Nondestructive detection method and system for sugar content of wu citrus based on resistance impedance and multi-source feature fusion

Abstract

This invention provides a non-destructive method and system for detecting the sugar content of Wogan mandarins based on the fusion of electrical impedance and multi-source features. The method involves placing the Wogan mandarin orange to be tested on an insulating support surface and attaching electrode patches to the equatorial region of the orange to obtain electrical impedance information. External phenotypic data, including the weight, first diameter, second diameter, equatorial circumference, and height of the orange, are acquired. The electrical impedance and external phenotypic data are preprocessed to obtain derived features, which are then concatenated with the electrical impedance and external phenotypic data to form a multi-source feature set. This multi-source feature set is input into a pre-trained nonlinear regression model for prediction, yielding a predicted sugar content value. This invention integrates the electrical impedance and external phenotypic data of the Wogan mandarin orange, enabling the output of predicted sugar content values ​​without damaging the orange sample, thus improving the convenience and predictive stability of single-fruit sugar content detection.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and smart agriculture, and in particular to a non-destructive testing method and system for the sugar content of Wogan oranges based on the fusion of electrical impedance and multi-source features. Background Technology

[0002] As the development direction of the Wogan mandarin orange industry gradually shifts from scale expansion to quality grading and branding, research on rapid, stable, and low-cost testing of the internal quality of individual Wogan mandarins has become a crucial issue for post-harvest sorting and refined sales. Current methods for testing the sugar content of Wogan mandarins mainly rely on measuring the sugar content using a refractometer after juice extraction. Each test requires cutting open the fruit and extracting the juice, rendering the sample no longer commercially viable after testing. This method is insufficient to meet the needs for rapid grading and on-site testing during orchard acquisition, packaging grading, or warehouse sampling.

[0003] With the rapid development of non-destructive testing technologies such as near-infrared spectroscopy, hyperspectral imaging, and machine vision, these technologies can indirectly characterize quality indicators of Wogan oranges, such as sugar content, acidity, firmness, and moisture content, through spectral response, image texture, or spatial spectral information, achieving good predictive results under experimental conditions. However, in practical applications, optical testing equipment often faces problems such as high equipment cost, large system size, sensitivity to ambient light and fruit peel condition, and limited model transferability. Especially when targeting rapid single-fruit testing, portable testing terminals, and embedded deployments in sorting lines, complex optical paths, calibration and maintenance, and computational loads increase the application threshold of the system. Unlike optical detection methods, bioelectrical impedance analysis (BIA) focuses on the electrical response of Wogan orange tissues under an electric field. Wogan orange pulp and peel tissues contain intracellular fluid, extracellular fluid, and cell membrane structures. Intracellular and extracellular fluids have conductive properties, while cell membranes exhibit dielectric properties under an alternating current electric field. When the water content, ion migration capacity, cell membrane integrity, and differences in fruit structure change within the tissue, the bioelectrical impedance also changes accordingly. Therefore, bioelectrical impedance can serve as supplementary information reflecting the internal state of the fruit.

[0004] However, existing research on bioelectrical impedance analysis of fruits focuses primarily on maturity identification, storage condition assessment, mechanical damage detection, or interpretation of equivalent circuit parameters. The sugar content of Wogan tangerines is influenced by multiple factors, including variety, batch, fruit shape, maturity, water content, tissue structure, and external phenotypic data. Relying on only one original measurement variable makes it difficult to fully characterize the patterns of sugar content variation.

[0005] Therefore, there is an urgent need for a non-destructive sugar content detection system and method that is more suitable for rapid detection of single Wogan mandarins. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a non-destructive detection method and system for the sugar content of Wogan mandarins based on the fusion of electrical impedance and multi-source features. This invention can utilize the external morphological information of Wogan mandarins and fully absorb the supplementary expressive ability of electrical impedance to the internal state of the fruit, thereby further improving the accuracy of Wogan mandarin sugar content prediction.

[0007] In a first aspect, the technical solution of the present invention is: a non-destructive detection method for the sugar content of Wogan oranges based on the fusion of electrical impedance and multi-source features, comprising: Place the Wogan orange to be tested on an insulating bearing surface, and set electrode patches in the equatorial region of the Wogan orange to obtain the impedance information of the Wogan orange; Obtain the external phenotypic data of the Wogan mandarin oranges to be tested, including weight, first diameter, second diameter, equatorial circumference, and fruit height. Preprocessing is performed on the electrical impedance information and external phenotypic data of Wogan oranges to obtain Wogan orange derived features. The Wogan orange derived features are then concatenated with the electrical impedance information and external phenotypic data of Wogan oranges to form a multi-source feature set. By inputting a multi-source feature set into a pre-trained nonlinear regression model for prediction, the predicted sugar content of Wogan oranges is obtained.

[0008] Preferably, the tangerine to be tested is placed on an insulating bearing surface, and electrode patches are set in the equatorial region of the tangerine to obtain the impedance information of the tangerine, including: Electrode patches are symmetrically attached to the four directions of the equatorial circumference of the Wogan orange to be tested: above, below, left, and right. Two of the electrode patches are used as excitation electrodes, and the other two electrode patches are used as sampling electrodes. An excitation signal with a set frequency is output through the impedance measurement chip AD5933. After being buffered, impedance matched, or having its driving capability enhanced by the voltage follower AD820, the excitation signal is applied to the inside of the Wogan tissue through the excitation electrode. The electrical response signal fed back by the Wogan tissue is collected in real time through the sampling electrode. The electrical response signal is amplified and converted from analog to digital to obtain a discrete time-domain signal. After the discrete time-domain signal is processed by discrete Fourier transform, the real part and imaginary part data of the response signal are output. The real part of the frequency domain transform result is extracted as the real part data of the response signal, and the imaginary part of the frequency domain transform result is extracted as the imaginary part data of the response signal. The real and imaginary parts of the response signal are read in real time through the IIC communication bus using the embedded microprocessor STM32F103C8T6 to calculate the signal amplitude. The gain coefficient is calculated by combining the standard impedance with the calibrated amplitude. Based on the gain coefficient and the measured signal amplitude, the impedance information of the measured Wogan is obtained. After multiple measurements, the average value of the measured Wogan orange impedance information is taken as the final Wogan orange impedance information. .

[0009] Preferably, the first fruit diameter is the diameter of the Wogan mandarin orange, measured with the line connecting the centers of the two excitation electrodes in the equatorial region as the reference.

[0010] Preferably, the second fruit diameter is the diameter of the Wogan mandarin orange, measured with the line connecting the centers of two sampling electrodes in the equatorial region as the measurement reference.

[0011] Preferably, the Wogan-derived features include average fruit diameter, fruit shape index, Wogan volume, density proxy, normalized electrical impedance, and Wogan logarithmic electrical impedance information spliced ​​together to form Wogan-derived features.

[0012] Preferably, the fruit shape index Based on average fruit diameter and fruit height Calculation, that is: .

[0013] Preferably, the density proxy Calculations based on the weight and volume of Wogan oranges, namely: ; In the formula, This refers to the weight of the Wogan oranges.

[0014] As a preferred approach, the normalized impedance information of the Wogan orange is used. First Fruit Path Second fruit diameter Fruit height Calculation, that is: ; In the formula, This is the normalized impedance.

[0015] Preferably, the multi-source feature set is input into a pre-trained nonlinear regression model for prediction to obtain the predicted sugar content of Wogan oranges, including: The nonlinear regression model mentioned is the CatBoost regression model. During training, multi-source feature training samples are input into the CatBoost regression model, and the predicted sugar content of Wogan oranges is calculated through a nonlinear mapping function, i.e.: ; In the formula, For the first Predicted sugar content of Wogan oranges from 10 multi-source feature training samples; For the first Multi-source feature training samples; These are the parameters of the CatBoost regression model; This is the nonlinear mapping function of the model; The loss function was calculated based on the predicted sugar content of Wogan oranges and the actual label. ,Right now: ; In the formula, The number of training samples for multi-source features; For the first The true labels of the multi-source feature training samples; Based on loss function The decision tree is iteratively constructed using gradient boosting, which gradually reduces the error between the model output value and the true label until the preset maximum number of iterations or the loss value converges, resulting in a trained CatBoost regression model. In actual testing, the multi-source feature set corresponding to the Wogan orange to be tested is input into the trained CatBoost regression model for prediction, and the predicted sugar content of the Wogan orange is obtained.

[0016] Secondly, the technical solution of the present invention is: a non-destructive testing system for the sugar content of Wogan oranges based on the fusion of electrical impedance and multi-source features, comprising: The impedance information acquisition module is used to place the Wogan orange under test on the insulating bearing surface and set electrode patches in the equatorial region of the Wogan orange to obtain the impedance information of the Wogan orange; The phenotypic data acquisition module is used to acquire the external phenotypic data of the tested Wogan mandarin orange, including its weight, first fruit diameter, second fruit diameter, equatorial circumference, and fruit height. The multi-source feature construction module is used to preprocess the electrical impedance information and external phenotypic data of Wogan oranges to obtain Wogan orange derived features, and then concatenate the Wogan orange derived features with the electrical impedance information and external phenotypic data of Wogan oranges to form a multi-source feature set. The prediction module is used to input the multi-source feature set into a pre-trained nonlinear regression model for prediction, and obtain the predicted sugar content of Wogan oranges.

[0017] Thirdly, the present invention provides an electronic device, comprising: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by at least one processor. When the computer program is executed by at least one processor, it implements the non-destructive detection method for the sugar content of Wogan oranges based on the fusion of electrical impedance and multi-source features.

[0018] The beneficial technical effects of this invention are as follows: 1. Based on electrical impedance information and external phenotypic data, this invention uses the CatBoost model to predict the sugar content of Wogan oranges, and further improves the prediction accuracy. 2. This invention collects electrical impedance information of Wogan oranges by setting electrode patches around the equatorial circumference, and calculates derived features by combining the external phenotypic data of Wogan oranges. Multi-source features are constructed by combining derived features with original features. The size, shape and density of the fruit are quantified by morphological derived features, and the interference of fruit size on electrical signals is reduced by electrical impedance transformation features, thereby achieving deep integration of electrical information and phenotypic information. 3. The CatBoost regression model used in this invention can effectively learn the complex mapping relationship between multi-source features and the sugar content of Wogan oranges, further improving the prediction accuracy; 4. This invention uses the STM32F103C8T6 microprocessor and the excitation signal of the AD5933 impedance measurement chip to acquire impedance information, and improves the transmission state of the AD5933 output excitation signal in the external circuit through a voltage follower, thereby reducing the impact of the impedance change of the measured sample on the stability of the measurement signal. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the method of an embodiment of the present invention; Figure 2 This is a schematic diagram of the electrode patch arrangement for Wogan oranges according to an embodiment of the present invention; Figure 3 This is a structural framework diagram of the system according to an embodiment of the present invention; Figure 4 This is a structural framework diagram of the impedance information acquisition module according to an embodiment of the present invention; In the figure, 10 is the electrical impedance information acquisition module; 20 is the phenotypic data acquisition module; 30 is the multi-source feature construction module; and 40 is the prediction module. Detailed Implementation

[0020] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings: like Figure 1 As shown, this embodiment provides a non-destructive testing method for the sugar content of Wogan oranges based on the fusion of electrical impedance and multi-source features, including: S1) Place the Wogan orange to be tested on the insulating bearing surface, and set the electrode patch in the equatorial region of the Wogan orange to obtain the impedance information of the Wogan orange; S2) Obtain the external phenotypic data of the Wogan mandarin orange to be tested, including weight, first fruit diameter, second fruit diameter, equatorial circumference, and fruit height; S3) Preprocess the electrical impedance information and external phenotypic data of Wogan oranges to obtain Wogan orange derived features, and then concatenate the Wogan orange derived features with the electrical impedance information and external phenotypic data of Wogan oranges to form a multi-source feature set; S4) Input the multi-source feature set into the pre-trained nonlinear regression model for prediction to obtain the predicted sugar content of Wogan oranges.

[0021] In this embodiment, step S1) involves placing the Wogan orange to be tested on an insulating bearing surface and setting an electrode patch in the equatorial region of the Wogan orange to obtain its impedance information, including: S11) Symmetrically attach corresponding electrode patches at the four directions of the equatorial circumference of the Wogan orange to be tested: top, bottom, left, and right. Two of the electrode patches serve as excitation electrodes, and the other two serve as sampling electrodes.

[0022] S12) An excitation signal of a set frequency is output through the impedance measurement chip AD5933. This excitation signal is buffered, impedance matched, or has its driving capability enhanced by the voltage follower AD820 before being applied to the Wogan tissue through the excitation electrode. The electrical response signal fed back by the Wogan tissue is acquired in real time through the sampling electrode. This electrical response signal is sequentially amplified and converted from analog to digital to obtain a discrete-time signal. After being processed by a discrete Fourier transform, the real and imaginary parts of the response signal are output. This includes: ; In the formula, The result is a frequency domain transform. It is a discrete-time signal; This represents the total number of sampling points; The imaginary unit; This refers to the frequency domain sampling sequence number; The time-domain sampling point number; Extracting frequency domain transform results The real part is used as the real part data of the response signal, and the frequency domain transformation result is extracted. The imaginary part is used as the imaginary part data of the response signal, that is: ; ; In the formula, The operator represents the operation for retrieving the real part of a complex number; The operator represents the operation for retrieving the imaginary part of a complex number; In this embodiment, the set frequency is 50 kHz, and the voltage of the excitation signal is 2V.

[0023] S13) The real and imaginary data of the response signal are read in real time through the IIC communication bus by the embedded microprocessor STM32F103C8T6 to calculate the signal amplitude. The gain coefficient is calculated by combining the standard impedance with the calibrated amplitude. Based on the gain coefficient and the signal amplitude measured in this measurement, the impedance information of the measured Wogan is obtained. In this embodiment, the signal amplitude The calculation formula is expressed as: ; In the formula, , These are the real and imaginary parts of the response signal, respectively. The gain factor is calculated by combining the standard impedance with the calibrated amplitude. ,Right now: ; In the formula, Standard impedance; The calibrated amplitude; The impedance information of the Wogan orange measured in this study is calculated based on the gain coefficient and the signal amplitude of this measurement, namely: ; In the formula, This provides the electrical impedance information for the Wogan oranges measured in this study. This is the amplitude of the Wogan signal measured in this study.

[0024] In this embodiment, after multiple measurements, the average value of the measured Wogan orange impedance information is taken as the final Wogan orange impedance information. .

[0025] In this embodiment, in step S2), the first fruit diameter The diameter of the Wogan orange is measured using the line connecting the centers of the two excitation electrodes in the equatorial region as the reference.

[0026] The second fruit diameter The diameter of the Wogan orange is measured using the line connecting the centers of two sampling electrodes in the equatorial region as the reference.

[0027] In this embodiment, in step S3), the electrical impedance information and external phenotypic data of the Wogan orange are preprocessed to obtain Wogan orange derived features, and the Wogan orange derived features are concatenated with the Wogan orange electrical impedance information and external phenotypic data to form a multi-source feature set, including: S31), based on the first fruit path Second fruit diameter Calculate the average fruit diameter ,Right now: ; S32), based on average fruit diameter and fruit height Calculate the fruit shape index ,Right now: ; S33) When the Wogan orange is approximated as an ellipsoid, calculate the volume of the Wogan orange. ,Right now: ; S34), Density Calculation Agent Based on Wogan Orange Weight and Volumetric Calculation ,Right now: ; In the formula, This refers to the weight of the Wogan oranges.

[0028] S35), based on the impedance information of Wogan oranges First Fruit Path Second fruit diameter Fruit height Calculate the normalized impedance, i.e.: ; In the formula, This is the normalized impedance.

[0029] S36), information on the impedance of Wogan oranges Perform a logarithmic transformation to obtain the logarithmic impedance information of the Wogan orange, log10(Z).

[0030] S37), average fruit diameter Fruit shape index Wogan orange volume Density proxy Normalized impedance The logarithmic impedance information log10(Z) of Wogan oranges is concatenated to form the derived features of Wogan oranges. ,Right now: ; S38) Combine the electrical impedance information and external phenotypic data of the Wogan orange into the original features. ,Right now: ; S39), Derivative characteristics of Wogan mandarin and original features Concatenate into a multi-source feature set ,Right now: .

[0031] During the model training phase, the true sugar content label was obtained using an electronic saccharimeter. After collecting electrical impedance information and external phenotypic data, the same Wogan mandarin orange sample was peeled, and four non-adjacent segments were juiced, mixed, filtered, and then the sugar content was measured using an electronic saccharimeter. For the same sample, three consecutive measurements were taken, and the average value was used as the true sugar content label for that sample. The above sugar content measurement process was only used for training sample labeling and model validation; in the actual application phase, the Wogan mandarin oranges to be tested do not need to be peeled, segmented, or juiced.

[0032] In this embodiment, step S4) involves inputting the multi-source feature set into a pre-trained nonlinear regression model for prediction to obtain the predicted sugar content of the Wogan orange, including: The nonlinear regression model mentioned is the CatBoost regression model. During training, multi-source feature training samples are input into the CatBoost regression model, and the predicted sugar content of Wogan oranges is calculated through a nonlinear mapping function, i.e.: ; In the formula, For the first Predicted sugar content of Wogan oranges from 10 multi-source feature training samples; For the first Multi-source feature training samples; These are the parameters of the CatBoost regression model; This is the nonlinear mapping function of the model; The loss function was calculated based on the predicted sugar content of Wogan oranges and the actual label. ,Right now: ; In the formula, The number of training samples for multi-source features; For the first The true labels of the multi-source feature training samples.

[0033] Based on loss function The decision tree is built iteratively through gradient boosting, which gradually reduces the error between the model output value and the true label until the preset maximum number of iterations or the loss value converges, thus obtaining a trained CatBoost regression model.

[0034] In actual testing, the multi-source feature set corresponding to the Wogan orange to be tested is input into the trained CatBoost regression model for prediction, and the predicted sugar content of the Wogan orange is obtained.

[0035] In some embodiments, the nonlinear regression model may be a random forest, XGBoost, support vector regression, multilayer perceptron, or other regression models that can establish a nonlinear mapping relationship between a multi-source feature set and the sugar content of Wogan oranges.

[0036] like Figure 3 As shown, the second embodiment of this application provides a non-destructive testing system for the sugar content of Wogan oranges based on the fusion of electrical impedance and multi-source features, comprising: The impedance information acquisition module 10 is used to place the Wogan orange under test on the insulating bearing surface and set electrode patches in the equatorial region of the Wogan orange to obtain the impedance information of the Wogan orange. The phenotypic data acquisition module 20 is used to acquire the external phenotypic data of the Wogan mandarin orange under test, including weight, first fruit diameter, second fruit diameter, equatorial circumference and fruit height. The multi-source feature construction module 30 is used to preprocess the electrical impedance information and external phenotypic data of Wogan oranges to obtain Wogan orange derived features, and to concatenate the Wogan orange derived features with the electrical impedance information and external phenotypic data of Wogan oranges to form a multi-source feature set; The prediction module 40 is used to input the multi-source feature set into a pre-trained nonlinear regression model for prediction, and obtain the predicted sugar content of Wogan oranges.

[0037] In this embodiment, the process by which the impedance information acquisition module 10 acquires the impedance information of the Wogan orange is as follows: like Figure 4 As shown, corresponding electrode patches are symmetrically attached to the four directions of the equatorial circumference of the Wogan orange to be tested: top, bottom, left, and right. Two of the electrode patches serve as excitation electrodes, and the other two serve as sampling electrodes. An excitation signal with a set frequency is output through an impedance measurement chip AD5933. This excitation signal is buffered, impedance matched, or has its driving capability enhanced by a voltage follower AD820 before being applied to the Wogan tissue through excitation electrodes. The electrical response signal from the Wogan tissue is acquired in real time through sampling electrodes. This electrical response signal is amplified and converted from analog to digital to obtain a discrete-time signal. After discrete Fourier transform processing, the real and imaginary parts of the response signal are output. This includes: ; In the formula, The result is a frequency domain transform. It is a discrete-time signal; This represents the total number of sampling points; The imaginary unit; This refers to the frequency domain sampling sequence number; The time-domain sampling point number; Extracting frequency domain transform results The real part is used as the real part data of the response signal, and the frequency domain transformation result is extracted. The imaginary part is used as the imaginary part data of the response signal, that is: ; ; In the formula, The operator represents the operation for retrieving the real part of a complex number; The operator represents the operation for retrieving the imaginary part of a complex number; In this embodiment, the set frequency is 50 kHz, and the voltage of the excitation signal is 2V.

[0038] The real and imaginary parts of the response signal are read in real time through the IIC communication bus using the embedded microprocessor STM32F103C8T6 to calculate the signal amplitude. The gain coefficient is calculated by combining the standard impedance with the calibrated amplitude. Based on the gain coefficient and the measured signal amplitude, the impedance information of the measured Wogan is obtained. In this embodiment, the signal amplitude The calculation formula is expressed as: ; In the formula, , These are the real and imaginary parts of the response signal, respectively. The gain factor is calculated by combining the standard impedance with the calibrated amplitude. ,Right now: ; In the formula, Standard impedance; The calibrated amplitude; The impedance information of the Wogan orange measured in this study is calculated based on the gain coefficient and the signal amplitude of this measurement, namely: ; In the formula, This provides the electrical impedance information for the Wogan oranges measured in this study. This is the amplitude of the Wogan signal measured in this study.

[0039] In this embodiment, after multiple measurements, the average value of the measured Wogan orange impedance information is taken as the final Wogan orange impedance information. .

[0040] In this embodiment, the first fruit diameter The diameter of the Wogan orange is measured using the line connecting the centers of the two excitation electrodes in the equatorial region as the reference.

[0041] The second fruit diameter The diameter of the Wogan orange is measured using the line connecting the centers of two sampling electrodes in the equatorial region as the reference.

[0042] In this embodiment, the multi-source feature construction module 30 preprocesses the electrical impedance information and external phenotypic data of the Wogan orange to obtain Wogan orange derived features, and concatenates the Wogan orange derived features with the Wogan orange electrical impedance information and external phenotypic data to form a multi-source feature set, including: Based on the first fruit path Second fruit diameter Calculate the average fruit diameter ,Right now: ; Based on average fruit diameter and fruit height Calculate the fruit shape index ,Right now: ;

[0043] When the Wogan orange is approximated as an ellipsoid, the volume of the Wogan orange is calculated. ,Right now: ; Density Calculation Agent Based on Weight and Volume of Wogan Tangerine ,Right now: ; In the formula, This refers to the weight of the Wogan oranges.

[0044] Based on the electrical impedance information of Wogan oranges First Fruit Path Second fruit diameter Fruit height Calculate the normalized impedance, i.e.: ; In the formula, This is the normalized impedance.

[0045] Information on the impedance of Wogan oranges Perform a logarithmic transformation to obtain the logarithmic impedance information of the Wogan orange, log10(Z).

[0046] Average fruit diameter Fruit shape index Wogan orange volume Density proxy Normalized impedance The logarithmic impedance information log10(Z) of Wogan oranges is concatenated to form the derived features of Wogan oranges. ,Right now: ; The electrical impedance information and extrinsic phenotypic data of Wogan oranges were combined to form the original features. ,Right now: ; Derivative characteristics of Wogan mandarin and original features Concatenate into a multi-source feature set ,Right now: .

[0047] In this embodiment, the prediction module 40 inputs a multi-source feature set into a pre-trained nonlinear regression model for prediction, obtaining the predicted sugar content of the Wogan oranges. This includes a model training phase and a prediction phase, as detailed below: The nonlinear regression model mentioned is the CatBoost regression model. During training, multi-source feature training samples are input into the CatBoost regression model, and the predicted sugar content of Wogan oranges is calculated through a nonlinear mapping function, i.e.: ; In the formula, For the first Predicted sugar content of Wogan oranges from 10 multi-source feature training samples; For the first Multi-source feature training samples; These are the parameters of the CatBoost regression model; This is the nonlinear mapping function of the model; The loss function was calculated based on the predicted sugar content of Wogan oranges and the actual label. ,Right now: ; In the formula, The number of training samples for multi-source features; For the first The true labels of the multi-source feature training samples.

[0048] Based on loss function The decision tree is built iteratively through gradient boosting, which gradually reduces the error between the model output value and the true label until the preset maximum number of iterations or the loss value converges, thus obtaining a trained CatBoost regression model.

[0049] In actual testing, the multi-source feature set corresponding to the Wogan orange to be tested is input into the trained CatBoost regression model for prediction, and the predicted sugar content of the Wogan orange is obtained.

[0050] In some embodiments, the nonlinear regression model may be a random forest, XGBoost, support vector regression, multilayer perceptron, or other regression models that can establish a nonlinear mapping relationship between a multi-source feature set and the sugar content of Wogan oranges.

[0051] A third embodiment of this application provides an electronic device, comprising: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by at least one processor. When executed by at least one processor, the computer program receives electrical impedance information of Wogan mandarins and external phenotypic data such as weight, first fruit diameter, second fruit diameter, equatorial circumference and fruit height, constructs a multi-source feature set, inputs the multi-source feature set into a pre-trained nonlinear regression model, and outputs a predicted value of Wogan mandarin sugar content.

[0052] In this embodiment, the memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. A processor, coupled to the memory, is used to execute computer programs stored in the memory.

[0053] The embodiments and descriptions above are merely illustrative of the principles and preferred embodiments of the present invention. Various changes and modifications may be made to the present invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed.

Claims

1. A non-destructive method for detecting the sugar content of Wogan oranges based on the fusion of electrical impedance and multi-source features, characterized in that, include: Place the Wogan orange to be tested on an insulating bearing surface, and set electrode patches in the equatorial region of the Wogan orange to obtain the impedance information of the Wogan orange; Obtain the external phenotypic data of the Wogan mandarin oranges to be tested, including weight, first diameter, second diameter, equatorial circumference, and fruit height. Preprocessing is performed on the electrical impedance information and external phenotypic data of Wogan oranges to obtain Wogan orange derived features. The Wogan orange derived features are then concatenated with the electrical impedance information and external phenotypic data of Wogan oranges to form a multi-source feature set. By inputting a multi-source feature set into a pre-trained nonlinear regression model for prediction, the predicted sugar content of Wogan oranges is obtained.

2. The non-destructive testing method for the sugar content of Wogan oranges based on the fusion of electrical impedance and multi-source features according to claim 1, characterized in that, The Wogan orange to be tested is placed on an insulating bearing surface, and electrode patches are set in the equatorial region of the Wogan orange to obtain the impedance information of the Wogan orange, including: Electrode patches are symmetrically attached to the four directions of the equatorial circumference of the Wogan orange to be tested: above, below, left, and right. Two of the electrode patches are used as excitation electrodes, and the other two electrode patches are used as sampling electrodes. An excitation signal with a set frequency is output through an impedance measurement chip AD5933. After being buffered, impedance matched, or having its driving capability enhanced by a voltage follower AD820, the excitation signal is applied to the inside of the Wogan tissue through excitation electrodes. The electrical response signal fed back by the Wogan tissue is collected in real time through sampling electrodes. The electrical response signal is amplified and converted from analog to digital to obtain a discrete time-domain signal. After the discrete time-domain signal is processed by discrete Fourier transform, the real part and imaginary part data of the response signal are output. The real part of the frequency domain transform result is extracted as the real part data of the response signal, and the imaginary part of the frequency domain transform result is extracted as the imaginary part data of the response signal. The real and imaginary parts of the response signal are read in real time through the IIC communication bus using the embedded microprocessor STM32F103C8T6 to calculate the signal amplitude. The gain coefficient is calculated by combining the standard impedance with the calibrated amplitude. Based on the gain coefficient and the measured signal amplitude, the impedance information of the measured Wogan is obtained. After multiple measurements, the average value of the measured Wogan orange impedance information is taken as the final Wogan orange impedance information. .

3. The non-destructive testing method for the sugar content of Wogan oranges based on the fusion of electrical impedance and multi-source features according to claim 1, characterized in that, The first fruit diameter is the diameter of the Wogan mandarin orange, measured with the line connecting the centers of the two excitation electrodes in the equatorial region as the measurement reference. The second fruit diameter is the diameter of the Wogan mandarin orange, measured with the line connecting the centers of the two sampling electrodes in the equatorial region as the measurement reference. The equatorial circumference is the circumferential length measured around the fruit body along the equatorial region of the Wogan mandarin orange, and the equatorial region is the transverse measurement area of ​​the fruit body where the four electrode patches are located.

4. The non-destructive testing method for the sugar content of Wogan oranges based on the fusion of electrical impedance and multi-source features according to claim 1, characterized in that, The derived characteristics of the Wogan mandarin include average fruit diameter, fruit shape index, Wogan mandarin volume, density profile, normalized electrical impedance, and Wogan mandarin logarithmic electrical impedance information. The multi-source feature set is obtained by splicing together the electrical impedance information, weight, first fruit diameter, second fruit diameter, equatorial circumference, fruit height, and the derived features of the Wogan mandarin orange.

5. The non-destructive testing method for the sugar content of Wogan oranges based on the fusion of electrical impedance and multi-source features according to claim 4, characterized in that, The average fruit diameter is calculated based on the first fruit diameter and the second fruit diameter, that is: In the formula, For the first fruit diameter, The second fruit diameter, This represents the average fruit diameter.

6. The non-destructive testing method for the sugar content of Wogan oranges based on the fusion of electrical impedance and multi-source features according to claim 5, characterized in that, The fruit shape index It is calculated based on the average fruit diameter and fruit height, that is: ； In the formula, The height of the fruit; Average fruit diameter; The non-destructive detection method for the sugar content of Wogan oranges based on the fusion of electrical impedance and multi-source features is characterized in that the density proxy... The calculations are based on the weight and volume of the Wogan oranges, namely: ； In the formula, This refers to the weight of the Wogan oranges; This refers to the volume of the Wogan orange.

7. The non-destructive testing method for the sugar content of Wogan oranges based on the fusion of electrical impedance and multi-source features according to claim 6, characterized in that, The normalized impedance is based on the impedance information of Wogan oranges. First Fruit Path Second fruit diameter Fruit height The calculation yields the following result: ； In the formula, This is the normalized impedance.

8. The non-destructive testing method for the sugar content of Wogan oranges based on the fusion of electrical impedance and multi-source features according to claim 1, characterized in that, The multi-source feature set is input into a pre-trained nonlinear regression model for prediction, resulting in a predicted sugar content value for Wogan oranges, including: The nonlinear regression model mentioned is the CatBoost regression model. During training, multi-source feature training samples are input into the CatBoost regression model, and the predicted sugar content of Wogan oranges is calculated through a nonlinear mapping function, i.e.: ； In the formula, For the first Predicted sugar content of Wogan oranges from 10 multi-source feature training samples; For the first One multi-source feature training sample; These are the parameters of the CatBoost regression model; This is the nonlinear mapping function of the model; The loss function was calculated based on the predicted sugar content of Wogan oranges and the actual label. ,Right now: ； In the formula, The number of training samples for multi-source features; For the first The true labels of the multi-source feature training samples; Based on loss function The decision tree is iteratively constructed using gradient boosting, which gradually reduces the error between the model output value and the true label until the preset maximum number of iterations or the loss value converges, resulting in a trained CatBoost regression model. In actual testing, the multi-source feature set corresponding to the Wogan orange to be tested is input into the trained CatBoost regression model for prediction, and the predicted sugar content of the Wogan orange is obtained.

9. A non-destructive testing system for the sugar content of Wogan oranges based on the fusion of electrical impedance and multi-source features, characterized in that, include: The impedance information acquisition module is used to place the Wogan orange under test on the insulating bearing surface and set electrode patches in the equatorial region of the Wogan orange to obtain the impedance information of the Wogan orange; The phenotypic data acquisition module is used to acquire the external phenotypic data of the tested Wogan mandarin orange, including its weight, first fruit diameter, second fruit diameter, equatorial circumference, and fruit height. The multi-source feature construction module is used to preprocess the electrical impedance information and external phenotypic data of Wogan oranges to obtain Wogan orange derived features, and then concatenate the Wogan orange derived features with the electrical impedance information and external phenotypic data of Wogan oranges to form a multi-source feature set. The prediction module is used to input the multi-source feature set into a pre-trained nonlinear regression model for prediction, and obtain the predicted sugar content of Wogan oranges.

10. An electronic device, comprising: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by at least one processor, characterized in that, when the computer program is executed by at least one processor, it implements the non-destructive detection method for sugar content of Wogan oranges based on impedance and multi-source feature fusion as described in any one of claims 1-8.

Images