A nondestructive testing method and equipment for black heart disease of pear

Abstract

The embodiment of the present application discloses a kind of pear black heart disease nondestructive testing method and equipment, the embodiment of the present application adopts the micro sampling cover of inner cavity volume not more than 30ml, in not more than 8 seconds sealing time, fast collection pear fruit head space gas, obtain the short time dynamic release characteristics of carbon dioxide and ethylene, and combine near infrared spectrum estimated multi-source information such as soluble solids, color and acoustic-mechanical texture, input fusion model is calculated to obtain a black heart early warning index, according to which pear fruit is divided into low, medium, high three risk levels, and linkage sorting mechanism is sorted.The matching equipment includes the micro sampling cover, multi-source sensing head, data fusion and sorting execution unit.The present application overcomes the defects that existing technology cannot obtain fruit physiological dynamic information under high-speed production rhythm, realizes the rapid, accurate, nondestructive early warning and sorting of pear black heart disease risk, effectively reduces storage loss.

Description

TECHNICAL FIELD

[0001] The embodiment of the present application relates to the field of non-destructive detection and grading of postharvest physiological diseases of fruits, in particular to a non-destructive detection method and equipment for black heart disease of pears. BACKGROUND

[0002] Pears are prone to internal tissue browning during long-term cold storage or controlled atmosphere storage, which manifests as darkening to black-brown in the core area (commonly known as "black heart disease"), which seriously affects the commercial value. Related studies have shown that preharvest internal quality (such as soluble solids SSC, titratable acid TA, ascorbic acid AA, etc.) is correlated with the occurrence of black heart disease during storage; in addition, the respiratory and ethylene metabolism characteristics of the fruit at the later stage of storage are also significantly related to the risk of black heart disease.

[0003] The existing technology mainly focuses on: ① non-destructive determination of SSC by near-infrared (NIR) for grading; ② reduction of browning by optimizing postharvest cooling rate and storage gas conditions. However, there is still a lack of a systematic non-destructive detection method and equipment that can quickly and quantitatively assess the risk of black heart before entering the warehouse and link the sorting, especially a lack of online equipment and decision algorithms that integrate SSC with color, acoustic / mechanical texture, short-term respiratory-ethylene dynamic response, and multi-spectral features for modeling. SUMMARY

[0004] Therefore, the embodiment of the present application provides a non-destructive detection method and equipment for black heart disease of pears to solve the blank in physiological dynamic detection and multi-modal fusion modeling of the existing technology, and to realize early risk warning and sorting of pears with multiple sources, dynamics, explainability, and linkability for black heart disease.

[0005] In order to achieve the above purpose, the embodiment of the present application provides the following technical solutions:

[0006] According to the first aspect of the embodiment of the present application, a non-destructive detection method for black heart disease of pears is provided, the method comprising:

[0007] Sampling and posture positioning step: placing the pear fruit in a detection station, positioning the posture and driving it to rotate to obtain multiple azimuth information near the equator;

[0008] Multi-source information acquisition step: synchronously or sequentially acquiring the following information of the pear fruit:

[0009] a. Optical information: obtaining spectral data by near-infrared or visible-near-infrared spectrometer, and calculating the estimated value of soluble solids based on this; and / or obtaining color feature values by image sensor, including at least one of lightness value, red-green chroma value, yellow-blue chroma value, hue angle, and saturation;

[0010] b. Acoustic-mechanical information: at least one of resonance frequency and damping ratio is extracted by exciting and collecting vibration signals;

[0011] c. Short-term gas dynamic information: using a micro-sampling hood to seal the pear, collecting the data of carbon dioxide concentration and ethylene concentration in headspace gas during the sealing period;

[0012] Feature extraction and calculation step: based on the short-term gas dynamic information, the instantaneous respiration rate and ethylene release rate are calculated;

[0013] Risk determination step: the soluble solid estimation value, the color feature value, the acoustic-mechanical parameter, the instantaneous respiration rate and the ethylene release rate are input into the trained fusion model to calculate a black heart warning index; according to the preset first threshold and second threshold, the pear is divided into low risk, medium risk or high risk level;

[0014] Sorting execution step: according to the risk level, the sorting execution mechanism is controlled to sort the pears into different paths.

[0015] Further, based on the short-term gas dynamic information, the instantaneous respiration rate and ethylene release rate are calculated, including:

[0016] Specifically, the sealing time is divided into a first stage of 0-3 seconds and a second stage of 3-8 seconds, the change slopes of carbon dioxide and ethylene are calculated respectively, and at least one of the respiration slope difference and the ethylene release slope difference between the two stages is calculated, and the slope difference is input into the fusion model as a feature.

[0017] Further, when collecting the short-term gas dynamic information, the detection limit of the ethylene sensor used is not higher than 20 ppb, the range of the carbon dioxide sensor is 0-5%, and the flow rate of the sampling gas path is 10-80 mL / min.

[0018] Further, the fusion model is a gradient boosting decision tree model; the method further includes outputting the contribution of each input feature to the black heart warning index.

[0019] Further, the first threshold and the second threshold are set by the following method: using the actual black heart disease occurrence of pears after eight months of cold storage and ten days of shelf life as labels for model training and calibration, so that the black heart disease occurrence rate of pears classified as low risk is not higher than 5%.

[0020] Further, when the origin or season of the detected pears changes, no more than one hundred samples with new labels are collected to fine-tune the last layer parameters of the fusion model, realizing online transfer learning and self-adaptation of the model.

[0021] Further, before the multi-source information collection step or at the beginning of each detection shift, a self-calibration step is performed, which includes optical calibration of the optical sensing unit using a spectral reference brick, and gas zero point and range calibration of the carbon dioxide sensor and ethylene sensor using zero point gas and span gas.

[0022] According to a second aspect of the embodiments of the present application, a nondestructive detection device for black heart of pears is provided, which comprises:

[0023] A feeding and posture setting module is configured to feed the pears and rotate them in a predetermined posture;

[0024] A multi-source sensing head is composed of a near-infrared spectral module and a color / multi-spectral image module, which is configured to acquire spectral data by a near-infrared or visible-near-infrared spectrometer and calculate soluble solid content estimates based on the spectral data; and / or acquire color feature values by an image sensor;

[0025] An acoustic-mechanical texture module is configured to acquire vibration signals by micro-amplitude excitation and extract at least one of resonance frequency and damping ratio;

[0026] A micro respiration-ethylene sampling cover has a cavity volume of not more than 30 ml and is provided with a lifting driving mechanism, an annular elastic sealing ring, a gas sampling pipeline, and a carbon dioxide sensor and an ethylene sensor, which are configured to seal the pears using the micro sampling cover and acquire data of changes in carbon dioxide concentration and ethylene concentration of headspace gas over time during the sealing;

[0027] A data fusion and operation unit is in communication connection with the multi-source sensing head, which is configured to receive multi-source data, execute a fusion model algorithm, calculate a black heart early warning index, and output a risk level and a sorting instruction;

[0028] A sorting execution mechanism is connected with the data fusion and operation unit and is configured to sort the pears into different channels according to the sorting instruction.

[0029] Further, an anti-condensation structure is arranged in the gas pipeline system of the micro respiration-ethylene sampling cover, and the device further comprises a self-calibration unit, which includes a reference brick for optical calibration and a zero point gas and span gas source for calibration of the gas sensor, and can execute an automatic calibration program at a set shift or period.

[0030] Further, when the device operates in an online mode, the detection and sorting pace of single fruits is not higher than 2 seconds; and when the device operates in a sampling inspection mode, the detection pace of single fruits is not higher than 30 seconds.

[0031] The embodiments of the present application have the following advantages:

[0032] The embodiment of the present application adopts a micro sampling cover with a cavity volume of no more than 30 ml, and in a sealing time of no more than 8 seconds, short-time dynamic release characteristics of carbon dioxide and ethylene are obtained by rapidly collecting the headspace gas of a pear fruit, and multi-source information such as soluble solids, color and acoustic-mechanical texture estimated by near-infrared spectroscopy is input into a fusion model to calculate a black heart warning index, according to which the pear fruit is divided into three risk levels of low, medium and high, and a sorting mechanism is linked to perform sorting. The matching equipment includes the micro sampling cover, the multi-source sensing head, the data fusion and sorting execution unit. The present application overcomes the defect that the prior art cannot obtain physiological dynamic information of the fruit at a high-speed production rhythm, and realizes rapid, accurate and non-destructive warning and sorting of the risk of pear black heart disease, effectively reducing storage loss. BRIEF DESCRIPTION OF DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the following embodiment or prior art description will be briefly introduced. Obviously, the drawings in the following description are only exemplary, and for those skilled in the art, other embodiments can be obtained from the provided drawings without creative labor.

[0034] The structures, proportions, sizes, etc. shown in the specification are only used to cooperate with the content disclosed in the specification, to be understood and read by those skilled in the art, and do not define the limiting conditions for the implementation of the present application, so they do not have technical significance. Any modification of structure, change of proportion relationship or adjustment of size, without affecting the effect and purpose that the present application can produce, should still fall within the scope of the technical content disclosed by the present application.

[0035] Figure 1 A flowchart of a non-destructive detection method for pear black heart disease provided by the embodiment of the present application;

[0036] Figure 2 A structural schematic diagram of a non-destructive detection equipment for pear black heart disease provided by the embodiment of the present application;

[0037] Figure 3 A multi-source sensor flowchart of a non-destructive detection method for pear black heart disease provided by the embodiment of the present application;

[0038] Figure 4 A data flowchart in the detection process of a non-destructive detection method for pear black heart disease provided by the embodiment of the present application;

[0039] Figure 5 A PBRI calculation flowchart in a non-destructive detection method for pear black heart disease provided by the embodiment of the present application;

[0040] Figure 6 A parameter criticality test schematic diagram in a pear black heart disease nondestructive detection method provided by the embodiment of the present application is shown in the figure;

[0041] Figure 7 An online sorting linkage schematic diagram in a pear black heart disease nondestructive detection method provided by the embodiment of the present application is shown in the figure;

[0042] Figure 8 A domain migration-self-calibration flow schematic diagram in a pear black heart disease nondestructive detection method provided by the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0043] The embodiments of the present application are described below by specific embodiments, and those skilled in the art can easily understand other advantages and effects of the present application from the disclosed content of the specification. Obviously, the described embodiments are part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0044] In order to solve the blank of physiological dynamic detection and multi-modal fusion modeling in the prior art, realize the early risk warning and sorting of pear black heart disease in multi-source, dynamic, interpretable and linkable way.

[0045] REFERENCE Figure 1 and Figure 2 The embodiment of the present application discloses a pear black heart disease nondestructive detection equipment, specifically comprising: a feeding and posture setting module, a multi-source sensing head, a data fusion and operation unit, a sorting execution mechanism and a man-machine interface, a self-calibration and safety monitoring unit. Its structure and function are as follows.

[0046] 1. Feeding and posture setting module

[0047] Feeding and spacing: roller or belt + star wheel spacing, mechanical pitch 80-120mm, up to ≥30000 fruits / h (single channel).

[0048] Posture setting: upper / side view camera identifies pedicel-calyx end, drives the gripper to make the equatorial zone enter the measurement position; rotation angle ≥180°, positioning error ≤±5°.

[0049] Clamping and buffering: flexible cup holder / gel wheel limiting, shock absorption design to suppress knock echo crosstalk.

[0050] 2. Multi-source sensing head (sensing and sampling assembly)

[0051] 2.1 Near-infrared (NIR) spectral module

[0052] Wavelength range: 900-1700 nm (optional 750-2000 nm), diffuse / annular diffuse light path; integration time < 20 ms / azimuth.

[0053] Calibration: built-in reference tile (white plate / diffuse reflector) with dark current timing calibration; drift compensation < 1% per shift.

[0054] 2.2 Color / multispectral image module

[0055] Imaging: RGB or 4-6 channel multispectral line / surface array camera; annular constant current light source; computation etc.

[0056] where, L is luminance, a is red-green chrominance, b is yellow-blue chrominance, h is hue angle, s is saturation.

[0057] Synchronization: with star wheel angle encoder, de-blurring exposure (< 1 ms).

[0058] 2.3 Acoustic-mechanical texture module

[0059] Excitation: micro-actuator single pulse (about 2 ms, displacement < 0.3 mm, force < 3 N);

[0060] Acquisition: MEMS accelerometer / force sensor, bandwidth 0-5 kHz; resonance frequency , damping ratio ζ, etc.

[0061] 2.4 Micro-respiration-ethylene sampling hood

[0062] Cavity: effective volume < 30 mL (preferably 15-30 mL), PC / PEEK anti-cracking transparent cover; bottom with annular elastic sealing ring (food-grade silicone).

[0063] Sealing: electric lifting / micro-cylinder driven, sealing-sampling-lifting < 1 s; sealing sampling time < 8 s (friendly to work station beat).

[0064] Gas path: 1 / 16" PTFE pipeline; micro diaphragm pump 10-80 mL / min; sampling port and exhaust port separated, with one-way valve and anti-back suction structure; with condensation / fog suppression (hydrophobic membrane + micro heating).

[0065] Sensor compartment:

[0066] CO2: NDIR, range 0-5% (v / v), response t90 < 2 s;

[0067] Ethylene: photoacoustic or electrochemical detection limit ≤20 ppb, linearity better than ±2%FS;

[0068] O2: Electrochemical or optical dissolved oxygen type, for monitoring diffusion / leakage.

[0069] Leak detection and self-test: Head space pressure / flow differential monitoring; zero-air self-test for leak rate (threshold alarm) every shift, automatic flushing for 2-5 seconds to reduce residue.

[0070] Materials and Hygiene: Parts in contact with gas are made of PTFE / PEEK / 316L; O-rings are quick-release and replaceable, and the cover can be wiped clean.

[0071] Note: The sampling hood can output short-term slopes Rs, Es and the slope difference ΔR, ΔE under the conditions of ≤30mL×≤8s, which significantly enhances the amount of discrimination information without sacrificing the production line cycle time.

[0072] 3. Data fusion and processing unit (edge ​​computing)

[0073] Hardware: Industrial computer (x86 / ARM, 4–8 cores, 8–16GB RAM), isolated digital / analog I / O, RS-485 / RS-232, Gigabit Ethernet.

[0074] Timing synchronization: hardware trigger + timestamp bus (spectral / image / acoustic / gas data alignment error ≤ 5ms).

[0075] Algorithm: Built-in feature normalization and fusion model (LogReg / GBDT / LightGBM) to calculate PBRI and risk level (A / B / C) in real time; supports online inference ≤2s / result.

[0076] Data management: unique ID for each fruit (time / channel / batch); local cache ≥30 days, MQTT / OPCUA uplink to MES / cloud.

[0077] 4. Sorting execution agency

[0078] Actuator: Push rod.

[0079] Key points summary

[0080] A) Miniature sampling hood: inner cavity ≤30mL, sealing ≤8s, pump 10–80mL / min, ethylene detection limit ≤20ppb, CO2 range 0–5%;

[0081] B) Synchronous acquisition: NIR / color / acoustic and short-time gas are triggered and timestamped simultaneously;

[0082] C) Self-calibration + leak detection: Zero / span gas and optical reference brick, voltage / current stabilization leak detection interlock;

[0083] D) Linked sorting: A / B / C three channels, cycle time ≤ 2s / fruit, data traceability.

[0084] Corresponding to the non-destructive testing equipment for pear black heart disease disclosed above, this invention also discloses a non-destructive testing method for pear black heart disease. The following details a non-destructive testing method for pear black heart disease disclosed in this invention, in conjunction with the aforementioned non-destructive testing equipment for pear black heart disease.

[0085] This invention aims to provide a non-destructive detection method and equipment for pear black heart disease, which can be used for rapid screening, early warning, and automatic sorting before post-harvest storage or during cold storage, achieving:

[0086] 1. Without damaging the fruit, provide a risk level for tainted fruit within ≤2 seconds / fruit (online) or ≤30 seconds / fruit (sampling / batch inspection);

[0087] 2. Improve prediction accuracy through multi-source sensing and robust modeling with small samples, and provide an interpretable “Black Heart Warning Index (PBRI)”;

[0088] 3. Linked with the warehousing / sorting line to achieve storage according to risk path (low-risk cold storage, higher-risk near-sales or process).

[0089] refer to Figures 3 to 8 This invention discloses a non-destructive detection method for pear black heart disease, comprising the following steps:

[0090] Sampling and orientation positioning: Place the pear on a conveyor belt or sampling tray, and establish an equatorial reference by visually positioning the fruit stem to the calyx end. Drive the rotation to scan the fruit surface in all directions (rotation angle ≥180°).

[0091] Multi-source non-destructive acquisition: Simultaneous acquisition of each fruit:

[0092] 2.1 Near-infrared / visible-near-infrared spectrum: 900–1700 nm (expandable to 750–2000 nm), diffuse reflectance mode, output SSC prediction values ​​and key spectral features;

[0093] 2.2 Color Image: Acquisition , Tone angle With saturation Equal characterization;

[0094] in, For brightness, For red-green color, For yellow-blue hue, For hue angle, Saturation;

[0095] 2.3 Acoustic-mechanical structure: Micro-amplitude non-destructive tapping or micro-compression (displacement ≤ 0.3 mm, force ≤ 3 N) was used to extract the resonant frequency f0, damping ratio ζ, etc.

[0096] 2.4 Short-term gas dynamics: A miniature breathing-ethylene sampling hood (internal cavity ≤30 mL, closure time ≤8 s) was placed outside the fruit surface, and the sampling head space was collected by a micro-flow pump; ΔCO2 / Δt and ΔC2H4 / Δt were recorded synchronously by infrared CO2 and photoacoustic / electrochemical ethylene sensors;

[0097] 2.5 Microenvironment detection: The miniature O2 probe records the rate of O2 change inside the enclosure and estimates gas diffusion-related parameters.

[0098] Feature engineering and computational metrics:

[0099] 3.1 Obtaining SSC estimates from spectral regression (PLS / Ridge / 1D-CNN) ;

[0100] 3.2 Calculated from the image ,in, For brightness, For red-green color, For yellow-blue hue, For hue angle, Saturation;

[0101] 3.3 Obtained from acoustic / mechanical testing , The resonant frequency, The damping ratio;

[0102] 3.4 Obtaining instantaneous respiratory rate from short-term dynamics With ethylene release rate ;

[0103] 3.5 Constructing a Black Heart Warning Index (PBRI):

[0104] in For normalization transformation, weights With bias The data is determined from labeled historical data using logistic regression / gradient boosting tree / LightGBM.

[0105] Risk assessment: Based on the PBRI, three levels of risk are output:

[0106] Low risk (A): PBRI ≥ T1;

[0107] Medium risk (B): T2 ≤ PBRI < T1;

[0108] High risk (C): PBRI < T2. Thresholds (T1, T2) are set by the black-hearted occurrence label of the training set "8 months of refrigeration + 10 days of shelf life", and the optimization is to make the black-hearted occurrence rate of Class A samples ≤5% in 8 months + 10 days.

[0109] Linked sorting and traceability: The control push rod / air separator / flip plate imports A / B / C into different turnover boxes respectively; each fruit generates a unique ID and is archived in the cloud, which facilitates batch tracking and re-judgment in the warehouse.

[0110] Modeling and calibration methods:

[0111] 1. Dual-domain self-calibration: Introducing "spectral reference brick + gas zero point / span gas" for automatic calibration every shift;

[0112] 2. Limited labeled transfer learning: When the place of origin / season changes, the parameters of the last layer of the model are updated with n≤100 re-judgment samples to achieve cross-place of origin generalization;

[0113] 3. Short-term dynamic enhancement: Compare the slope difference of Rs and Es in the two stages of 0–3 s and 3–8 s after the shield is opened, and construct ΔR and ΔE as energy metabolism robustness characteristics;

[0114] 4. Interpretable output: Displays contribution bar charts (contributions of SSC, color, and dynamic gases to PBRI), facilitating adjustments to sorting strategies.

[0115] This embodiment is applicable to online high-speed detection and sorting of pears in post-harvest processing centers. The pears are conveyed via a roller conveyor line, and the fruit spacing is adjusted to approximately 100 millimeters by a star wheel spacing mechanism. A vision system consisting of upward and side-view cameras identifies the fruit stalk and calyx end to determine the equatorial plane of the fruit. Flexible clamping wheels, driven by a servo motor, rotate the pears at least 180 degrees to ensure that the equatorial region is fully scanned.

[0116] In the multi-source information acquisition stage, the near-infrared spectrometer collects diffuse reflectance spectra at three equally divided positions during the fruit's rotation, with an integration time of no more than 20 milliseconds at each position. Simultaneously, a linear RGB camera acquires images under a ring LED light source and calculates the average brightness value, red-green value, yellow-blue value, hue angle, and saturation of the fruit surface. A micro-force electromagnetic actuator applies a micro-pulse with a duration of about two milliseconds and a displacement of no more than 0.3 millimeters to the equatorial part of the fruit, and a micro accelerometer collects the vibration response signal to extract its principal resonant frequency and damping ratio.

[0117] The core short-term gas dynamic acquisition is performed by a miniature sampling hood with a cavity volume of 25 ml. After receiving the fruit positioning signal, it is driven to descend by an electric push rod. The bottom annular food-grade silicone sealing ring forms a sealed cavity with the fruit surface. The sealing duration is set to eight seconds. During this period, a miniature diaphragm pump pumps the gas in the head space into the parallel gas sensor chamber at a flow rate of 50 ml per minute. The non-dispersive infrared carbon dioxide sensor and the photoacoustic ethylene sensor record the gas concentration change curves over time, respectively.

[0118] In the data processing and risk assessment stage, the soluble solids value of spectral data is estimated in real time using a partial least squares regression model. The gas dynamic data is divided into two time periods, 0 to 3 seconds and 3 to 8 seconds, for linear fitting to obtain the initial instantaneous respiration rate, ethylene release rate, and the slope difference between the two periods. Finally, eight features, including the estimated soluble solids value, hue angle, saturation, resonance frequency, damping ratio, instantaneous respiration rate, ethylene release rate, and respiration slope difference, are input into a pre-trained LightGBM fusion model to output a black-heart warning index between zero and one. Based on the preset first threshold of 0.62 and the second threshold of 0.38, the fruit is divided into three risk levels: low, medium, and high.

[0119] The sorting and traceability are performed by a pneumatic push rod controlled by a data fusion and computing unit, which guides fruits of different risk levels into their corresponding channels. At the same time, the unique identification code of each fruit and all grading data are uploaded to the cloud database via the MQTT protocol to achieve full-chain traceability.

[0120] This embodiment is suitable for precise sampling and testing of specific batches at the entrance of laboratories or cold storage facilities. It uses a 16-position parallel testing tray instead of an online conveyor line, with operators manually placing pears into flexible cup holders on the trays. A rotating mechanism at each station rotates the pears 360 degrees, allowing the near-infrared spectrometer to collect spectral data from more directions, improving the robustness of soluble solids estimation. The sealing time of the miniature sampling hood can be extended to 15-30 seconds to further reduce the uncertainty in low-concentration ethylene measurements. Its data processing and grading procedures are consistent with the online mode, but the cycle time is relaxed to no more than 30 seconds per fruit, emphasizing data accuracy and reliability.

[0121] This embodiment illustrates the construction and maintenance method of the core model. Initial model training requires collecting over 1,000 pear fruit samples covering at least two major production areas, three main varieties, and three harvest periods (early, mid, and late). After collecting multi-source information using the equipment of this invention, eight months of standard cold storage simulation and ten days of shelf-life simulation are conducted. Finally, the actual black heart disease severity index of each fruit is obtained through fruit dissection and detection as the gold standard label. During the model construction phase, the collected multi-source features are aligned with the black heart disease label. A five-fold cross-validation method is used to select the best-performing algorithm from multiple algorithms as the fusion model, and its hyperparameters are determined. Online transfer learning is initiated when the equipment is deployed to a new production area. Approximately sixty samples are randomly selected from the first batch of pears stored for testing and subsequent storage experiments. The acquired new label data is used to fine-tune the last layer or some parameters of the existing model, thereby enabling the model to quickly adapt to the characteristics of fruits from the new production area.

[0122] This embodiment illustrates the key maintenance process for ensuring the long-term stable operation of equipment. Daily shift calibration includes two parts: optical calibration and gas calibration. Optical calibration is performed by automatically aligning the near-infrared spectroscopy module with the built-in ceramic reference white plate to calibrate diffuse reflectance and collect dark current signals. Gas calibration is performed by automatically introducing zero-point gas and span gas into the sampling hood and gas path to calibrate the zero point and range of carbon dioxide and ethylene sensors. The system records the calibration curves and performs real-time drift compensation in subsequent measurements. In the leak detection and cleaning process, the airtightness of the sampling hood's sealed gas path is checked every shift using a micro-differential pressure sensor, and an alarm is issued when the leakage rate exceeds the limit. The surface of the sampling hood and gas path can be quickly wiped with sterile wet wipes automatically or manually to prevent cross-contamination.

[0123] Example 1 (Online mode, 30,000 fruits / hour)

[0124] Conveying speed: ≥1.2 m / s, star wheel 12-position spacing;

[0125] Spectral acquisition: 3 orientations per fruit, integration time ≤20 ms per orientation, and the final SSC is output as the weighted median;

[0126] Color acquisition: Linear scan camera + ring light, calculation ,in, For brightness, For red-green color, For yellow-blue hue, For hue angle, Saturation;

[0127] Acoustic percussion: Microactuator pulse 2 ms, sampling bandwidth 0–5 kHz, extracting f0, ζ. The resonant frequency, The damping ratio;

[0128] Sampling hood: Immediately after lowering and sealing, data collection begins for 8 seconds. ΔCO2 and ΔC2H4 are obtained through linear fitting. ;

[0129] Model: LightGBM, with 36 features; thresholds set to T1=0.62 and T2=0.38 (example values).

[0130] Example 2 (Sampling / Batch Inspection Mode)

[0131] 16-bit parallel measurement on the tray; each fruit rotates 360° to acquire more azimuth spectrum for improved robustness;

[0132] Allow the sampling hood to be closed for 15–30 s to improve the signal-to-noise ratio of low-concentration ethylene (≤20 ppb detection limit).

[0133] Example 3 (Model Training and Transfer)

[0134] Collect 1,000 tagged samples (including different origins, tree ages, and harvest periods), and use the black heart index of the cut fruit after 8 months of cold storage + 10 days of shelf life as the gold standard;

[0135] Model and hyperparameter selection were performed using 5-fold cross-validation;

[0136] When a new production area is established, approximately 60 re-judgment samples are extracted for final parameter fine-tuning to achieve domain migration.

[0137] The system monitors the actual rate of tainted products in Category A batches online, and automatically triggers an adaptive threshold if the rate exceeds 5%.

[0138] Example 4 (Interlocking Control)

[0139] Category A (low risk) items are stored in deep cold storage; Category B (medium risk) items are stored in conventional cold storage; Category C (high risk) items are transferred to near-market sales or further processing.

[0140] The system binds the grading results to the pallet / box code, providing traceability and statistical reports.

[0141] Comparative experiment and effect data:

[0142] To support the unexpected technical effects of "short-term breathing - ethylene dynamics + multimodal fusion", the following comparative and parameter criticality tests were set up; the test data can be used as appendices / figures in the instruction manual:

[0143] A. Baseline Comparison

[0144] Sample size and stratification: N≥1,000, covering at least 2 production areas × 2 varieties × 2 harvest periods; the gold standard is “8 months of cold storage + 10 days of shelf life followed by cutting open the fruit and finding black-hearted labels”.

[0145] Model group:

[0146] G1: NIR only (SSC regression + discriminant);

[0147] G2: NIR+ Color / Acoustics;

[0148] G3: NIR + Color / Acoustics + Short-Time Gas Dynamics (Invention).

[0149] Evaluation metrics: AUC, sensitivity at 95% specificity, F1, batch-to-batch drift stability (across production areas / seasons); statistical analysis was performed using the DeLong test (AUC) and McNemar test (hit / miss), with 95% CI obtained by bootstrap method (1,000 trials).

[0150] Preset compliance criteria: Under the same false alarm constraint, the AUC of G3 is ≥0.04 higher than that of G2 and ≥0.08 higher than that of G1; when migrating across production sites, the sensitivity of G3 decreases by ≤10%.

[0151] B. The necessity of second-level dynamics

[0152] Compare the two sealing sampling times: 8 s and 60 s. Record ΔCO2 / Δt, ΔC2H4 / Δt, and the slope difference between the two stages (0–3 s vs 3–8 s). This demonstrates that a stable and distinguishable signal can be formed within 8 s, meeting the requirements for online cycle time; while 60 s slightly improves the SNR, it is unusable on the production line.

[0153] C. Parameter Criticality

[0154] Full factorial tests were conducted with cavity volumes of 15 / 30 / 60 mL and sealing times of 4 / 8 / 20 s; the slope SNR, leakage rate, cycle time loss, and discrimination AUC were output. The preferred range is: volume 15–30 mL, sealing time 6–10 s; where cavity volume ≤30 mL and sealing time ≤8 s balances SNR and cycle time, supporting the parameter window of the claims.

[0155] D. Online beat and stability

[0156] A 48-hour continuous operation test was conducted on a 30,000 fruits / hour line: the production line throughput, equipment self-calibration drift (spectral brick and zero / span gas), and zero drift and compensation effect of the ethylene / CO2 sensor were recorded.

[0157] E. Threshold Calibration and Traceability

[0158] The actual incidence of black-hearted samples in Category A is used as an online quality control indicator on a batch basis; if the incidence is >5% for two consecutive batches, the threshold is automatically adjusted and the difference before and after adjustment is recorded.

[0159] The embodiments of the present invention have the following advantages:

[0160] 1. By combining multi-source fusion with short-term gas dynamics, the accuracy and stability of black market risk prediction are significantly improved;

[0161] 2. Create a miniature breathing-ethylene sampling mask to introduce dynamic physiological information without significantly reducing the beat rate;

[0162] 3. Provide interpretable and interconnected PBRI indicators to achieve "storage according to risk path" and reduce losses during deep storage period;

[0163] 4. Supports both random sampling and online modes, suitable for origin sorting and pre-warehouse acceptance.

[0164] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A non-destructive detection method for pear black heart disease, characterized in that, The method includes: Sampling and attitude positioning steps: Place the pear at the detection station, perform attitude positioning and drive its rotation to obtain multiple orientation information at least near the equator; Multi-source information acquisition steps: Collect the following information about the pear fruit synchronously or sequentially: a. Optical information: acquiring spectral data via a near-infrared or visible-near-infrared spectrometer and calculating an estimate of soluble solids based thereon; and / or acquiring color feature values ​​via an image sensor, including at least one of lightness, red-green hue, yellow-blue hue, hue angle, and saturation; b. Acoustic-mechanical information: By micro-excitation and acquisition of vibration signals, at least one of the resonant frequency and damping ratio is extracted; c. Short-term gas dynamics information: A miniature sampling hood was used to seal the pear fruit, and data on the changes in carbon dioxide and ethylene concentrations in the head space over time were collected during the sealing period; Feature extraction and calculation steps: Based on the short-term gas dynamic information, the instantaneous respiration rate and ethylene release rate are calculated; Based on the aforementioned short-term gas dynamics information, the instantaneous respiration rate and ethylene release rate are calculated, including: Specifically, the sealing time is divided into a first stage of 0 to 3 seconds and a second stage of 3 to 8 seconds. The slopes of carbon dioxide and ethylene changes are calculated respectively, and at least one of the breathing slope difference and ethylene release slope difference between the two stages is calculated. The slope difference is used as a feature input to the fusion model. When collecting the short-term gas dynamic information, the detection limit of the ethylene sensor used is no higher than 20 ppb, the range of the carbon dioxide sensor is 0-5%, and the flow rate of the sampling gas path is 10-80 mL / min. Risk assessment steps: The estimated value of soluble solids, the color feature value, the acoustic-mechanical parameters, the instantaneous respiration rate, and the ethylene release rate are input into a trained fusion model to calculate a black-heart warning index; based on a preset first threshold and second threshold, the pears are classified into low-risk, medium-risk, or high-risk levels. Sorting execution steps: Based on the risk level, control the sorting execution agency to send the pears to different paths.

2. The non-destructive detection method for pear black heart disease as described in claim 1, characterized in that, The fusion model is a gradient boosting decision tree model; the method also includes outputting the contribution of each input feature to the black-hearted warning index.

3. The non-destructive detection method for pear black heart disease as described in claim 1, characterized in that, The first and second thresholds are set in the following way: the actual occurrence of black heart disease in pears after eight months of cold storage and ten days of shelf life is used as a label for model training and calibration, so that the actual incidence of black heart disease in pears classified as low risk is no higher than 5%.

4. The non-destructive detection method for pear black heart disease as described in claim 1, characterized in that, When the origin or season of the pears being tested changes, no more than one hundred samples with new labels are collected to fine-tune the parameters of the last layer of the fusion model, thereby enabling online transfer learning and adaptation of the model.

5. The non-destructive detection method for pear black heart disease as described in claim 1, characterized in that, Before the multi-source information acquisition step or at the beginning of each testing shift, a self-calibration step is performed, which includes: optical calibration of the optical sensing unit using a spectral reference brick, and gas zero-point and range calibration of the carbon dioxide sensor and ethylene sensor using zero-point gas and span gas.

6. A non-destructive testing device for pear black heart lesion, used for performing the non-destructive testing method for pear black heart lesion as described in any one of claims 1 to 5, characterized in that, The equipment includes: The feeding and orientation module is used to transport pears and rotate them in a predetermined orientation. A multi-source sensor head, comprising a near-infrared spectroscopy module and a color / multispectral image module, is used to acquire spectral data through a near-infrared or visible-near-infrared spectrometer and calculate an estimate of soluble solids based thereon; and / or acquire color feature values ​​through an image sensor. An acoustic-mechanical structure module is used to extract at least one of the resonant frequency and damping ratio by micro-excitation and acquisition of vibration signals. The miniature breathing-ethylene sampling hood has a cavity volume of no more than 30 ml and is equipped with a lifting drive mechanism, an annular elastic sealing ring, a gas sampling pipeline, and carbon dioxide and ethylene sensors. It is used to seal pears with the miniature sampling hood and collect data on the changes in carbon dioxide and ethylene concentrations in the head space over time during the sealing period. Based on the aforementioned short-term gas dynamics information, the instantaneous respiration rate and ethylene release rate are calculated, including: Specifically, the sealing time is divided into a first stage of 0 to 3 seconds and a second stage of 3 to 8 seconds. The slopes of carbon dioxide and ethylene changes are calculated respectively, and at least one of the breathing slope difference and ethylene release slope difference between the two stages is calculated. The slope difference is then used as a feature input into the fusion model. When collecting the short-term gas dynamic information, the detection limit of the ethylene sensor used is no higher than 20 ppb, the range of the carbon dioxide sensor is 0-5%, and the flow rate of the sampling gas path is 10-80 mL / min. The data fusion and processing unit is communicatively connected to the multi-source sensor head and is used to receive multi-source data, execute fusion model algorithms, calculate the black-hearted warning index, and output risk level and sorting instructions. The sorting execution mechanism is connected to the data fusion and computing unit and is used to sort pears into different channels according to the sorting instructions.

7. The non-destructive testing equipment for pear black heart disease as described in claim 6, characterized in that, The gas path system of the miniature breathing-ethylene sampling hood is equipped with an anti-condensation structure. The equipment also includes a self-calibration unit, which includes a reference brick for optical calibration and a zero-point gas and span gas source for gas sensor calibration. It can perform automatic calibration procedures according to a set shift or cycle.

8. The non-destructive testing equipment for pear black heart disease as described in claim 6, characterized in that, When the equipment is operating in online mode, the detection and sorting cycle for a single fruit is no more than two seconds; when operating in sampling inspection mode, the detection cycle for a single fruit is no more than thirty seconds.

Images