Fruit quality detection method based on two-dimensional data fusion

Abstract

The invention discloses a fruit quality detection method based on two-dimensional data fusion, and relates to the technical field of infrared detection.The method comprises the steps that near infrared spectrum data and quality data of known fruits are collected; extracting a reflectivity spectrum; determining an optimal latent variable number according to the reflectivity spectrum; screening acquisition sensors; acquiring smell data of the known fruits by using an acquisition sensor; fusing the optimal latent variable number and the smell data into a multivariable data set; establishing a correlation model between the multivariable data set and the quality data; collecting a multivariable data set of a to-be-detected fruit; and predicting quality data of the to-be-detected fruit according to the multivariable data set of the to-be-detected fruit and the association model. According to the method, different types of data from the near infrared spectrum and the electronic olfaction system are subjected to conjoint analysis, so that the comprehensiveness of fruit quality detection and sensory evaluation is improved.

Description

Technical Field

[0001] The present application relates to the field of infrared detection technology, and in particular to a fruit quality detection method based on two-dimensional data fusion. Background Art

[0002] Fruit quality testing is an important process for evaluating the quality and health of fruits. For farmers and producers, the quality of fruits is directly related to the market value of the products and consumer satisfaction. Common fruit quality testing indicators include appearance inspection, sugar content, acidity, hardness, dry matter content and internal defects. For these indicators, the commonly used detection methods are near-infrared spectroscopy and electronic olfaction technology, which respectively collect fruit optical properties and flavor data to evaluate the quality and sensory properties of fruits.

[0003] Near-infrared spectroscopy (NIR) is a spectral analysis technique that uses the absorption, reflection or transmission of light in the near-infrared region (wavelength is about 780nm to 2500nm) to analyze the composition of a substance. NIR mainly detects stretching vibrations and bending vibrations in molecules, and these vibration modes are mainly related to functional groups such as CH, NH, and OH.

[0004] An electronic olfactory system is an instrument that simulates the biological olfactory system and can identify odors by detecting mixtures of gases or volatile compounds. An electronic olfactory system is often composed of multiple different types of chemical sensors, each with different sensitivities to specific gases. When these sensors are exposed to odors, they produce specific response patterns, which can be matched to specific odors through pattern recognition technology.

[0005] Near-infrared spectroscopy and electronic olfactory systems play complementary roles in the quality and sensory evaluation of fruits. Near-infrared spectroscopy detects the internal chemical composition of fruits, such as sugar content, acidity, moisture content, and dry matter content, in a non-destructive way, helping to assess the maturity, sweetness, and internal defects of fruits. These indicators are key factors in judging the intrinsic quality of fruits. Electronic olfactory systems evaluate the external quality of fruits from a sensory perspective by detecting volatile organic compounds released by fruits and analyzing the aroma and flavor characteristics of fruits, such as sweetness and fruity intensity. In addition, electronic olfactory systems can also identify abnormal odors caused by corruption or disease during storage or transportation.

[0006] In the prior art, although near infrared spectroscopy and electronic olfactory systems have been widely used in fruit quality detection, these two types of data are usually analyzed independently, which may lead to a lack of comprehensiveness in the evaluation of fruit quality. Summary of the invention

[0007] The embodiment of the present application provides a fruit quality detection method based on two-dimensional data fusion, which is used to solve the problem of lack of comprehensiveness in evaluating fruit quality by near-infrared spectroscopy and electronic olfactory system alone in the prior art.

[0008] The present application embodiment provides a fruit quality detection method based on two-dimensional data fusion, comprising:

[0009] Collect near infrared spectral data and quality data of known fruits;

[0010] Extract reflectance spectra from near-infrared spectral data;

[0011] Find the minimum root mean square error of the reflectance spectrum at different wavelengths, and determine the optimal number of latent variables of the reflectance spectrum at different wavelengths based on the minimum root mean square error;

[0012] Different gas sensors are used to detect the odor of known fruits, and multiple gas sensors with the highest responsiveness are selected as collection sensors;

[0013] Using a collection sensor to collect smell data of known fruits;

[0014] Fusion of optimal number of latent variables and odor data into a multivariate dataset;

[0015] Build association models between multivariate data sets and quality data;

[0016] Collecting a multivariate data set of a fruit to be tested, wherein the fruit to be tested and the known fruit are of the same species;

[0017] The quality data of the fruits to be tested are predicted based on the multivariate data set of the fruits to be tested and the association model.

[0018] In a possible implementation, a black-white correction is performed on the near-infrared spectrum data to obtain a reflectance spectrum.

[0019] In a possible implementation, after the reflectance spectrum is obtained, the reflectance spectrum is preprocessed, and the preprocessing is one or more of smoothing, normalization, multivariate scattering correction, first-order derivative processing and second-order derivative processing.

[0020] In a possible implementation, the minimum root mean square error of the reflectance spectrum at different wavelengths is calculated according to the following formula:

[0021]

[0022] Where j represents the number of latent variables, RMSECV(j) represents the minimum root mean square error when the number of latent variables is j, M represents the number of samples, Y(i,j) represents the predicted value of the i-th sample under the j-th latent variable, and y(i) represents the actual value of the i-th sample;

[0023] The optimal number of latent variables for reflectance spectra at different wavelengths is determined by the following formula:

[0024]

[0025] Where LV represents the optimal number of latent variables.

[0026] In a possible implementation, after the smell data of the fruit is collected by using the collection sensor, the smell data is further subjected to dimensionality reduction processing, and the dimensionality reduction processing method is:

[0027] Decentralize the odor data collected by each collection sensor:

[0028]

[0029] Among them, X s ' represents the odor data after decentralized processing, X s represents the odor data collected by the sth sensor, M represents the number of samples, and X si represents the odor data of the i-th sample collected by the s-th collection sensor;

[0030] Find the covariance matrix:

[0031]

[0032] Where C represents the covariance matrix, and the superscript T represents the transpose;

[0033] Perform eigenvalue decomposition on the covariance matrix to obtain the corresponding eigenvalues ​​and eigenvectors, arrange the eigenvectors from large to small according to the eigenvalues, and take the first k columns to form the dimension reduction matrix P n×k , n represents the number of acquisition sensors;

[0034] Reduce the dimension of the odor data:

[0035]

[0036] Among them, Y k×M represents the odor data after dimensionality reduction. The superscript T represents transposition. n×m Represents a matrix composed of odor data.

[0037] In a possible implementation, the association model is established by the least square method, and the association model is expressed as:

[0038] Y=XW+E

[0039] Where Y represents the quality data of known fruits, X represents the multivariate data set of known fruits, W represents the weight matrix, and E represents the error term.

[0040] In a possible implementation, after the association model is trained using the training set data consisting of the multivariate data set and quality data of known fruits, the association model is also verified using the root mean square error.

[0041] A fruit quality detection method based on two-dimensional data fusion in this application has the following advantages:

[0042] Different types of data from near-infrared spectroscopy and electronic olfactory systems are jointly analyzed to improve the comprehensiveness of fruit quality detection and sensory evaluation. This method can make full use of the internal component analysis capability of near-infrared spectroscopy and the odor recognition capability of electronic olfactory systems, introduce the concept of two-dimensional data fusion, combine multi-source data through principal component analysis and other methods, and conduct a comprehensive evaluation of fruits from multiple perspectives. It has a wider applicability and can be applied to the detection of different types of fruits. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1 This is a flow chart of a fruit quality detection method based on two-dimensional data fusion provided in an embodiment of the present application.

[0045] Figure 2 This is a radar chart of the response of the gas sensor to kiwi fruit in the embodiment of the present application.

[0046] Figure 3 These are principal component analysis results for different numbers of latent variables and different gas sensors in the embodiments of the present application.

[0047] Figure 4 A schematic diagram of the curves of the training data, test data, and fitted association model provided in the embodiments of the present application. DETAILED DESCRIPTION

[0048] 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, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0049] Figure 1 A flow chart of a fruit quality detection method based on two-dimensional data fusion provided in an embodiment of the present application. An embodiment of the present application provides a fruit quality detection method based on two-dimensional data fusion, comprising:

[0050] S100, collects near-infrared spectral data and quality data of known fruits.

[0051] Exemplarily, known fruits are used to establish the association model in the laboratory. It should be understood that the known fruits and the fruits to be tested are different. The known fruits are pre-selected fruits that can better reflect the relationship between the near-infrared spectrum data, the odor data and the quality data. The selection can be based on different states of the fruits, such as representative fruits in unripe, fully ripe and over-ripe states. The fruits to be tested are fruits of unknown quality, which may be in any state.

[0052] The present application uses a reflection method to collect near-infrared spectral data. In the specific operation, the fruit used to establish the model or the fruit to be tested is placed on a reflective substrate made of stainless steel, aluminum, mirror, etc., and the fruit is irradiated with near-infrared light emitted by a near-infrared light source, and the reflected near-infrared light is received. The near-infrared spectral data can be obtained by performing spectral analysis on the reflected near-infrared light.

[0053] Quality data are the data that the market pays more attention to for fruits, such as sugar content, acidity, water content, etc. When collecting quality data, invasive collection methods can be used.

[0054] S110, extracting a reflectance spectrum from the near-infrared spectrum data.

[0055] Exemplarily, the reflectance spectrum can be obtained by performing black-white correction on the near-infrared spectrum data.

[0056] Furthermore, after obtaining the reflectance spectrum, the reflectance spectrum is preprocessed, and the preprocessing is one or more of smoothing processing, normalization processing, multivariate scattering correction, first-order derivative processing and second-order derivative processing.

[0057] S120, finding the minimum root mean square error of the reflectance spectrum at different wavelengths, and determining the optimal number of latent variables of the reflectance spectrum at different wavelengths according to the minimum root mean square error.

[0058] For example, a latent variable is a variable that affects a quality of fruit that can be detected by near-infrared spectroscopy, such as sugar content, acidity, water content, etc. This variable cannot be directly observed, but can be inferred based on other observable variables. Since latent variables cannot be directly observed, there are many possibilities, and the number of latent variables under each possibility is different. Different numbers of latent variables have different effects on the prediction accuracy of the model. Generally, the fewer the number of latent variables, the higher the prediction accuracy of the model, and vice versa. Therefore, in general, the optimal number of latent variables is only a small value, such as 1 or 3.

[0059] Furthermore, the minimum root mean square error of the reflectance spectrum at different wavelengths is calculated according to the following formula:

[0060]

[0061] Among them, j represents the number of latent variables, RMSECV(j) represents the minimum root mean square error when the number of latent variables is j, M represents the number of samples, Y(i,j) represents the predicted value of the i-th sample under the j-th latent variable, and y(i) represents the actual value of the i-th sample. The predicted value is the estimated value of a certain quality of the fruit obtained by analyzing the reflectance spectrum, such as sugar content, acidity, moisture content, etc., while the actual value is the actual value of a certain quality of the fruit obtained by directly measuring the fruit using various instruments, such as a sugar meter, acidity meter, etc.

[0062] The optimal number of latent variables for reflectance spectra at different wavelengths is determined by the following formula:

[0063]

[0064] Among them, LV represents the optimal number of latent variables. After using the above formula, the optimal number of latent variables LV is the number of latent variables when the minimum root mean square error value is minimized. When the minimum root mean square error is minimized, the difference between the predicted value and the actual value of a certain quality of the fruit is also at the minimum state, that is, the predicted value and the actual value are basically the same. At this time, it can be considered that the model obtained by analyzing the near-infrared spectrum has a high accuracy rate in predicting the quality of the fruit.

[0065] S130, using different gas sensors to detect the smell of known fruits, and selecting multiple gas sensors with the highest responsiveness as collection sensors.

[0066] For example, fruits release aromas when they mature. These aromas are essentially volatile gases, which usually contain substances such as alkyl esters and alkenyl esters, alcohols, aldehydes and ketones. The smell of each fruit is different. Therefore, before testing the quality of the fruit, it is necessary to select several models of commonly used gas sensors that have a high response to the smell of the current fruit, so as to reduce the amount of odor data in the subsequent fruit quality detection, and at the same time eliminate the adverse effects of odor data collected by gas sensors with poor response on the accuracy of the detection results.

[0067] When selecting a collection sensor, each gas sensor can be used to collect odor data of the same fruit at the same time, and a radar chart can be drawn according to the size of the odor data. The radar chart can intuitively understand the size difference of odor data collected by different gas sensors for the same fruit, and multiple gas sensors with the highest responsiveness, that is, the largest odor data, for example, 4 gas sensors, can be used as collection sensors.

[0068] S140, using a collection sensor to collect smell data of known fruits.

[0069] For example, after the smell data of known fruits are collected by using the collection sensor, the smell data is further subjected to dimensionality reduction processing, and the dimensionality reduction processing method is:

[0070] Decentralize the odor data collected by each collection sensor:

[0071]

[0072] Among them, X s ' represents the odor data after decentralized processing, X s represents the odor data collected by the sth sensor, M represents the number of samples, that is, the number of known fruits involved in the experiment, and X si represents the odor data of the i-th sample collected by the s-th collection sensor;

[0073] Find the covariance matrix:

[0074]

[0075] Where C represents the covariance matrix, and the superscript T represents the transpose;

[0076] Perform eigenvalue decomposition on the covariance matrix to obtain the corresponding eigenvalues ​​and eigenvectors, arrange the eigenvectors from large to small according to the eigenvalues, and take the first k columns to form the dimension reduction matrix P n×k , n represents the number of acquisition sensors;

[0077] Reduce the dimension of the odor data:

[0078]

[0079] Among them, Y k×M represents the odor data after dimensionality reduction. The superscript T represents transposition. n×m Represents a matrix composed of odor data.

[0080] S150, fusing the optimal number of latent variables and the odor data into a multivariate dataset.

[0081] Exemplarily, the fused multivariate data set includes the optimal number of latent variables extracted from the near-infrared spectral data and the data after dimensionality reduction processing of the odor data. Therefore, the multivariate data set contains information of both the near-infrared spectral data and the odor data. When the multivariate data set is used to detect the quality of fruit, the near-infrared spectral data and the odor data can be comprehensively considered, thereby improving the comprehensiveness and accuracy of the fruit quality detection results.

[0082] Furthermore, after the multivariate data set of known fruits is obtained by fusion, principal component analysis is performed on the multivariate data set to obtain the main components, and then the Pearson correlation coefficients of the reflectance spectra and odor data corresponding to the main components are determined.

[0083] Specifically, the Pearson correlation coefficient of the reflectance spectrum and odor data corresponding to the main component was determined according to the following formula:

[0084]

[0085] Where r represents the Pearson correlation coefficient, N represents the number of principal components, and X r represents the odor data in the main components, represents the mean of the odor data in the main components, Y r Data representing the reflectance spectra of the main components, Represents the mean of the data for the reflectance spectra in the principal components.

[0086] S160, establishing a correlation model between the multivariate data set and the quality data.

[0087] Exemplarily, the association model is established by the least squares method, and the association model is expressed as:

[0088] Y=XW+E

[0089] Where Y represents the quality data of known fruits, X represents the multivariate data set of known fruits, W represents the weight matrix, and E represents the error term.

[0090] Furthermore, after training the association model using the training set data consisting of the multivariate data set and quality data of known fruits, the association model was verified using the root mean square error using the test data set.

[0091] The performance of the correlation model can be verified by the root mean square error (RMSEP) as shown below:

[0092]

[0093] Where RMSEP(i) is the root mean square error of the i-th sample, Y i is the actual quality data of the i-th sample, is the prediction quality data of the i-th sample. The smaller the RMSEP, the more stable the model is.

[0094] If the verification result shows that the prediction result accuracy of the association model meets the requirement, the association model can be applied to subsequent practical use, otherwise the association model will continue to be trained until it passes the verification.

[0095] Furthermore, the present application also uses the determination coefficient R 2 Verify the correlation between predicted quality data and multivariate data sets. 2 It is called the coefficient of determination and is an important indicator for measuring the goodness of fit of the model. Its expression is as follows:

[0096]

[0097] In the formula, is the mean of the known fruit quality data. 2 The closer the value is to 1, the more relevant the fused multivariate dataset is to the fruit quality data, and the fused multivariate dataset can predict the quality of the fruit.

[0098] S170, collecting a multivariate data set of the fruit to be detected, wherein the fruit to be detected and the known fruit are of the same type.

[0099] For example, since the optimal number of latent variables has been determined in steps S100-S120, when the types of fruits are the same, the optimal number of latent variables does not need to be calculated again, and the result obtained in step S120 can be directly used. When collecting the odor data of the fruit to be detected, the collection sensor selected in step S130 can be used. At this point, the optimal number of latent variables and the odor data of the fruit to be detected can be obtained, and the multivariate data set of the fruit to be detected can be obtained by fusing the two.

[0100] S180, predicting quality data of the fruit to be tested based on the multivariate data set of the fruit to be tested and the association model.

[0101] Exemplarily, after the association model between the quality data and the multivariate data set is established in step S160, the relationship between the quality data and the multivariate data set has been determined, so after the multivariate data set of the fruit to be tested is obtained by fusion in step S170, it can be input into the association model as an independent variable to obtain the predicted quality data of the fruit to be tested. After the predicted quality data is obtained, the quality of the fruit to be tested can be determined according to the specific value of the quality data.

[0102] The method of the present application is further described below using kiwi fruit as an example.

[0103] The near-infrared spectral data, odor data, and sugar content data of known kiwifruit were collected. The near-infrared spectral data is the reflectance spectrum of kiwifruit at a wavelength of 600-1100nm. The odor data includes the concentration data of substances such as alkyl esters and alkenyl esters, alcohols, aldehydes, and ketones. A total of 10 gas sensors from the MQ and TGS series were used to collect the odor data.

[0104] The near-infrared spectral data was preprocessed, and the reflectance spectral data of kiwifruit was obtained by black-and-white correction. The reflectance spectrum was preprocessed by integrating one of the following methods: smoothing, normalization, multivariate scattering correction, first-order derivative processing, and second-order derivative processing. The minimum root mean square error was obtained by K-fold cross-validation of the partial least squares method, and the corresponding optimal latent variable was determined.

[0105] MQ and TGS gas sensors are detection devices based on metal oxide semiconductor (MOS) technology and are widely used in environmental monitoring, industrial safety, and food quality control. They detect the reaction of gas molecules with sensing materials through internal heating elements, causing resistance changes, thereby measuring the concentration of specific gases (such as methane, hydrogen, alcohol, carbon monoxide, etc.). These sensors have become common gas detection tools due to their high sensitivity, low cost, and ease of use. When collecting kiwifruit odor data, MQ and TGS gas sensors that have a strong response to the aroma of kiwifruit were selected as the main data support.

[0106] Radar images of the responses of 10 gas sensors to kiwi fruit Figure 2 As shown in the radar chart, it can be seen that MQ2, MQ4, TGS2610 and TGS2620 gas sensors react most obviously to the aroma of kiwi fruit, so these four gas sensors are selected as sensors for collecting kiwi fruit odor data, that is, collection sensors.

[0107] The optimal latent variable number extracted from the near-infrared spectral data and the odor data collected by the acquisition sensor are fused. The optimal latent variable number is related to the sugar content of kiwifruit. The latent variables related to the sugar content of kiwifruit are extracted by partial least squares regression. The principal component analysis of the fused data set is performed to observe the correlation between the optimal latent variable number LV of kiwifruit samples and the acquisition sensors MQ2, MQ4, TGS2610 and TGS2620. Figure 3 As shown, LV1, LV3, MQ2, MQ4 and TGS2610 were selected as the main components for data analysis.

[0108] Calculate the Pearson correlation coefficient of the reflectance spectrum corresponding to the best latent variable in the principal component in the near infrared spectrum and the odor data collected by the acquisition sensor. In the calculation, X r represents the odor data collected by the sensors MQ2, MQ4 and TGS2610, Y r Represents the reflectance spectra corresponding to the latent variables LV1 and LV3.

[0109] After the optimal number of latent variables and odor data are integrated to obtain a multivariate data set, the least squares method is used to establish the association model, and the association model is trained and verified. The curves of the training data, test data, and the finally fitted association model are shown in Figure 4 After verification, the RMSEP value of the association model in this application is 0.41, R 2 The value is 0.89, indicating that the association model between the fused multivariate dataset and the sugar content of kiwifruit is stable and the multivariate dataset has an excellent correlation with the sugar content.

[0110] In the experiment, the number of kiwifruit samples was 90, and the reflectance spectra of the kiwifruit samples were collected to determine the corresponding optimal number of latent variables. After the optimal number of latent variables and the odor data of different acquisition sensors were fused, the Pearson correlation coefficient of the optimal number of latent variables and the acquisition sensors was calculated. The results showed that the acquisition sensors MQ2, MQ4 and TGS2610 had a high correlation with the latent variables LV1 and LV3, with the highest correlation reaching 0.91, indicating that the two-dimensional fusion of near-infrared spectral data and odor data in this application has a high accuracy.

[0111] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0112] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A fruit quality detection method based on two-dimensional data fusion, characterized in that: include: Collect near infrared spectral data and quality data of known fruits; extracting a reflectance spectrum from the near infrared spectrum data; Calculating the minimum root mean square error of the reflectivity spectrum at different wavelengths, and determining the optimal number of latent variables of the reflectivity spectrum at different wavelengths according to the minimum root mean square error; Using different gas sensors to detect the smell of known fruits, and selecting a plurality of gas sensors with the highest responsiveness as collection sensors; Using the collection sensor to collect smell data of known fruits; fusing the optimal number of latent variables and the odor data into a multivariate data set; establishing a correlation model between the multivariate data set and the quality data; Collecting a multivariate data set of a fruit to be tested, wherein the fruit to be tested and the known fruit are of the same species; The quality data of the fruit to be detected is predicted according to the multivariate data set of the fruit to be detected and the association model.

2. A fruit quality detection method based on two-dimensional data fusion according to claim 1, characterized in that: The near-infrared spectrum data is subjected to black-and-white correction to obtain the reflectance spectrum.

3. A fruit quality detection method based on two-dimensional data fusion according to claim 1, characterized in that: After obtaining the reflectivity spectrum, the reflectivity spectrum is also preprocessed, and the preprocessing is one or more of smoothing, normalization, multivariate scattering correction, first-order derivative processing and second-order derivative processing.

4. A fruit quality detection method based on two-dimensional data fusion according to claim 1, characterized in that: The minimum root mean square error of the reflectance spectrum at different wavelengths is calculated according to the following formula: Where j represents the number of latent variables, RMSECV(j) represents the minimum root mean square error when the number of latent variables is j, M represents the number of samples, Y(i,j) represents the predicted value of the i-th sample under the j-th latent variable, and y(i) represents the actual value of the i-th sample; The optimal number of latent variables for the reflectance spectrum at different wavelengths is determined according to the following formula: Wherein, LV represents the optimal number of latent variables.

5. A fruit quality detection method based on two-dimensional data fusion according to claim 1, characterized in that: After the smell data of the fruit is collected by the collection sensor, the smell data is further subjected to dimensionality reduction processing, and the dimensionality reduction processing method is: Decentralize the odor data collected by each of the collection sensors: Among them, X s ' represents the odor data after decentralized processing, X s represents the odor data collected by the sth collection sensor, M represents the number of samples, X si represents the odor data of the i-th sample collected by the s-th collection sensor; Find the covariance matrix: Where C represents the covariance matrix, and the superscript T represents the transpose; Perform eigenvalue decomposition on the covariance matrix to obtain corresponding eigenvalues ​​and eigenvectors, arrange the eigenvectors from large to small according to the eigenvalues, and take the first k columns to form a dimension reduction matrix P n×k , n represents the number of the acquisition sensors; The odor data is subjected to dimensionality reduction processing: Among them, Y k×M represents the odor data after dimensionality reduction processing, the superscript T represents transposition, and X n×m A matrix representing the odor data.

6. A fruit quality detection method based on two-dimensional data fusion according to claim 1, characterized in that: The association model is established by the least square method, and the association model is expressed as: Y=XW+E Wherein, Y represents the quality data of the known fruit, X represents the multivariate data set of the known fruit, W represents the weight matrix, and E represents the error term.

7. A fruit quality detection method based on two-dimensional data fusion according to claim 6, characterized in that: After the association model is trained using the training set data consisting of the multivariate data set and the quality data of known fruits, the association model is also verified using the root mean square error.

Images