mancozeb residues, characterized in that: The intelligent detection and analysis method for mancozeb residue is applied to the intelligent detection and analysis system for mancozeb residue, and the method comprises: After the mancozeb residue intelligent detection and analysis system interacts with the detection unit, a product data set of the product to be tested is read, wherein the product data set includes product quantity data and product batch data; After determining the sample distribution according to the product data set, random sample collection under distribution constraints is performed, and after the analysis samples are subjected to multidimensional residue detection, a multidimensional data set is established. The multidimensional residue detection includes spectral detection, chromatographic detection, and electrochemical sensor detection. The multidimensional data set is marked with environmental factors; Extracting features of the multidimensional data set using a deep feature extraction channel to establish a deep feature set; After globally and locally aligning the deep feature set, the alignment result and the environmental factor identifier are input into the decision layer. After high-dimensional feature fusion is performed using the decision layer, a residual detection decision is generated after reverse verification based on the distribution constraint.

2. The intelligent detection and analysis method for mancozeb residue according to claim 1, characterized in that: The step of performing global-local feature alignment on the deep feature set comprises: After performing feature normalization processing on the deep feature set, the information entropy of each modal feature after normalization is calculated, and information amount compensation is performed based on the information entropy to establish a global feature alignment result; After detecting feature points of the deep feature set, feature matching of feature points between different modalities is performed through cross-modal attention to perform feature point alignment; After the aligned feature points are subjected to local dynamic weight adjustment, the local feature alignment result is established.

3. A method for intelligent detection and analysis of mancozeb residues as claimed in claim 2, characterized in that: The using the decision layer to perform high-dimensional feature fusion includes: The fusion channel of the decision layer is called to perform cross-modal high-dimensional feature fusion as follows: ； in, Characterization The feature matrix of the layer, Characterization The feature matrix of the layer, is the activation function, Characterization The trainable weight matrix of the layer, Characterization Node The neighbor set of Characterization Node Any neighbor of Characterization Node The number of neighbors of Characterization Node No. The feature matrix of the layer; Calculate the correlation data between the high-dimensional feature fusion result and the target label, perform mutual information loss minimization optimization according to the correlation data, and update the high-dimensional feature fusion result according to the optimization result.

4. A method for intelligent detection and analysis of mancozeb residues as claimed in claim 2, characterized in that: The using the decision layer to perform high-dimensional feature fusion includes: Calling the first channel of the decision layer to perform spectrum-chromatogram mutual information calculation, perform feature fusion, and generate a first fusion focus result; Calling the second channel of the decision layer to perform spectral-electrochemical nonlinear relationship feature fusion to generate a second fusion focus result; Calling the third channel of the decision layer to perform chromatography-electrochemistry high-dimensional feature mapping, and establishing a third fusion focus result according to the high-dimensional feature mapping result; After compensating the first fusion focus result, the second fusion focus result, and the third fusion focus result using the environmental factor identifier, a high-dimensional feature fusion result is established.

5. The intelligent detection and analysis method for mancozeb residue according to claim 1, characterized in that: The sample is analyzed through multidimensional residue testing, including: Establish a self-monitoring database for residue testing; Active error identification of equipment drift during multi-dimensional residue detection is performed based on the self-supervisory database, and an active error correction mechanism is established; The active error correction mechanism is used to compensate for the detection result, so as to complete the multi-dimensional residue detection.

6. The intelligent detection and analysis method for mancozeb residue according to claim 1, characterized in that: After generating a residual detection decision after reverse verification according to the distribution constraint, the method includes: Obtain product source data of the product to be tested, perform dynamic risk assessment based on the product source data and residue detection decision, and establish a residue diffusion simulation trajectory; An abnormal residue warning is performed based on the residue diffusion simulation trajectory.

7. The intelligent detection and analysis method for mancozeb residue according to claim 1, characterized in that: The step of extracting features of the multidimensional data set using a deep feature extraction channel to establish a deep feature set includes: The residual attention convolutional network is used to extract the key peaks of spectral data in multidimensional data sets and establish spectral features; The gated cycle unit was used to analyze the chromatographic peak shape and retention time of the chromatographic data in the multidimensional dataset and extract the chromatographic features; Use temporal convolutional networks to analyze long-term dependencies of electrochemical sensing data in multidimensional datasets and extract electrochemical features; A depth feature set is established according to the spectral features, the chromatographic features, and the electrochemical features.

8. An intelligent detection and analysis system for mancozeb residues, characterized in that: The system comprises: A product data interaction module, wherein the product data interaction module is used to read a product data set of the product to be tested after the mancozeb residue intelligent detection and analysis system interacts with the detection unit, wherein the product data set includes product quantity data and product batch data; A multidimensional residue detection module, which is used to determine the sample distribution according to the product data set, perform random sample collection under distribution constraints, and establish a multidimensional data set after the analysis samples pass the multidimensional residue detection. The multidimensional residue detection includes spectral detection, chromatographic detection, and electrochemical sensor detection. The multidimensional data set is marked with environmental factors; A feature extraction module, wherein the feature extraction module is used to extract features of the multidimensional data set using a deep feature extraction channel to establish a deep feature set; The residual detection decision module is used to perform global-local feature alignment on the deep feature set, input the alignment result and environmental factor identification into the decision layer, use the decision layer to perform high-dimensional feature fusion, and generate a residual detection decision after reverse verification based on the distribution constraints.

Images