CN-121995017-A - Fruit and vegetable pesticide residue detection system and detection method

CN121995017ACN 121995017 ACN121995017 ACN 121995017ACN-121995017-A

Abstract

The invention relates to the technical field of detection, in particular to a fruit and vegetable pesticide residue detection system and a detection method, which are used for solving the problems that when a plurality of batches of fruits and vegetables are in a variable temperature environment, a substrate difference and a reagent batch fluctuate, the temperature lag is coupled with the substrate drift in pesticide residue detection, an evaluation sample set is fixed, so that implicit adaptation deviation is caused, compensation parameters are asynchronous with a data version, and detection deviation amplification and out-of-standard erroneous judgment are caused.

Inventors

XU MIAOMIAO
CAO ZHIQIANG
WANG WENXIAO
LIN JIJIE
WANG JIAYAO
CAO YIFAN

Assignees

烟台联蕾食品有限责任公司

Dates

Publication Date: 20260508
Application Date: 20260410

Claims (10)

1. A fruit and vegetable pesticide residue detection system comprises an acquisition module, a building module, a first generation module, an output module and a second generation module, and is characterized by further comprising a Wen Jijie coupling module and a batch updating module, wherein the Wen Jijie coupling module is used for compensating environmental temperature hysteresis to obtain a representative temperature and a drift mark, the batch updating module is used for updating an influence coefficient based on batch and substrate deviation, the building module is used for building a dynamic comparison library based on the environmental temperature, sample substrate and chemical detection reagent information output by the acquisition module, the first generation module is used for generating mixed detection liquid, the output module comprises a data evaluation module, a time sequence spectrum coding module, a temperature base condition coupling module and a deep learning evaluation module, the data evaluation module is used for acquiring a candidate sample set from historical detection data and field newly increased detection records, invoking the deep learning evaluation module to select an initial network, training the candidate network by using the target sample set to obtain the candidate network, using the candidate sample set and the target sample set as an input predicted output deviation ratio iteration by the data correction model to generate a target candidate sample set and writing the target candidate sample set into the dynamic comparison library, the time sequence coding module is used for converting and coding characteristics of detection signals time-frequency, the temperature base condition coupling module is used for generating mixed detection liquid, the characteristics and the temperature base condition coupling module is used for acquiring the characteristics and the representative temperature, the characteristic and the influence coefficient and the pesticide residue value is input to the initial dynamic residual value to the initial value.
2. The fruit and vegetable pesticide residue detection system according to claim 1, wherein the acquisition module comprises a temperature time sequence acquisition unit, a substrate identification reading unit, a reagent batch reading unit and a sample recording and packaging unit, wherein the temperature time sequence acquisition unit, the substrate identification reading unit and the test batch reading unit are all connected with the sample recording and packaging unit, and the sample recording and packaging unit is also connected with a pesticide residue detector.
3. The fruit and vegetable pesticide residue detection system according to claim 2, wherein in the acquisition module, the temperature time sequence acquisition unit acquires an environmental temperature time sequence of pesticide residue detection according to a preset sampling rate, the substrate identification reading unit acquires a substrate type identification of the fruit and vegetable samples, the reagent batch reading unit identifies and acquires a chemical detection reagent batch identification, the sample record packaging unit acquires detection signals and signal acquisition parameters of transmission light intensity changing along with time from a pesticide residue detector interface, and the detection signals and the signal acquisition parameters are packaged into candidate sample records according to a data field structure of a target sample set and written into historical detection data with the output of the temperature time sequence acquisition unit, the substrate identification reading unit and the reagent batch reading unit.
4. The fruit and vegetable pesticide residue detection system according to claim 1, wherein the establishing module comprises a target sample set management unit, a candidate sample set storage unit, a deviation index unit and a test data set hooking unit, the target sample set management unit stores and maintains a target sample set with a pesticide residue reference concentration label and a data field structure thereof, the candidate sample set storage unit receives candidate sample records packaged by the obtaining module and writes the candidate sample records into a dynamic comparison library according to the data field structure of the target sample set to form a candidate sample set, the deviation index unit records a deviation ratio and an iterative round identifier output by the data evaluation module and establishes a correlation index with the candidate sample records in the candidate sample set, and the test data set hooking unit hooks the test data set output by the data correction model to the dynamic comparison library by a data set version number when the deviation ratio meets a preset threshold value, and establishes a mapping relation with an ambient temperature, a sample matrix type, a reagent batch identifier, a representative temperature, a drift identifier and an influence coefficient.
5. The system for detecting pesticide residues on fruits and vegetables according to claim 1, wherein the first generating module is connected with the acquiring module to read the substrate type identifier of the sample, and connected with the establishing module to call the sample preparation parameter table of the substrate type, the rotary cutting of the fruit and vegetable samples is driven and controlled under the constraint of the sample preparation parameter table to generate the crushed samples, meanwhile, the stirring of the crushed samples and the chemical detection reagent is driven and controlled under the constraint of the sample preparation parameter table, the mixed detection liquid is output, and the sample preparation parameter table comprises rotary cutting rotation speed, rotary cutting duration, stirring rotation speed, stirring duration and chemical detection reagent adding amount.
6. The fruit and vegetable pesticide residue detection system according to claim 1, wherein the Wen Jijie coupling module comprises a temperature hysteresis compensation unit and a matrix drift identification unit, the temperature hysteresis compensation unit is connected with the acquisition module, receives an ambient temperature time sequence, performs time shift alignment and exponential smoothing on the ambient temperature time sequence based on a preset thermal inertia discrete state model to obtain a representative temperature, the thermal inertia discrete state model synchronously acquires the ambient temperature sequence and a mixed detection liquid temperature sequence under a plurality of groups of known temperature variation conditions, performs recursive least square parameter identification to determine model orders, time constants and sampling period discretization coefficients, the matrix drift identification unit is connected with the output module, receives detection signals and time-frequency spectrum point sets thereof, counts baseline offsets, peak position offsets and peak width variation amounts of the spectrum point sets under the representative temperature constraint, and generates matrix drift identification.
7. The fruit and vegetable pesticide residue detection system according to claim 1, wherein the data evaluation module extracts a candidate sample set with the same field structure as the target sample set from the historical detection data and the field newly-added detection record, invokes the deep learning evaluation module to select a plurality of different initial pesticide residue detection networks from the preset network configuration set, trains and stores each initial pesticide residue detection network by using the target sample set to obtain a plurality of candidate networks, respectively takes the candidate sample set and the target sample set as inputs for each candidate network to generate a first prediction output and a second prediction output, calculates a global deviation ratio between the first prediction output and the second prediction output, uses the global deviation ratio as an iterative driving amount control data correction model, performs deletion or addition to the candidate sample set to update the candidate sample set, invokes the global deviation ratio calculation in a circulating manner until the global deviation ratio falls into a preset threshold range, outputs the target candidate sample set and writes the target candidate sample set into the dynamic comparison library.
8. The fruit and vegetable pesticide residue detection system according to claim 7, wherein in the data evaluation module, when the data correction module updates the candidate sample set, the candidate sample set is set up into a layering index according to the sample matrix type identifier and the temperature interval to which the representative temperature belongs, and the sample number constraint of each layering is set, for each candidate sample record in each layering, the sample deviation contribution degree is calculated based on the first prediction output and the second prediction output statistics of the candidate sample records by a plurality of candidate networks, and samples are deleted from the layering of which the deviation contribution degree is higher than a first threshold value and the number of layered samples exceeds the sample number constraint; The hierarchical index uses a sample matrix type identifier and a temperature interval to which a representative temperature belongs as a joint key, maps candidate sample records to corresponding hierarchies, and maintains a sample ID set, a sample number constraint, a hierarchy deviation statistic and a truncated or supplemented sample pool pointer for each hierarchy; The lower limit of the constraint of the number of samples of each layer is obtained by calculating residual error variance between the reference concentration of pesticide residue of each layer and the output of a detection network after layering according to matrix type mark multiplied by representative temperature interval in the historical detection data and substituting a sample size formula of preset tolerance error and confidence parameter, and the upper limit is determined according to the product of the duty ratio of each layer in the historical detection data and the lumped scale of preset test data and takes the minimum value compared with the preset maximum number of samples.
9. A detection method, using a fruit and vegetable pesticide residue detection system as claimed in any one of claims 1 to 8, comprising the steps of: step 1, detection information acquisition, namely acquiring an environmental temperature time sequence of pesticide residue detection by an acquisition module, reading a matrix type identifier and a chemical detection reagent batch identifier of a fruit and vegetable sample, and acquiring a detection signal and a signal acquisition parameter of the change of transmitted light intensity along with time; Step 2, temperature hysteresis compensation, namely performing time shift alignment and exponential smoothing on an environmental temperature time sequence based on a preset thermal inertia discrete state model by a Wen Jijie coupling module to obtain a representative temperature; Step 3, generating a matrix drift identifier by Wen Jijie coupling modules based on a time-frequency spectrum point set of the detection signal and generating the matrix drift identifier under the representative temperature constraint; updating the influence coefficient by a batch updating module based on the chemical detection reagent batch identification and the matrix deviation; Step 5, a dynamic control library is built, wherein the dynamic control library is built or updated by a building module based on the environmental temperature, the sample matrix type, the chemical detection reagent information, the representative temperature, the matrix drift identification, the influence coefficient and the historical detection data; Step 6, generating a test data set, namely extracting a candidate sample set with the same field structure as a target sample set from historical detection data and field newly-added detection records by a data evaluation module, calling a deep learning evaluation module to select a plurality of initial pesticide residue detection networks, training the target sample set to obtain a plurality of candidate networks, respectively taking the candidate sample set and the target sample set as inputs based on the candidate networks to obtain a first prediction output and a second prediction output, calculating a global deviation ratio, driving a data correction model by the global deviation ratio to execute deletion or supplement iteration on the candidate sample set until the global deviation ratio falls into a preset threshold range, outputting the target candidate sample set as the test data set, and writing the test data set into a dynamic comparison library; step 7, generating a sample preparation mixed solution, namely driving the fruit and vegetable sample to rotary cut and stirring the minced sample and the chemical detection reagent by a first generation module according to a sample preparation parameter table of a matrix type to obtain a mixed detection solution; Step 8, time-frequency coding and conditional coupling, namely performing time-frequency conversion and coding on the detection signal by a time sequence spectrum coding module to obtain a feature vector, and performing conditional coupling on the feature vector and a representative temperature, a matrix drift mark and an influence coefficient by a temperature-based conditional coupling module; Step 9, evaluating the initial residual value, namely outputting the initial pesticide residual value by a deep learning evaluation module based on the feature after conditional coupling; And step 10, dynamically compensating and outputting, namely calling a dynamic comparison library by a second generation module to dynamically compensate the initial pesticide residue value to obtain a target residue value.
10. The detection method according to claim 9, wherein the step S4 includes the steps of extracting the reference concentration of pesticide residues and the initial pesticide residue value output by the deep learning evaluation module from the historical detection records retrieved by the batch updating module according to the chemical detection reagent batch identification, the sample matrix type identification and the temperature interval to which the representative temperature belongs in the dynamic control library, calculating the batch deviation amount between the reference concentration and the initial pesticide residue value, and weighting and summarizing the batch deviation amount according to the detection timestamp to form a current batch statistic, recursively fusing the current batch statistic with the batch influence coefficients existing in the dynamic control library according to the preset forgetting factor to obtain updated batch influence coefficients, writing the updated batch influence coefficients into the dynamic control library, and synchronously outputting the updated batch influence coefficients to the temperature-based condition coupling module and the second generation module for calling.

Description

Fruit and vegetable pesticide residue detection system and detection method Technical Field The invention relates to the technical field of detection, in particular to a fruit and vegetable pesticide residue detection system and a detection method. Background The Chinese patent publication No. CN120404672A discloses a pesticide residue detection method and a system for a pesticide residue detector, which are used for acquiring environmental temperature data, sample matrix type information and chemical detection reagents of fruit and vegetable samples, establishing a dynamic control database based on the environmental temperature data and the sample matrix type information and historical detection data, inputting the prepared mixed detection liquid and the environmental temperature data into the pesticide detector to acquire an initial pesticide residue detection value, dynamically compensating the initial pesticide residue detection value based on the dynamic control database to generate a target pesticide residue detection value, and solving the problems of low pesticide residue detection precision and poor efficiency of a pesticide residue detection scheme adopting a fixed temperature compensation coefficient under multi-batch fruit and vegetable samples and variable temperature environments. The detection method for deep learning can utilize a third party to detect samples, obtain spectrum points through batch, environmental temperature, fruit and vegetable matrixes, signals of transmitted light intensity changing along with time and time-frequency conversion, combine sample records with pesticide residue reference concentration labels, and obtain a model with optimal performance on the current samples through training and evaluating different deep learning models. While the current sample does not participate in the counter-propagation training of model parameters, the evaluation label is continuously acted on the model network structure and the iteration direction to cause the final model to gradually adapt to the environmental temperature interval distribution of the sample and the fruit and vegetable matrix, so that the evaluation index is continuously improved, meanwhile, under the influence of the new fruit and vegetable batch and matrix internal difference and the rapid temperature change and heat balance hysteresis caused by cold chain in-out warehouse, the increase of the deviation and out-of-standard misjudgment of the pesticide residue detection concentration occurs, and meanwhile, the field lacks of independent sample sources which are distributed in the same way and provided with real labels, namely the training scale is changed by directly reclassifying the training sample, the external random construction sample is difficult to ensure the consistency with the original sample in the fruit and vegetable matrix and the environmental temperature, and the information leakage is caused by constructing the new sample by the detection label of the original sample, so that the objectivity of the pesticide residue detection result output by the model is lost. Disclosure of Invention The invention aims to solve the technical problem of providing a fruit and vegetable pesticide residue detection system and a detection method, wherein Wen Jijie is coupled with a batch updating linkage dynamic comparison library, and a versionable test data set is generated by driving according to a deviation ratio, so that detection and evaluation are synchronously aligned with compensation parameters, the pesticide residue concentration deviation under a variable temperature and multi-matrix scene is reduced, and the exceeding error judgment is avoided. In order to achieve the above purpose, the present invention provides the following technical solutions: The invention provides a fruit and vegetable pesticide residue detection system, which comprises an acquisition module, a building module, a first generation module, an output module, a second generation module, a Wen Jijie coupling module, a batch updating module, a Wen Jijie coupling module, a batch updating module, a time sequence frequency spectrum coding module, a temperature base condition coupling module and a deep learning evaluation module, wherein the acquisition module, the building module, the first generation module, the output module and the second generation module are used for generating mixed detection liquid, the data evaluation module comprises a data evaluation module, a time sequence frequency spectrum coding module, a temperature base condition coupling module and a deep learning evaluation module, the data evaluation module acquires a candidate sample set from historical detection data and on-site newly increased detection records, the deep learning evaluation module is called to select an initial network, the candidate network is obtained by training of the target sample set, the candidate network is compared with a predicted output deviation