A nondestructive detection method and system for apple quality based on photoacoustic biphasic spectrum

Abstract

The application discloses a kind of based on photoacoustic two-phase spectrum apple quality nondestructive testing method and system, belong to agricultural product nondestructive testing technical field, including: the near-infrared absorption spectrum of collection apple sample and the time-domain vibration signal of stimulated vibration, vibration spectrum is obtained by fast fourier transform;Spectrum is pretreated;Based on the hardness standard value and the soluble solid content standard value of apple, the characteristic signal related to hardness and soluble solid content is screened;The screened optical characteristics and acoustic characteristics are fused using data level fusion, feature level fusion or decision level fusion mode, to obtain photoacoustic two-phase fusion characteristics;The fusion characteristics are input into quality prediction model, and the hardness prediction value and the soluble solid content prediction value of apple are output.The application can simultaneously, nondestructively detect the texture properties and chemical components of apple, with high detection precision, to provide a reliable basis for comprehensive evaluation of apple maturity.

Description

Technical Field

[0001] This invention belongs to the field of non-destructive testing technology for agricultural products, and particularly relates to a non-destructive testing method and system for apple quality based on photoacoustic dual-phase spectroscopy. Background Technology

[0002] Currently, non-destructive testing technologies for apple quality mainly include physicochemical analysis, machine vision, single near-infrared spectroscopy, and acoustic vibration methods. Physicochemical analysis (such as starch-iodine staining) can accurately determine ripeness, but it is a destructive test and cannot meet the needs of online, batch testing. Machine vision judges ripeness based on appearance and color, but it only obtains surface information and cannot reflect internal quality changes. Single near-infrared spectroscopy can detect soluble solids and other chemical components, and has the advantages of being non-destructive and rapid, but its accuracy in detecting texture properties such as hardness and brittleness is poor. Acoustic vibration methods are sensitive to physical characteristics such as fruit hardness, and are simple to operate and low in cost. In recent years, multi-source information fusion technology has been gradually applied to fruit quality testing, such as the fusion of acoustic vibration and near-infrared spectroscopy (see CN201910273251.X, CN202410761858.3), which can improve detection accuracy to a certain extent and shows good application prospects in areas such as fruit firmness measurement and internal lesion identification.

[0003] However, existing fusion detection technologies still have the following shortcomings: First, the detection indicators are relatively singular, mainly targeting texture indicators such as hardness or firmness, failing to simultaneously achieve a comprehensive characterization of texture properties such as hardness and brittleness, as well as chemical components such as soluble solids and pectin, making it difficult to meet the needs of multi-dimensional evaluation of apple maturity. Second, the fusion strategies are relatively simple, with most methods only employing data-level or decision-level fusion, lacking a systematic comparison and optimization mechanism for multiple fusion strategies at the data-level, feature-level, and decision-level. Third, existing methods do not consider the dual interference of the peel on near-infrared spectroscopy and acoustic signals, lacking feature correction methods for the peel's influence. Fourth, in terms of maturity grading, there is a lack of threshold verification methods based on theories such as maximum entropy and maximum inter-class variance, resulting in insufficient objectivity and accuracy in maturity discrimination. Therefore, how to achieve high-precision simultaneous non-destructive detection of apple internal quality (chemical composition) and texture quality (physical properties), and establish a scientific and systematic multi-source information fusion and maturity discrimination system, remains an urgent technical problem to be solved. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a non-destructive testing method and system for apple quality based on photoacoustic dual-phase spectroscopy, thereby resolving the issues present in the prior art.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a non-destructive testing method for apple quality based on photoacoustic dual-phase spectroscopy, comprising: Near-infrared absorption spectra and time-domain vibration signals during stimulated vibration of apple samples were collected, and the time-domain vibration signals were subjected to fast Fourier transform to obtain the vibration spectrum. The near-infrared absorption spectrum and the vibrational spectrum are preprocessed; Based on the standard values ​​of apple hardness and soluble solids content, feature signals related to hardness and soluble solids content in the near-infrared absorption spectrum and the vibration spectrum are screened. The selected optical feature signals and acoustic feature signals are fused using data-level fusion, feature-level fusion, or decision-level fusion methods to obtain photoacoustic two-phase fusion features; The photoacoustic two-phase fusion features are input into a pre-established quality prediction model, which outputs the predicted values ​​of apple hardness and soluble solids content.

[0006] Preferably, the preprocessing includes correcting the near-infrared absorption spectrum and the vibrational spectrum using one or more of the following: normalized ratio algorithm, multi-source scattering correction, normal variable transformation, first derivative, and second derivative.

[0007] Preferably, when screening feature signals, one or more of the following methods are used: correlation analysis, continuous projection algorithm, competitive adaptive resampling algorithm, iterative population analysis of information-preserving variables or combinations of variables.

[0008] Preferably, before fusing the screened optical and acoustic feature signals, a step of removing the influence of the fruit peel is included: Near-infrared spectral and acoustic characteristics of whole apples and peeled apples were collected separately. The difference between the two was used to obtain a data matrix. Singular value decomposition was performed on the data matrix to extract the projection vectors in the peel, and a spatial projection matrix was constructed. The feature information of the whole apples was corrected using the spatial projection matrix to obtain the feature information after removing the influence of the peel.

[0009] Preferably, the data-level fusion involves directly splicing the original near-infrared absorption spectrum data and the original time-domain vibration data; the feature-level fusion involves extracting feature vectors from the near-infrared absorption spectrum and the vibration spectrum respectively and then splicing them; the decision-level fusion involves establishing a spectral prediction model and a vibration prediction model respectively and then taking a weighted average of the output results of the two models.

[0010] Preferably, the quality prediction model is a principal component regression model, a multilayer perceptron model, or a support vector machine regression model.

[0011] Preferably, it also includes a maturity discrimination threshold verification step: determining the maturity classification threshold based on the maximum inter-class variance method or the maximum entropy method; The formula for the maximum inter-class variance is: ; in, q otsu,th When the threshold is th The inter-class variance of the two types of data; u It is the average of the sample predicted data, which can be derived from... Calculated; u 0 and u 1 represents the average of the first and second category sample data; h i This is the predicted maturity value of the sample; n The total number of samples; l The number of samples in the first category; The formula for the maximum entropy method is: ; in, q entropy,th It is the sum of the entropy values ​​of the two types of data when the threshold is selected as th; p t The probability that the predicted maturity value of the apple sample belongs to t; P 1 represents the first type of probability, which can be derived from... The calculation yields the result; the symbol ln represents the logarithmic transformation; t represents the range of the predicted maturity values ​​for the apple sample, which is determined by the number of iterations and the iteration step size.

[0012] Preferably, the excited vibration is applied by a vibration excitation device, and the frequency range of the sweep signal output by the vibration excitation device is 200Hz to 1200Hz, and the sweep time is 3 seconds.

[0013] Preferably, the near-infrared absorption spectrum is collected in a wavelength range that covers the characteristic absorption bands of apple soluble solids and pectin.

[0014] Secondly, the present invention also discloses an apple quality non-destructive testing system based on photoacoustic two-phase spectrum, including: a photoacoustic two-phase spectrum acquisition device and photo-vibration two-phase fruit and vegetable quality non-destructive testing software; The photoacoustic phase spectrum acquisition device includes a support unit for placing apple samples, a near-infrared spectrum acquisition unit for acquiring near-infrared absorption spectra, a vibration spectrum acquisition unit for acquiring time-domain vibration signals, and a control and communication unit for coordinated control and data synchronous transmission. The optical-vibration two-phase fruit and vegetable quality non-destructive testing software includes a parameter setting module, a hardware driver module, a data processing and display module, and a device log module, and the optical-vibration two-phase fruit and vegetable quality non-destructive testing software is configured to execute the method described in the first aspect.

[0015] Compared with the prior art, the present invention has the following advantages and technical effects: This invention achieves simultaneous, non-destructive detection of both the texture properties (firmness) and chemical composition (soluble solids content) of apples by simultaneously acquiring the near-infrared absorption spectrum and the stimulated vibration spectrum of apples, and then filtering characteristic signals based on standard values ​​for firmness and soluble solids content, respectively. This technical solution overcomes the limitation of a single information source in accurately characterizing both internal chemical composition and texture properties simultaneously, providing multi-dimensional quality indicators for the comprehensive assessment of apple maturity.

[0016] This invention employs data-level fusion, feature-level fusion, or decision-level fusion methods to fuse optical and acoustic feature signals, forming a photoacoustic two-phase fusion feature. This multi-strategy fusion mechanism can select the optimal fusion path based on different quality indicators, significantly improving the model's predictive adaptability and accuracy for hardness and soluble solids content compared to single-modal or fixed fusion methods.

[0017] This invention effectively reduces noise interference and redundant information in the original signal by preprocessing the near-infrared absorption spectrum and vibration spectrum, combined with a feature signal screening step based on standard values. This technical feature makes the features of the input quality prediction model more representative and robust, thereby improving the accuracy and stability of the predicted values ​​for hardness and soluble solids content.

[0018] This invention inputs the fused photoacoustic two-phase characteristics into a pre-established quality prediction model, simultaneously outputting predicted hardness and soluble solids content values, achieving parallel output of multiple indicators. This scheme simplifies the testing process, eliminating the need to establish multiple single-indicator models separately, and improving the efficiency and practicality of non-destructive testing of apple quality. Attached Figure Description

[0019] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of a non-destructive testing method for apple quality based on photoacoustic dual-phase spectroscopy, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the apple quality prediction model structure according to an embodiment of the present invention; Figure 3 This is a schematic diagram of an apple photoacoustic dual-phase acquisition system according to an embodiment of the present invention; Figure 4 This is a three-dimensional diagram of the system device according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the apple photoacoustic two-phase spectrum acquisition software interface according to an embodiment of the present invention. Detailed Implementation

[0020] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0021] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0022] Example 1 like Figure 1 As shown, this embodiment provides a non-destructive testing method for apple quality based on photoacoustic dual-phase spectroscopy, including: S1: Collect the near-infrared absorption spectrum and time-domain vibration signal of apple samples under excited vibration, and perform fast Fourier transform on the time-domain vibration signal to obtain the vibration spectrum; Furthermore, the excited vibration is applied by a vibration excitation device, the frequency range of the sweep signal output by the vibration excitation device is 200Hz to 1200Hz, and the sweep time is 3 seconds.

[0023] Furthermore, the near-infrared absorption spectrum is collected in a wavelength range that covers the characteristic absorption bands of apple soluble solids and pectin.

[0024] Specifically, in this embodiment, the near-infrared absorption spectrum and time-domain vibration signal of the apple sample are acquired by controlling the photoacoustic phase spectrum acquisition device; the vibration spectrum is obtained by performing a fast Fourier transform on the time-domain vibration signal.

[0025] S2: Preprocess the near-infrared absorption spectrum and the vibrational spectrum; Furthermore, the preprocessing includes correcting the near-infrared absorption spectrum and the vibrational spectrum using one or more of the following: normalized ratio algorithm, multi-source scattering correction, normal variable transformation, first derivative, and second derivative.

[0026] S3: Based on the standard values ​​of apple hardness and soluble solids content, screen the feature signals related to hardness and soluble solids content in the near-infrared absorption spectrum and the vibration spectrum; Furthermore, when screening feature signals, one or more of the following methods are employed: correlation analysis, continuous projection algorithm, competitive adaptive resampling algorithm, iterative population analysis of information-preserving variables or combinations of variables.

[0027] Specifically, this embodiment uses the material basis (including texture property parameters and chemical composition content) as the standard value, and uses one or more of the following methods: Correlation Analysis (CA), Successive Projections Algorithm (SPA), Competitive Adaptive Reweighted Sampling (CARS), Iteratively Retains Informative Variables (IRIV), and Variable Combination Population Analysis (VCPA) to screen characteristic signals related to maturity in the near-infrared absorption spectrum and acoustic frequency domain spectrum.

[0028] S4: The selected optical feature signals and acoustic feature signals are fused using data-level fusion, feature-level fusion, or decision-level fusion methods to obtain photoacoustic two-phase fusion features; Furthermore, before fusing the selected optical and acoustic feature signals, a step is included to remove the influence of the fruit peel: Near-infrared spectral and acoustic characteristics of whole apples and peeled apples were collected separately. The difference between the two was used to obtain a data matrix. Singular value decomposition was performed on the data matrix to extract the projection vectors in the peel, and a spatial projection matrix was constructed. The feature information of the whole apples was corrected using the spatial projection matrix to obtain the feature information after removing the influence of the peel.

[0029] Furthermore, the data-level fusion involves directly concatenating the original near-infrared absorption spectrum data and the original time-domain vibration data; the feature-level fusion involves extracting feature vectors from the near-infrared absorption spectrum and the vibration spectrum respectively and then concatenating them; and the decision-level fusion involves establishing a spectral prediction model and a vibration prediction model respectively and then performing a weighted average of the output results of the two models.

[0030] Specifically, the near-infrared spectral-acoustic characteristic information selected is optimized. After collecting the near-infrared spectral-acoustic characteristics of the complete apple, the apple is peeled, and the near-infrared spectral-acoustic characteristic information of the peeled apple is collected. The difference between the two is used to obtain the data matrix D, and then the matrix D is decomposed by SVD (Equation (1)) to extract the projection vector in the peel: (1) Where T represents the transpose of the matrix; s and f represent the number of valid information and noise factors, respectively. Using V in formula (1) s The submatrices are used to construct the spatial projection matrix, i.e., EV. sV s T Substitute the collected complete apple feature information (X1) into equation (2): (2) Where X1 represents the extracted near-infrared spectral or acoustic characteristics of apples; E is the identity matrix; and X2 represents the characteristics after removing the influence of the peel.

[0031] The optimized optical feature signals are fused with the acoustic feature signals to obtain photoacoustic two-phase fusion features. The fusion method includes data-level fusion, feature-level fusion, or decision-level fusion.

[0032] S5: Input the photoacoustic two-phase fusion feature into the pre-established quality prediction model, and output the predicted values ​​of apple hardness and soluble solids content.

[0033] Furthermore, the quality prediction model is a principal component regression model, a multilayer perceptron model, or a support vector machine regression model.

[0034] Furthermore, it also includes a maturity discrimination threshold verification step: determining the maturity classification threshold based on the maximum inter-class variance method or the maximum entropy method; Specifically, based on the fused feature information, prediction models such as partial least squares regression (PLSR), multi-layer perceptron (MLP), and support vector machine regression (SVR) are established. The prediction effects of different modeling methods on apple quality are compared and analyzed to identify the best modeling method.

[0035] like Figure 2 As shown in the diagram, the apple quality prediction model is a multimodal fusion deep learning model based on a one-dimensional convolutional neural network (1D-CNN) and optimal transport (OT) alignment. Designed for industrial sample quality inspection scenarios, it simultaneously achieves dual-task outputs: regression prediction of physicochemical indicators and multi-class classification of quality grades. The model uses optical absorption spectrum data and vibration response spectrum data as dual-modal inputs. Through a dual-branch encoder, feature alignment and pyramid fusion modules, and dual-task output heads, it completes end-to-end feature extraction, cross-modal fusion, and multi-task inference. The overall architecture presents a progressive structure of "input-encoding-alignment-fusion-output".

[0036] Based on the theories of maximum entropy and maximum inter-class variance, a maturity discrimination threshold verification algorithm is created to verify the optimal threshold and the threshold selection method: the maximum inter-class variance method. The predicted and actual values ​​of models at different maturity levels are used as the basis for threshold determination, and the maximum inter-class variance is used as the threshold determination criterion.

[0037] The formula for the maximum inter-class variance is: (3) in, q otsu,th When the threshold is th The inter-class variance of the two types of data; u It is the average of the sample predicted data, which can be derived from... Calculated; u 0 and u 1 represents the average of the first and second category sample data; h i This is the predicted maturity value of the sample; n The total number of samples; l The number of samples in the first category; Maximum Entropy Method: Calculate the entropy values ​​of the first and second classes of data. When the sum of the entropy values ​​of the two classes is maximized, the threshold of the two classes of data can be obtained. The sum of the entropy values ​​of the two classes of data can be calculated by equation (4): (4) in, q entropy,th It is the sum of the entropy values ​​of the two types of data when the threshold is selected as th; p t The probability that the predicted maturity value of the apple sample belongs to t; P 1 represents the first type of probability, which can be derived from... The calculation yields the result; the symbol ln represents the logarithmic transformation; t represents the range of the predicted maturity values ​​for the apple sample, which is determined by the number of iterations and the iteration step size.

[0038] By combining maturity time and sensory evaluation scores, the effectiveness of threshold establishment using maximum entropy, maximum inter-class variance, and traditional receiver operating characteristic (ROC) curves was compared to confirm the optimal threshold and threshold selection method. The maturity prediction model and threshold were validated based on the correlation coefficient, root mean square error, detection accuracy, sensitivity, and specificity of the prediction set data. The photoacoustic fusion features of the samples were input into the trained model, outputting predicted values ​​for quality indicators.

[0039] Example 2 Based on the same inventive concept, such as Figure 3 As shown, this embodiment also provides a non-destructive testing system for apple quality based on photoacoustic two-phase spectrum, including: a photoacoustic two-phase spectrum acquisition device and photo-vibration two-phase fruit and vegetable quality non-destructive testing software; The photoacoustic phase spectrum acquisition device includes a support unit for placing apple samples, a near-infrared spectrum acquisition unit for acquiring near-infrared absorption spectra, a vibration spectrum acquisition unit for acquiring time-domain vibration signals, and a control and communication unit for coordinated control and data synchronous transmission. Specifically, such as Figure 4 The diagram shows the spatial layout of a system device including components such as a laser Doppler vibration meter, a halogen tungsten lamp, a lead screw slide, a spectrometer, a fruit holder, and a vibrating horn.

[0040] The photoacoustic dual-phase spectrum acquisition device includes: Carrying unit: used to place the apple sample to be tested; Near-infrared spectral acquisition unit: includes a halogen tungsten lamp light source and a near-infrared spectrometer, used to acquire near-infrared diffuse reflectance spectral information of apple samples; Vibration spectrum acquisition unit: includes vibration excitation device (such as vibration horn, power amplifier, DAC decoder) and laser vibrometer, used to apply controllable vibration excitation to apple sample and acquire vibration response signal on its surface; Control and communication unit: connected to the near-infrared spectrum acquisition unit, vibration spectrum acquisition unit and data processing terminal, used to realize coordinated control and synchronous data transmission.

[0041] The optical-vibration two-phase fruit and vegetable quality non-destructive testing software includes a parameter setting module, a hardware driver module, a data processing and display module, and an equipment log module, and the optical-vibration two-phase fruit and vegetable quality non-destructive testing software is configured to execute the method described in Embodiment 1.

[0042] Specifically, such as Figure 5 The diagram shown is a schematic of the main interface, which includes a parameter setting area, a hardware driver button area, an image display area, and a device log area.

[0043] The optical two-phase fruit and vegetable quality non-destructive testing software (built into the data processing terminal) includes: Parameter setting module: used to set detection parameters such as spectral integration time, sweep frequency range, and excitation amplitude according to the size, weight, and surface characteristics of the fruits and vegetables to be tested; Hardware driver module: Used to drive the various hardware units of the photoacoustic binary spectrum acquisition device and perform operations such as device connection, calibration, and data acquisition; Data processing and display module: used to receive and process the acquired spectral and vibration signals, display the photoacoustic two-phase spectrum in real time, and output the quality prediction results; Equipment log module: Used to monitor the device's operating status and data calculation process, providing operation prompts and error records.

[0044] Specifically, the system in this embodiment includes a photoacoustic two-phase spectrum acquisition device placed in a dark box, and a computer equipped with a self-developed "photoacoustic two-phase fruit and vegetable quality non-destructive testing software".

[0045] The experimental steps include: 1. Sample preparation; The experimental apple variety was Aksu Red Fuji. Harvested in batches from an orchard in Aksu, Xinjiang Uygur Autonomous Region on November 25 (S1, unripe stage), December 1 (S2, ripe stage), and December 6 (S3, overripe stage), 2024, with 88 apples per batch, totaling 264 samples. All samples were selected from fruits 80-85mm in size, uniform in shape, and free from mechanical damage and pests / diseases. The fruit stems were retained during harvesting, and surface impurities were cleaned with a soft brush. After being covered with foam netting, the samples were transported back to the laboratory and placed in a (0±1)℃ cold storage for settling. Before the experiment, the samples were placed in a laboratory environment at 25±1℃ for 8 hours to eliminate moisture caused by low-temperature condensation and allow the internal physiological metabolic activities to return to normal room temperature levels before subsequent experiments were conducted.

[0046] 2. Equipment initialization and parameter setting; Connect the photoacoustic two-phase spectrum acquisition device to a power source, and connect the USB data transmission cable to a computer equipped with the photoacoustic two-phase fruit and vegetable quality non-destructive testing software. Double-click the software icon, click "Device Connection," and perform a device self-test to initialize the hardware. Based on the average size, weight, and surface smoothness of the test samples, set parameters such as spectral integration time, integration count, vibration reference amplitude, starting frequency, cutoff frequency, and sweep time in the software interface.

[0047] 3. Data collection; Place the apple sample parallel to the equatorial plane on the excitation surface of the support unit. Click the "Photoacoustic Diphasic Spectrum Detection" button, and the system will synchronously or time-divisionally acquire near-infrared spectra and vibration signals: Spectral acquisition: Using a near-infrared spectrometer, the spectra of four regions of interest evenly distributed along the equator of the apple were acquired and the average spectrum was calculated. The near-infrared absorption spectrum was obtained by absorbance conversion.

[0048] Acoustic signal acquisition: Frequency response curves under impact vibration were acquired using a laser Doppler vibration measurement platform. The baseline value for impact vibration was set to 0.02 (confirmed through preliminary experiments that it would not damage the apples), the frequency sweep range was 200-1200 Hz, the sweep time was 3 seconds, and the sampling frequency was set to 9765.6 Hz. After acquiring the time-domain vibration signal, a fast Fourier transform was performed to obtain the acoustic frequency domain spectrum.

[0049] 4. Data preprocessing; The raw near-infrared spectrum and acoustic spectrum were corrected using a normalized ratio algorithm, respectively. (8) Where Xi,NSR represents the spectrum of the i-th sample after correction by the normalized ratio algorithm; Xi is the original spectrum of the i-th sample; xh and xc represent data points that are only related to the addition coefficients and only related to the multiplication coefficients, respectively. Monte Carlo random sampling (MCS) is used for 10 iterations to select data points h and c based on the principle of minimum error.

[0050] 5. Feature signal screening and fusion; Using hardness values ​​and soluble solids content measured by standard laboratory methods as reference values, feature signals were screened using correlation analysis (CA), continuous projection algorithm (SPA), and competitive adaptive resampling (CARS). The screened optical and acoustic features were then fused using the following three methods: Data-level fusion: The original spectral data and vibration time-domain data are directly spliced ​​together and then input into the model.

[0051] Feature-level fusion: Feature vectors are extracted from the spectral and vibration signals respectively and then spliced ​​together.

[0052] Decision-level fusion: Establish spectral prediction model and vibration prediction model separately, and then perform a weighted average of the output results of the two models.

[0053] 6. Quality prediction model establishment and performance comparison; Three modeling methods—Partial Least Squares Regression (PLSR), Multilayer Perceptron (MLP), and Support Vector Machine Regression (SVR)—were used to compare the prediction performance of single-modal (spectral, vibrational) and multi-modal fusion (data-level, feature-level, decision-level) methods. The prediction results for hardness are shown in Table 1.

[0054] Table 1 Experimental results show that for hardness prediction, the data-level fusion effect under the PLSR model is the best, with a test R of 0.8406, a test RMSE of 0.5417, and a test RPD of 1.8419, which is significantly better than single spectrum and single vibration mode.

[0055] The prediction results for soluble solids are shown in Table 2.

[0056] Table 2 Experimental results show that for soluble solids prediction, the feature-level fusion model under the MLP model exhibits the best performance, with an R-value of 0.9125, an RMSE of 0.6592, and an RPD of 2.4242. The feature-level fusion model under the PLSR model also performs excellently, with an R-value of 0.9126, an RMSE of 0.6802, and an RPD of 2.3495. Both are superior to single spectral modes and single vibrational modes.

[0057] The experimental data above fully demonstrate that the non-destructive testing method for apple quality based on photoacoustic two-phase spectroscopy proposed in this invention can significantly improve the prediction accuracy of apple hardness and soluble solids by fusing near-infrared spectroscopy and vibrational spectrum. For texture quality indicators such as hardness, the vibrational single-mode already performs well (PLSR test R=0.767), but after fusion, it is further improved to 0.8406; for chemical composition indicators such as soluble solids, the spectral single-mode performs well (PLSR test R=0.8995, MLP test R=0.9077), while the feature-level fusion further improves it to over 0.9126, proving the effectiveness of multi-source information fusion.

[0058] Compared with the prior art, this embodiment has the following beneficial effects: 1. Significantly Improved Detection Accuracy: This invention achieves high-precision prediction of apple firmness and soluble solids content by fusing near-infrared spectroscopy and vibrational spectra. Experimental verification (see detailed implementation method) shows that using a feature-level fusion combined with the PLSR model, the correlation coefficient (R) for apple firmness on the test set reaches 0.8406, the root mean square error (RMSE) is 0.5417, and the relative prediction deviation (RPD) is 1.8419. All these indicators are superior to the single spectral model (R=0.4008, RMSE=0.9079, RPD=1.099) and the single vibrational model (R=0.767, RMSE=0.6496, RPD=1.5359). For soluble solids content, a characteristic-level fusion combined with MLP model was used, and the test set R reached 0.9125, RMSE was 0.6592, and RPD was 2.4242, which is also better than the single spectral model (R=0.9077, RMSE=0.7081, RPD=2.2568) and the single vibration model (R=0.4633, RMSE=1.4254, RPD=1.1212).

[0059] 2. Flexible and Effective Multimodal Fusion Strategy: This invention validates three strategies: data-level fusion, feature-level fusion, and decision-level fusion. Experimental results show that for hardness prediction, both data-level fusion (PLSR, test R=0.8406) and feature-level fusion (PLSR, test R=0.8201; MLP, test R=0.7118; SVR, test R=0.7983) achieve good results; for soluble solids prediction, feature-level fusion (MLP, test R=0.9125; PLSR, test R=0.9126) performs best. This invention allows for the selection of the optimal fusion strategy based on different quality indicators.

[0060] 3. Abundant testing information: It can simultaneously and non-destructively detect multiple indicators related to apple quality, such as firmness and soluble solids content, providing more comprehensive data support for the refined grading of apples.

[0061] 4. High level of intelligence: It combines feature screening and data fusion to achieve intelligent, rapid and non-destructive testing of apple quality, and is easy to integrate into automated grading production lines.

[0062] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A non-destructive testing method for apple quality based on photoacoustic dual-phase spectroscopy, characterized in that, Includes the following steps: Near-infrared absorption spectra and time-domain vibration signals during stimulated vibration of apple samples were collected, and the time-domain vibration signals were subjected to fast Fourier transform to obtain the vibration spectrum. The near-infrared absorption spectrum and the vibrational spectrum are preprocessed; Based on the standard values ​​of apple hardness and soluble solids content, feature signals related to hardness and soluble solids content in the near-infrared absorption spectrum and the vibration spectrum are screened. The selected optical feature signals and acoustic feature signals are fused using data-level fusion, feature-level fusion, or decision-level fusion methods to obtain photoacoustic two-phase fusion features; The photoacoustic two-phase fusion features are input into a pre-established quality prediction model, which outputs the predicted values ​​of apple hardness and soluble solids content.

2. The method according to claim 1, characterized in that, The preprocessing includes correcting the near-infrared absorption spectrum and the vibrational spectrum using one or more of the following: normalized ratio algorithm, multi-source scattering correction, normal variable transformation, first derivative, and second derivative.

3. The method according to claim 1, characterized in that, When screening feature signals, one or more of the following methods are used: correlation analysis, continuous projection algorithm, competitive adaptive resampling algorithm, iterative population analysis of information-preserving variables or variable combinations.

4. The method according to claim 1, characterized in that, Before fusing the selected optical and acoustic feature signals, a step is also included to remove the influence of the fruit peel: Near-infrared spectral and acoustic characteristics of whole apples and peeled apples were collected separately. The difference between the two was used to obtain a data matrix. Singular value decomposition was performed on the data matrix to extract the projection vectors in the peel, and a spatial projection matrix was constructed. The feature information of the whole apples was corrected using the spatial projection matrix to obtain the feature information after removing the influence of the peel.

5. The method according to claim 1, characterized in that, The data-level fusion involves directly concatenating the original near-infrared absorption spectrum data and the original time-domain vibration data; the feature-level fusion involves extracting feature vectors from the near-infrared absorption spectrum and the vibration spectrum respectively and then concatenating them; the decision-level fusion involves establishing a spectral prediction model and a vibration prediction model respectively and then weighting the output results of the two models.

6. The method according to claim 1, characterized in that, The quality prediction model is a principal component regression model, a multilayer perceptron model, or a support vector machine regression model.

7. The method according to claim 1, characterized in that, It also includes a maturity discrimination threshold verification step: determining the maturity classification threshold based on the maximum inter-class variance method or the maximum entropy method; The formula for the maximum inter-class variance is: ； in, q otsu,th When the threshold is th The inter-class variance of the two types of data; u It is the average of the sample predicted data, which can be derived from... Calculated; u 0 and u 1 represents the average value of the first and second category sample data; h i This is the predicted maturity value of the sample; n The total number of samples; l The number of samples in the first category; The formula for the maximum entropy method is: ； in, q entropy,th It is the sum of the entropy values ​​of the two types of data when the threshold is selected as th; p t The probability that the predicted maturity value of the apple sample belongs to t; P 1 represents the first type of probability, which can be derived from... The calculation yields the result; the symbol ln represents the logarithmic transformation; t represents the range of the predicted maturity values ​​for the apple samples, which is determined by the number of iterations and the iteration step size.

8. The method according to claim 1, characterized in that, The excited vibration is applied by a vibration excitation device, the frequency range of the sweep signal output by the vibration excitation device is 200Hz to 1200Hz, and the sweep time is 3 seconds.

9. The method according to claim 1, characterized in that, The near-infrared absorption spectrum is collected in a wavelength range that covers the characteristic absorption bands of apple soluble solids and pectin.

10. A non-destructive testing system for apple quality based on photoacoustic dual-phase spectroscopy, characterized in that, include: Photoacoustic two-phase spectrum acquisition device and optical-vibration two-phase non-destructive testing software for fruit and vegetable quality; The photoacoustic phase spectrum acquisition device includes a support unit for placing apple samples, a near-infrared spectrum acquisition unit for acquiring near-infrared absorption spectra, a vibration spectrum acquisition unit for acquiring time-domain vibration signals, and a control and communication unit for coordinated control and data synchronous transmission. The optical-vibration two-phase fruit and vegetable quality non-destructive testing software includes a parameter setting module, a hardware driver module, a data processing and display module, and an equipment log module, and the optical-vibration two-phase fruit and vegetable quality non-destructive testing software is configured to perform the method of any one of claims 1-9.

Images