A method for testing physical parameters to detect quality changes in cucurbit vegetables during storage

Abstract

The application discloses a physical parameter testing method for detecting quality change of melon vegetables during storage, relates to the technical field of optical detection, and is used for solving the problem of poor melon storage quality monitoring; the application realizes sensing and prediction of melon vegetable storage quality by constructing a two-layer optical model and a self-adaptive correction detection system, completes condition standardization through stable multi-band light sources, co-view field acquisition and environment benchmark data, generates a surface water film index through polarization contrast and diffuse reflectance ratio, forms two-state difference by cooperating with a micro-heat pulse, deducts surface interference, inverses surface film thickness, tissue absorption parameters and tissue scattering parameters, extracts internal moisture indicators and cell integrity indicators, and carries out joint correction with temperature, humidity and dew point, thereby constructing a quality mapping model which integrates internal moisture, tissue scattering characteristics and chroma parameters, outputting quality grades and change trends and triggering early warning and regulation suggestions, so as to improve detection accuracy and on-site availability.

Description

Technical Field

[0001] This invention relates to the field of optical detection technology, and more specifically, to a method for testing physical parameters to detect quality changes in cucurbit vegetables during storage. Background Technology

[0002] Cucurbit vegetables are an important part of the daily food consumption of Chinese residents. Among them, bitter melon and winter melon are widely cultivated and consumed due to their high nutritional value, cooling and heat-relieving properties, and medicinal and edible characteristics. Bitter melon is rich in various functional components, such as momordicin, polypeptides, and vitamin C, and is believed to have effects such as lowering blood sugar, anti-oxidation, and enhancing immunity. Therefore, the market demand is stable. However, cucurbit vegetables are characterized by high water content, strong respiratory metabolism, and crisp and tender tissues. They are extremely sensitive to external temperature, humidity, and gaseous environments. After harvesting, they are prone to wilting, browning, texture deterioration, and nutrient loss during storage and distribution, which seriously affects their commercial value and food safety.

[0003] The shortcomings of existing technologies are as follows: In the sample testing stage, stable positioning and optical benchmark calibration are not performed for the complex surface structure of melons such as bitter melon and winter melon, resulting in poor repeatability of collected data. Especially in the environmental interference stage, traditional methods do not identify and quantify the condensation film commonly found under storage conditions, and the distortion effect of surface liquid water on the spectrum is not recognized, making it easy to mistake the surface water film for internal quality changes. At the same time, in the data acquisition stage, existing methods usually only perform single-state measurements and cannot separate the film-containing and film-free states through differential methods, making it difficult to distinguish between internal moisture and external interference. They ignore the two-layer optical structure of "surface film and internal tissue", and the inverted parameters do not match the true quality, resulting in the lack of an early warning mechanism and the inability to achieve real-time monitoring and dynamic control in the storage chain. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the existing technology, the following solution is proposed to solve the problem of poor monitoring of melon storage quality in the above-mentioned background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for testing physical parameters to detect quality changes in cucurbit vegetables during storage, comprising the following steps:

[0007] Representative cucurbit vegetable samples were selected and the detection points were fixed. A stable multi-band light source, spectrometer and imaging unit were configured in the transparent measurement chamber to complete the dark field and white board calibration and lock the incident and acquisition geometric parameters.

[0008] The temperature, humidity and dew point information of the storage environment are recorded simultaneously. The presence of a water film on the sample surface is determined based on polarization contrast and diffuse reflectance ratio, and a surface water film index is generated.

[0009] Two-state rapid acquisition is performed on the same detection point. State 1 is the original state, and state 2 is the weakened surface water film state under the action of micro-thermal pulse. Multi-band reflectance spectrum and color image are acquired in the two states respectively, and differential spectrum and differential image are formed.

[0010] Based on a two-layer optical model of surface film and internal tissue, differential data is inverted to obtain surface film thickness, tissue absorption parameters and tissue scattering parameters. Internal water content and cell integrity indicators are extracted and adaptively corrected by combining surface water film index and environmental parameters.

[0011] Establish a quality mapping model based on internal moisture indicators, tissue scattering parameters, and color parameters. Output the quality grade and its changing trend during storage. When the quality indicators exceed the preset quality threshold, trigger an early warning and provide storage control suggestions.

[0012] Furthermore, the cucurbit vegetable samples mainly consist of bitter melon, but are also applicable to similar crops such as winter melon and loofah. The detection points are fixed on representative areas of the sample epidermis by marking or coordinate indexing to maintain consistency and traceability of repeated measurements.

[0013] Furthermore, the stable multi-band light source inside the transparent measurement chamber is composed of a combination of visible light and near-infrared light. The spectrometer and imaging unit are fixedly arranged with a common field of view and form a locked incident and acquisition geometry with the sample surface. The measurement chamber is equipped with polarization components, a white board and a dark field calibration board, as well as mechanical positioning fixtures, to complete the dark field and white board calibration and ensure the repeatability of the detection points.

[0014] Furthermore, the environmental parameter acquisition method involves setting up temperature sensors, humidity sensors, and dew point sensors inside the transparent measurement chamber to collect temperature, humidity, and dew point information in a coordinated manner.

[0015] The collected temperature, humidity, and dew point information are associated with timestamps and detection point numbers and stored to generate environmental baseline data. This environmental baseline data is used for surface water film index determination and subsequent adaptive correction.

[0016] Furthermore, based on polarization contrast and diffuse reflectance ratio, the presence of a water film on the sample surface is determined, and a surface water film index is generated, including the following steps:

[0017] The incident light and the analyzer were set to two working conditions: parallel polarization and orthogonal polarization. The parallel polarization reflection intensity and the orthogonal polarization reflection intensity were collected sequentially at the same detection point, and the two were normalized to obtain the polarization comparison value.

[0018] Cross-polarization or off-axis incident and acquisition geometry is used to suppress specular reflection, obtain the diffuse reflection intensity of the sample, and use a white board or built-in diffuse reflection standard as a reference to calculate the diffuse reflection ratio of the sample relative to the reference.

[0019] The polarization contrast value and diffuse reflectance ratio are normalized and combined according to the fusion rules obtained from the calibration to generate the surface water film index.

[0020] When the surface water film index exceeds the preset water film threshold, it is determined that a surface water film exists; when it is below the preset water film threshold, it is determined that no surface water film exists.

[0021] Furthermore, the process of weakening the surface water film includes the following steps:

[0022] While keeping the sample positioning and incident and acquisition geometry unchanged, the micro-thermal pulse unit is activated to apply one or more short-term heating pulses to the detection point. The pulse action area covers the detection point and does not extend beyond the preset boundary to avoid heat diffusion to non-test areas.

[0023] Micro-thermal pulses are generated by infrared micro-heating elements or miniature PTC heaters. The pulse duration and power are controlled to limit the temperature rise of the sample surface to a range that does not cause internal water migration and tissue denaturation. A stabilization time is set after the pulse ends to eliminate transient thermal disturbances.

[0024] The surface water film index is calculated before and after the pulse. When the decrease in the surface water film index reaches the preset water film threshold, it is determined that the water film has been effectively weakened. If the preset water film threshold is not reached, the micro-thermal pulse is repeated once without changing the incident and acquisition geometry.

[0025] After the film-cutting process is completed, data acquisition for state two is performed within the set waiting time window to obtain the reflectance spectra and color images of each band corresponding to state one, which are used to construct the differential dataset.

[0026] Furthermore, the two states are used to acquire multi-band reflectance spectra and color images respectively, and differential spectra and differential images are formed, including the following steps:

[0027] Multi-band reflectance spectra and color images were simultaneously acquired at the detection points. The raw data were corrected for dark field and white board respectively, and the exposure time, light source power, timestamp and detection point number were recorded.

[0028] After weakening the surface water film, data for state two was acquired while maintaining the same incident and acquisition geometry, focal length and exposure parameters, and the same dark field and whiteboard corrections as for state one were performed.

[0029] The two-state spectra are resampled and corrected according to a common wavelength grid, and the two-state color images are geometrically registered to establish a one-to-one correspondence between pixels;

[0030] A differential spectrum is constructed by subtracting the data of state 2 from that of state 1 wavelength by wavelength, and a differential image is constructed by subtracting each pixel by pixel.

[0031] Furthermore, based on a two-layer optical model of the surface film and the internal tissue, the differential data is inverted to obtain the surface film thickness, tissue absorption parameters, and tissue scattering parameters, including the following steps:

[0032] A two-layer optical model was established, with the sample surface layer modeled as an equivalent liquid water film and the lower layer modeled as a homogeneous scattering and absorbing tissue medium. The parameters to be determined were set as the surface film thickness, tissue absorption parameters, and tissue scattering parameters, and the known quantities were the device incident and acquisition geometry, the light source response, and the whiteboard calibration constant.

[0033] Differential spectroscopy is used as the primary constraint data, and brightness and chromaticity changes in the differential image are used as auxiliary constraints. When shallow and deep geometric acquisition data exist, they are incorporated into the joint objective to enhance the sensitivity to tissue layers.

[0034] The initial value and boundary of the surface film thickness are determined based on the surface water film index, and the initial values ​​of tissue absorption parameters and tissue scattering parameters are determined based on the device baseline obtained from the non-film reference sample or the built-in standard sample.

[0035] A nonlinear minimization fitting algorithm is used to solve the two-layer model, and a regularization term is introduced to suppress parameter correlation.

[0036] When using a joint objective, the solution is obtained by weighting the spectral domain residuals and the image domain residuals;

[0037] When the quality control criteria meet the preset quality threshold, the inversion results are output, and the surface film thickness, tissue absorption parameters and tissue scattering parameters are obtained.

[0038] Furthermore, internal moisture indicators and cell integrity indicators are extracted and adaptively corrected by combining the surface water film index with environmental parameters, including the following steps:

[0039] Based on the band intensity and band shape of tissue absorption parameters in the moisture-related near-infrared sensitive region, baseline correction and band consistency correction after dark field and whiteboard calibration are performed to generate an internal moisture indicator, which is then normalized to a fixed range for cross-batch comparison.

[0040] Based on the amplitude, wavelength variation trend, and scattering slope characteristics of tissue scattering parameters, combined with the local texture and edge integrity features of the differential image, a cell integrity indicator is generated and normalized.

[0041] The surface water film index is used as the weighting coefficient for the residual contribution of the surface layer, and the residual surface layer effect is deducted from the internal water indicator and the cell integrity indicator.

[0042] Using temperature, humidity, and dew point from environmental baseline data as adaptive correction factors, the baseline offset and range drift of the two indicators are jointly corrected by temperature, humidity, and dew point.

[0043] After adaptive correction, the final internal moisture indicator and cell integrity indicator are output and stored in association with the detection point number and timestamp for quality mapping and trend analysis.

[0044] Furthermore, a quality mapping model based on internal moisture indicators, tissue scattering parameters, and color parameters is established to output the quality grade and its changing trend during storage. When the quality indicators exceed the preset quality threshold, an early warning is triggered and storage control suggestions are provided, including the following steps:

[0045] Chromaticity parameters are calculated based on the difference image, the image is converted to a standard chromaticity space to obtain the luminance component and two chromaticity components, and the internal moisture index and tissue scattering parameters are combined to form a feature vector;

[0046] Select calibration samples to obtain reference hardness and color difference level labels, and pair the feature vectors with the reference hardness and reference color difference level labels to train the quality mapping model;

[0047] In online detection, the input feature vector outputs hardness estimation and color difference grade, and the change trend is calculated based on the sliding time window. The change trend represents the speed of quality evolution.

[0048] Set early warning judgment rules to trigger an early warning event when the hardness estimate is lower than the hardness threshold, the change trend exceeds the change threshold, or the color difference level reaches the critical level.

[0049] Based on environmental baseline data and early warning types, storage control recommendations are given, and early warnings and recommendations are archived along with monitoring point numbers and timestamps.

[0050] The technical effects and advantages of the physical parameter testing method for detecting quality changes in cucurbit vegetables during storage according to the present invention are as follows:

[0051] This invention achieves dynamic perception and trend prediction of quality changes in cucurbit vegetables during storage by constructing a physical parameter detection system based on a two-layer optical model and an adaptive correction mechanism. By configuring a stable multi-band light source, spectrometer, and imaging unit in a transparent measurement chamber, and combining environmental benchmark data collection of temperature, humidity, and dew point, the traceability and standardization of detection conditions are ensured. Furthermore, the surface water film index is generated by utilizing polarization contrast and diffuse reflectance ratio, and the surface water film is weakened under the action of micro-thermal pulses to form a two-state differential spectrum and differential image, thus eliminating the influence of surface interference on the detection of internal tissues.

[0052] Based on differential data inversion of a two-layer optical model, surface film thickness, tissue absorption parameters, and tissue scattering parameters are obtained. Internal moisture indicators and cell integrity indicators are extracted, and adaptive correction is performed by combining surface water film index and environmental baseline data to ensure the stability and consistency of test results in different batches and under different environments. On this basis, a quality mapping model integrating internal moisture, tissue scattering characteristics, and color parameters is constructed to output quality grades and trends. When quality indicators exceed preset quality thresholds, early warnings and storage control suggestions are generated, thus realizing closed-loop monitoring of quality changes based on optical detection and environmental perception. This effectively solves the detection bias problems caused by surface interference, environmental fluctuations, and single indicators in traditional detection, and significantly improves the accuracy of quality detection and the reliability of early warnings during the storage of cucurbit vegetables. Attached Figure Description

[0053] Figure 1 This is a flowchart illustrating a method for testing physical parameters to detect quality changes in cucurbit vegetables during storage, according to the present invention. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] In order to achieve the above objectives, Figure 1 A schematic diagram of the physical parameter testing method for detecting quality changes of cucurbit vegetables during storage is provided, which specifically includes the following steps;

[0056] Representative cucurbit vegetable samples were selected and the detection points were fixed. A stable multi-band light source, spectrometer and imaging unit were configured in the transparent measurement chamber to complete the dark field and white board calibration and lock the incident and acquisition geometric parameters.

[0057] The temperature, humidity and dew point information of the storage environment are recorded simultaneously. The presence of a water film on the sample surface is determined based on polarization contrast and diffuse reflectance ratio, and a surface water film index is generated.

[0058] Two-state rapid acquisition is performed on the same detection point. State 1 is the original state, and state 2 is the weakened surface water film state under the action of micro-thermal pulse. Multi-band reflectance spectrum and color image are acquired in the two states respectively, and differential spectrum and differential image are formed.

[0059] Based on a two-layer optical model of surface film and internal tissue, differential data is inverted to obtain surface film thickness, tissue absorption parameters and tissue scattering parameters. Internal water content and cell integrity indicators are extracted and adaptively corrected by combining surface water film index and environmental parameters.

[0060] Establish a quality mapping model based on internal moisture indicators, tissue scattering parameters, and color parameters. Output the quality grade and its changing trend during storage. When the quality indicators exceed the preset quality threshold, trigger an early warning and provide storage control suggestions.

[0061] Using bitter melon as the main sample, it is also compatible with similar crops such as winter melon and loofah.

[0062] Step 1: Select representative cucurbit vegetable samples and fix the detection points. Configure a stable multi-band light source, spectrometer, and imaging unit inside the transparent measurement chamber. Complete dark-field and white-board calibration and lock the incident and acquisition geometric parameters. Specific steps include:

[0063] First, after receiving the samples, an appearance screening is performed to remove individuals with mechanical damage, significant lesions, or abnormal morphology. Samples with intact skin and good representativeness are retained. To ensure consistency and traceability in repeated measurements, a unique number is assigned to each sample, and the detection points are fixed in representative areas of the sample skin using markings or coordinate indexing.

[0064] Specifically, the center of the detection point is marked on the sample surface with a removable micro-label or fine dot, and a local rectangular coordinate system is established with the longitudinal main axis of the sample and the most prominent longitudinal groove as two reference lines. The shortest distance and orientation of the detection point relative to the two reference lines are recorded to form a coordinate index.

[0065] Marking or coordinate indexing refers to using physical marks or coordinate records to determine the center of the detection point as a fixed position. These two methods can be used individually or in combination, but consistency must be maintained within the same batch. To avoid secondary positioning errors, the sample is placed in a mechanical positioning fixture during measurement. The mechanical positioning fixture limits the sample's rotation and axial position through a V-shaped bracket and end limit blocks that contact the sample at three points, ensuring that the sample maintains the same posture and reference plane relative to the device each time it is placed.

[0066] The transparent measurement chamber provides a stable and repeatable optical measurement environment. The inner walls of the chamber are treated with an anti-reflective coating to reduce stray reflections, and the chamber has a fixed mounting surface to support the optical components. The stable multi-band light source consists of a combination of visible and near-infrared light, and the light source is controlled to a stable output state through constant current drive and preheating. The spectrometer and imaging unit are fixedly arranged with a shared field of view, meaning they observe the same shared field of view area through coaxial or near-coaxial optical paths and are fixed by a rigid connection bracket to prevent relative displacement.

[0067] The measurement chamber is equipped with a polarization assembly, a white board and a dark field calibration board, as well as a mechanical positioning fixture. The polarization assembly is used for subsequent measurements requiring polarization conditions. The white board serves as a reflection reference. The dark field calibration board serves as a reference for dark current and stray light. The mechanical positioning fixture is used to fix the sample posture and detection point. To avoid semantic ambiguity, it is hereby stated that the incident and acquisition geometric parameters refer to the incident direction and angle of light from the light source to the sample surface, the acquisition direction and acquisition angle from the sample to the detector (spectrometer and imaging unit), and the corresponding working distance and common field of view. These parameters are locked with rigid brackets and stops after the device is installed and are not changed during routine measurements.

[0068] During the device startup phase, dark-field and white-board calibration is performed to lock the incident and acquisition geometric parameters. The dark-field and white-board calibration process is as follows: First, the stable multi-band light source is turned off or the incident light path is blocked. The detector end is aligned with the dark-field calibration board, and the dark-field baseline is recorded to subtract the detector dark current and environmental background. Then, the stable multi-band light source is turned on, the detector end is aligned with the white board, and the white board reflection baseline is recorded to establish a reference for reflectance conversion. During the calibration process, the spectrometer and imaging unit maintain a fixed arrangement with a shared field of view to ensure that the two types of data correspond to the same spatial region. If there is a slight deviation, the shared alignment target (including the central cross mark or regular pattern) is adjusted to align the centers of the fields of view of both until the mark within the shared field of view is centered in both the imaging unit and the spectrometer sampling area.

[0069] During measurement, place the marked or coordinate-indexed sample in the mechanical positioning fixture, aligning the coordinate index reference line with the fixture's reference edge. Slowly apply pressure to ensure full contact at the three points, confirming the center of the detection point is within the common field of view before data acquisition. Before each acquisition, verify the sample number and detection point number, and select the corresponding coordinate index record in the software. If the detection point center deviates from the common field of view center, fine-tune the sample's axis and rotation within the mechanical positioning fixture to bring the detection point center back to the field of view center before data acquisition.

[0070] For similar crops such as winter melon and loofah, the same sample numbering, marking or coordinate indexing, mechanical positioning and calibration process as bitter melon is still used. The only difference is that the surface characteristics of the crop are followed when selecting representative areas for point placement: for example, winter melon avoids areas covered with thick wax that may cause abnormal reflection, while for loofah, areas with stable texture between the longitudinal main ridge and the groove are selected as detection points. Regardless of the crop type, the device conditions remain unchanged, including a stable multi-band light source composed of a combination of visible and near-infrared light, a spectrometer and imaging unit with a fixed arrangement of common field of view, complete polarization components, white board and dark field calibration board, and mechanical positioning fixtures, and locked incident and acquisition geometric parameters. This ensures consistency and traceability across crops.

[0071] Step 2: Simultaneously record the temperature, humidity, and dew point information of the storage environment. Based on polarization contrast and diffuse reflectance ratio, determine whether a water film exists on the sample surface and generate a surface water film index. Specific steps include:

[0072] Temperature, humidity, and dew point sensors are fixedly installed inside the transparent measurement chamber. The probes of these three sensors are located near the sample but do not enter the optical path or imaging field of view, ensuring in-situ monitoring of the microenvironment around the sample without interfering with the incident and acquired geometric parameters.

[0073] After the device is powered on and preheated, it enters a stable state and begins to collect temperature, humidity, and dew point information in a synchronized manner with the optical acquisition. The three parameters are read immediately before each spectral and imaging acquisition and again after the acquisition ends. The average of these two readings is taken as the environmental record value for that optical acquisition. The collected temperature, humidity, and dew point information is associated with and stored along with the timestamp and detection point number to form environmental baseline data. The environmental baseline data record fields include at least: sample number, detection point number, timestamp, temperature, humidity, dew point, assembly number and status marker of the incident and acquisition geometric parameters, used for subsequent surface water film index discrimination and adaptive correction. To ensure traceability, the environmental baseline data and the original optical data are bound together using the same timestamp and detection point number key.

[0074] The polarization contrast value acquisition process is as follows: After completing the dark field and white board calibration and locking the incident and acquisition geometric parameters, reflection acquisition is performed sequentially at the same detection point under parallel polarization and orthogonal polarization conditions. Specifically, the azimuth angles of the polarizer and analyzer are set to be parallel to each other, and the parallel polarization reflection intensity is acquired; then, only the azimuth angle of the analyzer is changed to make it orthogonal to the polarizer, and the orthogonal polarization reflection intensity is acquired. The same exposure or integration time, the same light source power, and the same incident and acquisition geometric parameters are used for both acquisitions. The two reflection intensities are normalized and converted using the white board as a reference to make them comparable dimensionless quantities.

[0075] The polarization contrast value is calculated as follows: first, the difference between the parallel polarization reflection intensity and the orthogonal polarization reflection intensity is calculated, and then the sum of the two is used to scale the difference to obtain a normalized result between zero and one. This result is usually closer to one when there is a water film on the surface, and usually closer to zero when the surface is dry and close to no film.

[0076] The diffuse reflection ratio acquisition process is as follows: In order to suppress the interference caused by specular reflection, the diffuse reflection intensity is obtained at the same detection point by means of cross polarization or off-axis incident and acquisition geometry. In the cross polarization mode, the polarizer and analyzer are kept orthogonal. In the off-axis mode, without changing the locked state of the incident and acquisition geometry parameters, the off-axis accessory built into the device is used to make the acquisition direction avoid the specular reflection main lobe, while keeping the common field of view unchanged.

[0077] Then, using a whiteboard or a built-in diffuse reflection standard as a reference, the reflection intensity of the reference is first collected, and then the reflection intensity of the sample is collected. Both are then subjected to the same dark field and whiteboard processing. The diffuse reflection ratio is calculated as follows: the diffuse reflection intensity of the sample under the above geometric and polarization conditions is compared with the diffuse reflection intensity of the reference under the same conditions. A dimensionless result is obtained by comparing the ratio of the two. The closer the result is to one, the stronger the diffuse reflection of the sample surface, which is usually related to the presence of a water film on the surface. The smaller the result, the more likely the surface is to be dry.

[0078] The process of generating and judging the surface water film index is as follows: before being put into use, calibration is performed. Dry reference samples and film-containing reference samples (e.g., the same surface is first naturally dried, and then a very thin water film is uniformly formed on the surface) are selected to measure the polarization contrast value and diffuse reflectance ratio respectively. The measurement results of each type are normalized so that the two quantities of the dry reference sample are used as the zero end anchor point after normalization, and the two quantities of the film-containing reference sample are used as the one end anchor point after normalization.

[0079] The process of generating the surface water film index is as follows: First, the polarization contrast value and diffuse reflectance ratio obtained in this test are mapped to the interval between zero and one to keep them consistent with the aforementioned anchor point. Then, the two normalized results are weighted and synthesized according to the weights determined in the calibration stage to obtain a single index between zero and one, which is the surface water film index. The weights are determined based on the criterion of maximizing the distinguishability of the calibration sample between the dry reference sample and the film-covered reference sample. The weights are obtained by gradual adjustment and fixed in the device configuration file.

[0080] The preset water film threshold is defined as follows: in the calibration dataset, a threshold point that minimizes both the total number of false positives and false negatives is selected, or a point with comparable false positive and false negative rates is selected as the weighted threshold. This threshold is used for subsequent discrimination. In actual detection, when the surface water film index exceeds the preset water film threshold, it is determined that a surface water film exists; when the surface water film index is lower than the preset water film threshold, it is determined that no surface water film exists.

[0081] After each determination, the surface water film index, polarization contrast value, diffuse reflectance ratio, and corresponding environmental baseline data are bound and stored with the same timestamp and detection point number. Simultaneously, the polarization conditions used for acquisition, whether cross-polarization or off-axis incident was used, the acquisition geometry, the calibration batch number for white board and dark field, and the light source power and exposure or integration time settings are recorded to ensure that subsequent retests can replicate the same measurement conditions. The surface water film index is used as the trigger criterion for two-state differential acquisition (to determine whether it is necessary to weaken the surface water film).

[0082] Step 3: Perform two-state rapid acquisition on the same detection point. State 1 is the original state, and State 2 is the weakened surface water film state under the action of micro-thermal pulses. Multi-band reflectance spectra and color images are acquired for the two states respectively, and differential spectra and differential images are formed. The specific steps include:

[0083] The process of using micro-thermal pulses to weaken the surface water film and confirm its effectiveness is as follows:

[0084] After completing the dark field and whiteboard calibration and locking the incident and acquisition geometry parameters, the sample is placed in the mechanical positioning fixture, ensuring the predetermined detection point is located at the center of the common field of view. To maintain consistency in comparison, the multi-band light source is stabilized and operates at constant power, and the exposure or integration time of the spectrometer and imaging unit remains fixed. The micro-thermal pulse unit is activated to apply a short heating pulse to the detection point; the micro-thermal pulse is generated by an infrared micro-heating element or a miniature PTC heater, and the heating area covers the detection point and does not extend beyond the preset boundary to prevent heat from diffusing into non-test areas.

[0085] The duration and power of the pulse are limited by the device control terminal to a range where the sample surface temperature rise is sufficient to weaken the surface water film without causing internal water migration and tissue degeneration. After the pulse ends, a set waiting time window is entered to allow the transient thermal disturbance to decay naturally. The surface water film index is calculated before and after each pulse, and whether the decrease in the surface water film index reaches the preset water film threshold is used as the criterion.

[0086] When the reduction reaches or exceeds the preset water film threshold, the water film is determined to be effectively weakened.

[0087] If the preset water film threshold is not reached, repeat the micro-thermal pulse once without changing the incident and acquisition geometric parameters, the light source power and the exposure or integration time, until the criterion is met or the upper limit of the safe retry number is reached. After the film trimming process is completed and the surface water film index confirms that the film has been effectively weakened, immediately enter the data acquisition of state two.

[0088] Two-state fast acquisition follows the principle of first state one, then state two, with all conditions identical except for the state itself. State one is the original state.

[0089] Without any film removal process, multi-band reflectance spectra and color images are simultaneously acquired at the detection points, and exposure or integration time, light source power, timestamp and detection point number are recorded; the acquired raw data are then subjected to dark field and white board correction according to a predetermined process to obtain benchmark data for comparison.

[0090] Subsequently, the micro-thermal pulse film was trimmed as described above, and its effectiveness was confirmed by the surface water film index. Within the set waiting time window, without changing the incident and acquisition geometric parameters, the focal length and exposure or integration time, or the power of the light source, the multi-band reflectance spectrum and color image of state two were acquired, and the same dark field and whiteboard correction as state one were performed.

[0091] It should be noted that, to maintain the comparability of the two-state data in space and time, the interval between the acquisition of state 2 and state 1 must be controlled by the timing of the micro-thermal pulse and the waiting time window, and must be completed under the same stable multi-band light source operating condition.

[0092] The process of establishing a common wavelength grid and image geometric registration is as follows:

[0093] To construct a difference result that can be compared point by point, the two-state spectra are first resampled and corrected using a common wavelength grid.

[0094] The common wavelength grid maps the spectral data acquired twice to the same set of discrete wavelength points according to the device's wavelength calibration table. The resampling process uses a combination of interpolation and intensity consistency correction to eliminate misalignment caused by slight differences between the detector's discrete sampling and the wavelength scale, so that each discrete wavelength point has two equally corrected reflection intensity values, one for state one and one for state two.

[0095] The image end uses geometric registration to establish a one-to-one correspondence between pixels: taking the first state image as a reference, using fixed markers or boundary textures in the common field of view as features, a small range of transformations involving only translation, rotation, and scaling are performed to make the second state image coincide with the first state image in the reference coordinate system; the criterion for the end of registration is that the feature point deviation does not exceed the pixel-level positioning accuracy of the device and the edge overlap error after registration is within the allowable range of the device. After completion, two state image data with one-to-one pixel correspondence in the same coordinate system can be obtained.

[0096] At the spectral end, a differential spectrum is constructed by comparing state 2 with state 1 wavelength by wavelength. The textual calculation process of the differential spectrum is as follows: for each discrete wavelength point on the common wavelength grid, the corrected reflection intensity of state 2 is taken as the comparison quantity, and the corrected reflection intensity of state 1 is taken as the reference quantity. The difference between the two is calculated and arranged in order of wavelength to obtain the differential spectrum. If necessary, the deviation of the ratio between the comparison quantity and the reference quantity can be generated at the same time as a backup result of the ratio-type differential spectrum for cross-batch robustness assessment.

[0097] At the image end, a difference image is constructed by subtracting pixels one by one. The specific process is as follows: for the same pixel position after registration, the corrected pixel value of state two is subtracted from the corrected pixel value of state one, and the result is used as the difference value of the pixel. After traversing all pixels, a difference image is formed.

[0098] Set signal-to-noise and stability criteria: including whether the baseline fluctuation of the differential spectrum in the non-absorption region is lower than the device's specified threshold, whether the residual of the differential image in the background region is close to zero, and whether the light source and exposure or integration time records of the two-state acquisition are consistent. If any criterion is not met, the subsequent inversion process will not be entered, but a review will be prompted: first check whether the incident and acquisition geometric parameters have been changed unexpectedly, whether the light source power and exposure or integration time are consistent, and whether the registration meets the standards; if necessary, re-execute the state two acquisition or restart the two-state process from state one.

[0099] Differential spectra and differential images must be traceably linked to their source data and environmental information. The differential spectra, differential images, raw and corrected data of state one and state two, surface water film index (state one, state two and their reduction magnitude), preset water film threshold, micro-thermal pulse duration and power, set waiting time window, assembly number of incident and acquisition geometric parameters, light source power, exposure or integration time, timestamp and detection point number are all written into the same recording unit. The reduction magnitude of the surface water film index is defined as the result of subtracting the surface water film index of state two from the surface water film index of state one, which is used to prove the effectiveness of film removal.

[0100] For example, after completing the film trimming and two-state difference at the same detection point in the cold storage site, and after completing the dark field and white board calibration and locking the incident and acquisition geometric parameters in the transparent measurement chamber, the bitter gourd sample with coordinate index is placed into the mechanical positioning fixture so that the predetermined detection point enters the center of the common field of view; temperature, humidity and dew point information are collected simultaneously and environmental baseline data is generated.

[0101] First, acquire state-one data: A stable multi-band light source maintains constant power. The spectrometer and imaging unit synchronously acquire multi-band reflectance spectra and color images and complete corrections under predetermined exposure or integration times. Then, perform film reduction: Start the micro-thermal pulse unit to apply a short heating pulse to the detection point. The pulse area covers the detection point and does not cross the preset boundary. After the pulse ends, enter a set waiting time window to eliminate transient thermal disturbances. Calculate the surface water film index before and after the pulse. If the decrease in the surface water film index reaches the preset water film threshold, it is determined that the water film has been effectively weakened.

[0102] Immediately acquire state 2 data: Under the condition of not changing the incident and acquisition geometric parameters, and not changing the light source power and exposure or integration time, simultaneously acquire multi-band reflectance spectra and color images and complete the correction. Then, establish a common wavelength grid and image geometric registration for the two-state data: resample the two spectra to the same set of discrete wavelength points, and establish a one-to-one pixel correspondence between the two images in the same coordinate system;

[0103] Finally, the differential results are constructed: the differential spectrum is obtained by subtracting state 1 from state 2 for each wavelength, and the differential image is obtained by subtracting each pixel one by one. The quality criteria such as baseline fluctuation, background residual and consistency with acquired metadata are checked. If the preset quality threshold is met, the differential spectrum and differential image are archived together with the timestamp, detection point number, reduction of surface water film index and environmental baseline data.

[0104] Through the above-mentioned consistency control, two-state rapid acquisition, common wavelength grid and geometric registration, differential construction by wavelength and by pixel, and traceable storage of full metadata, the differential dataset is ensured to meet the stability and comparability requirements of subsequent inversion.

[0105] Step 4: Based on the two-layer optical model of the surface film and internal tissue, the differential data is inverted to obtain the surface film thickness, tissue absorption parameters, and tissue scattering parameters. Internal water content and cell integrity indicators are extracted, and adaptive correction is performed by combining the surface water film index with environmental parameters. Specific steps include:

[0106] After completing the two-state acquisition and differential construction, the differential spectrum is used as the main constraint data, and the brightness and chromaticity changes in the differential image are used as auxiliary constraints to enter the inversion process.

[0107] The two-layer optical model divides the sample into two layers: the outermost layer is an equivalent liquid water film, the unknown of which is the surface film thickness (used to quantify the contribution of the residual surface water film to reflection); the lower layer is a homogeneous scattering and absorbing tissue medium, the unknowns of which are the tissue absorption parameters and tissue scattering parameters (both vary with wavelength, are taken point-by-point on a discrete wavelength grid, forming two vectors that correspond one-to-one with the wavelength). On the device side, the incident and acquisition geometry, the light source response, and the whiteboard calibration constant are fixed as known quantities during the calibration phase and used to convert the detection signal to a comparable reflection scale. On the data side, the differential spectrum obtained from the two-state acquisition is used as the main constraint, and the brightness and chromaticity changes in the differential image are used as auxiliary constraints. If there is geometric acquisition data from shallow and deep layers, it is used as additional constraints that are more sensitive to the tissue layer.

[0108] When establishing the two-layer optical model, the sample surface is equivalent to a liquid water film, and its unknown quantity is defined as the surface film thickness (representing the thickness of the equivalent liquid water layer, used to quantitatively describe the optical contribution of the residual water film on the surface); the sample body below the film is approximated as a homogeneous scattering and absorbing tissue medium, and its unknown quantities are defined as the tissue absorption parameter and the tissue scattering parameter (representing the strength of effective absorption and effective scattering of incident radiation within the operating wavelength range of the device and their variation with wavelength).

[0109] The incident and acquisition geometry, light source response, and whiteboard calibration constant of the device have been determined during the calibration phase and are considered as known quantities for the solution. If the device obtains shallow and deep geometric acquisition data during the acquisition phase, these data are included in the inversion target to enhance the sensitivity to the tissue layer response. To ensure that the inversion is convergent and physically reasonable, the initial value and boundary of the surface film thickness are first set based on the calculated surface water film index (for example, when the surface water film index is high, the initial value is taken as a relatively large film thickness range). At the same time, the initial values ​​and variation range of the tissue absorption parameters and tissue scattering parameters are given by the device baseline obtained from the film-free reference sample or the built-in standard sample.

[0110] The inversion is solved using a nonlinear minimization fitting algorithm. The process is as follows: a two-layer optical model is used to give the predicted values ​​of the differential spectrum and the predicted values ​​of the brightness and chromaticity changes of the differential image for the same set of unknowns. The difference between the predicted value and the measured value is calculated and all the differences are accumulated into a comprehensive residual. When there are two types of data, spectral and image, the spectral domain residual and the image domain residual are weighted according to the weights determined in the calibration stage to form a single joint objective.

[0111] To suppress the correlation between tissue absorption and scattering parameters, a regularization term is introduced, and non-negativity and band smoothing constraints are applied to the unknowns to ensure that their shape varies with wavelength conforms to the physical characteristics of a continuous medium. The solution algorithm progressively updates the unknowns until convergence conditions are met. Convergence conditions are determined by a combination of three quantitative indicators: the magnitude of the comprehensive residual decreases to a preset range, the parameter change magnitude in two consecutive iterations is lower than a preset threshold, and the numerical stability, expressed as condition number or equivalent stability measure, reaches a preset quality threshold. If any indicator is not met, the algorithm automatically backtracks, adjusts the initial values ​​or weights, and repeats the solution. If the conditions are still not met after multiple repetitions, the algorithm prompts for a review of the two-state data quality and requests the acquisition of differential data again according to the two-state acquisition process. After meeting the quality control criteria, the algorithm outputs the surface film thickness, tissue absorption parameters, and tissue scattering parameters, along with uncertainty and stability indicators for subsequent weighting and correction.

[0112] The steps for generating internal water indicators and cell integrity indicators are as follows:

[0113] After obtaining the inversion results, the internal moisture index is generated based on the band intensity and band shape of the tissue absorption parameters in the moisture-related near-infrared sensitive region.

[0114] The generation process is as follows: the baseline of the spectral line of the tissue absorption parameters is corrected using dark field and white plate calibration information to eliminate the overall rise or fall caused by device drift; then, band shape consistency correction is performed on the sensitive region related to moisture to bring the band width and peak position shift back to the allowable range of the calibration reference; on this basis, the weighted combination of the representative value of the band intensity and the consistency of the band shape in this region is used as the initial value of the internal moisture indicator, and it is normalized according to the upper and lower limits determined during device calibration so that samples from different batches can be compared within a fixed range.

[0115] The cell integrity indicator is jointly generated from tissue scattering parameters and differential image features: First, the amplitude level and slope features as a function of wavelength are extracted from the tissue scattering parameters. The amplitude represents the overall scattering intensity, and the slope represents the trend of changes in particle size and tissue structure scale. Then, local texture and edge integrity features are extracted from the differential image (within the neighborhood of the detection point, it is statistically analyzed whether the differential response from state one to state two in the boundary region and texture region remains continuous and whether structural breaks occur). The two results are combined according to the calibrated weights to form the initial value of the cell integrity indicator, and normalization is also performed for cross-batch comparison.

[0116] The adaptive correction steps based on surface water film index and environmental baseline data are as follows: The surface water film index is used as a weighting coefficient for the contribution of surface residue. Surface residue is then subtracted from the internal moisture indicator and cell integrity indicator obtained in the previous step. The process is as follows:

[0117] When the surface water film index is close to zero, the residual deduction amount approaches zero. When the surface water film index increases, the share corresponding to the residual surface contribution is deducted from the two indicated amounts according to the calibrated proportional relationship. The proportional relationship is fixedly stored in the device configuration file.

[0118] Subsequently, environmental adaptive correction was performed on the two indicators: environmental reference data bound to the same timestamp and the same test point number as this test was read, and the temperature, humidity and dew point were used as correction factors to correct the baseline offset and range drift of the two indicators, respectively. The correction process is described in words as follows: during the device calibration stage, a one-to-one correspondence was established between the influence of temperature on the baseline and the influence of humidity and dew point on the range, and during online testing, the two indicators were moved to the reference state consistent with the calibration reference based on the current environmental reference data.

[0119] After completing the above adaptive correction, the final internal moisture indicator and cell integrity indicator are output and stored in association with the detection point number and timestamp to form a traceable record for subsequent quality mapping and trend analysis.

[0120] It should be noted that, in order to improve the reliability of the results, the uncertainty and stability indicators output during the inversion stage can be weighted by confidence level for the two indicators. When the weighted consistency index does not reach the preset threshold, a recalculation is triggered, and if necessary, the system will prompt the user to repeat the data collection process according to the two-state acquisition procedure.

[0121] Step 5: Establish a quality mapping model based on internal moisture indicators, tissue scattering parameters, and color parameters. During storage, output the quality grade and its changing trend. When quality indicators exceed preset quality thresholds, trigger an early warning and provide storage control suggestions. Specific steps include:

[0122] For the input features used in quality mapping, the internal moisture index and tissue scattering parameters are taken. Based on the requirements for calculating chromaticity parameters using the difference image, the difference image is converted to a standard chromaticity space to obtain the luminance component and two chromaticity components. These three are collectively referred to as chromaticity parameters. The specific steps are as follows:

[0123] Perform the same operation on the difference image as the original Figure 1 After the dark field and whiteboard calibration, the device calls the built-in standard color space conversion process (which is determined and fixed in the factory calibration) to obtain the luminance component and two color components pixel by pixel. The average of the neighborhood of the detection point is weighted by area and used as the color parameter of that point. Then, a feature vector for training is constructed. This feature vector is formed by splicing the internal moisture indicator, tissue scattering parameters and color parameters in a predetermined order.

[0124] Obtain supervisory labels, select representative calibration samples, and acquire reference hardness and reference color difference grade labels for each testing point using an industry-standard and traceable method:

[0125] The reference hardness is derived from a calibrated mechanical testing device or an industry-recognized equivalent method. The reference color difference grade label is derived from the grade given by a trained evaluation process based on a comparison with a standard color card or standard image. The feature vector is paired one by one with the reference hardness and reference color difference grade label and input into the training process of the quality mapping model.

[0126] The training process completes the model type selection and parameter determination without changing the meaning and name of the features, and confirms the generalization ability by reserving samples or cross-validation.

[0127] After training is complete, the version number of the standard chromaticity space conversion process used by the quality mapping model for generating chromaticity parameters, as well as the list of calibration samples used during training, will be archived together.

[0128] In actual storage testing, internal moisture indicators and tissue scattering parameters are generated in real time for each testing point according to the timestamp sequence. Following the same steps as the offline stage, chromaticity parameters (i.e., the result of averaging the luminance component and the two chromaticity components by neighborhood area weighting) are generated from the current differential image. The three are combined in a predetermined order to form a feature vector and input into the quality mapping model to obtain two outputs: hardness estimation and color difference grade.

[0129] The changing trend is calculated, and a sliding time window is maintained for each detection point. The sliding time window covers the most recent detection results for that point with a fixed window length, and slides forward one time step when a new result arrives. The changing trend is calculated by subtracting the previous and next values ​​of the same indicator (e.g., hardness estimation) sorted by timestamp within the window, and taking the signed average of these adjacent differences as the changing trend of that indicator. Positive values ​​indicate that the indicator increases over time, negative values ​​indicate that the indicator decreases over time, and the larger the absolute value, the faster the change. The changing trends of hardness estimation and color difference grade can be obtained in this way. The hardness estimation, color difference grade, and their respective changing trends are recorded together with the detection point number and timestamp for subsequent early warning judgment and traceability.

[0130] During online operation, a warning event is triggered based on the pre-set warning judgment rules: when the hardness estimate is lower than the hardness threshold determined and solidified in the calibration stage, or when the absolute value of the change trend of any index exceeds the corresponding change threshold, or when the color difference level reaches or exceeds the critical level determined in the calibration stage.

[0131] Once triggered, the corresponding warning type is invoked and combined with the environmental baseline data (including temperature, humidity and dew point) bound to this detection to generate on-site storage control recommendations;

[0132] When environmental baseline data shows that the temperature is above the target range and the hardness estimate is decreasing or showing a rapid decreasing trend, it is recommended to lower the temperature to the target range and keep it stable.

[0133] When environmental baseline data shows high humidity or dew point and an increasing or rapidly increasing color difference level, it is recommended to appropriately reduce humidity or increase ventilation to reduce surface condensation.

[0134] When both types of problems occur simultaneously, prioritize the factors that have a more direct impact on quality according to the warning type, and retest and confirm after handling.

[0135] All early warning events and storage control recommendations must be archived along with the testing point number and timestamp, and the version information of the hardness threshold, change threshold and critical level used at that time must also be recorded to ensure that subsequent reviews can restore the judgment basis and verify the consistency with the quality mapping model.

[0136] For example, for the same batch of samples, based on the generated internal moisture indicator, the luminance component and two chromaticity components are obtained by converting the differential image to the standard chromaticity space after each detection, and used as chromaticity parameters; then the internal moisture indicator, tissue scattering parameters and chromaticity parameters are concatenated into a feature vector in a predetermined order and input into the quality mapping model.

[0137] During online operation, a feature vector is assembled for each detection point in a predetermined order after each acquisition: internal moisture indicator, tissue scattering parameter and chromaticity parameter (weighted average of the neighborhood area of ​​the luminance component and the two chromaticity components), input into the quality mapping model that has been trained and archived offline, and outputs hardness estimate and color difference grade in real time.

[0138] Each testing point is configured with a maintenance sliding time window: Historical hardness estimates and color difference grades are sorted by timestamp, and the values ​​of each estimate are subtracted sequentially and a signed average is calculated to obtain the trends in hardness estimation and color difference grade. Positive values ​​indicate an increase, negative values ​​indicate a decrease, and larger absolute values ​​indicate a faster rate of change. A warning judgment is then made based on these trends.

[0139] When the hardness estimate is lower than the hardness threshold, or the absolute value of any trend exceeds the change threshold, or the color difference level reaches the critical level, an early warning event is triggered. After triggering, the environmental baseline data associated with this test is read, and storage control suggestions are given based on the warning type: for example, when the hardness estimate decreases and the trend is rapid, and the environmental baseline data shows that the temperature is high, it is recommended to lower the temperature and keep it stable; when the color difference level increases and the trend is rapid, and the environmental baseline data shows that the humidity or dew point is high, it is recommended to reduce the humidity or increase the ventilation to reduce surface condensation.

[0140] Early warning events and storage control recommendations are archived along with the monitoring point number and timestamp, while also recording the version information of the hardness threshold, change threshold, and critical level used in that instance.

[0141] It should be noted that the threshold information in this embodiment was set in advance by professionals and will not be explained in detail here. Some parameters in the embodiment may have the same English letters, but they are explained with different meanings when used, and will not be explained one by one here.

[0142] This invention achieves dynamic perception and trend prediction of quality changes in cucurbit vegetables during storage by constructing a physical parameter detection system based on a two-layer optical model and an adaptive correction mechanism. By configuring a stable multi-band light source, spectrometer, and imaging unit in a transparent measurement chamber, and combining environmental benchmark data collection of temperature, humidity, and dew point, the traceability and standardization of detection conditions are ensured. Furthermore, the surface water film index is generated by utilizing polarization contrast and diffuse reflectance ratio, and the surface water film is weakened under the action of micro-thermal pulses to form a two-state differential spectrum and differential image, thus eliminating the influence of surface interference on the detection of internal tissues.

[0143] Based on differential data inversion of a two-layer optical model, surface film thickness, tissue absorption parameters, and tissue scattering parameters are obtained. Internal moisture indicators and cell integrity indicators are extracted, and adaptive correction is performed by combining surface water film index and environmental baseline data to ensure the stability and consistency of test results in different batches and under different environments. On this basis, a quality mapping model integrating internal moisture, tissue scattering characteristics, and color parameters is constructed to output quality grades and trends. When quality indicators exceed preset quality thresholds, early warnings and storage control suggestions are generated, thus realizing closed-loop monitoring of quality changes based on optical detection and environmental perception. This effectively solves the detection bias problems caused by surface interference, environmental fluctuations, and single indicators in traditional detection, and significantly improves the accuracy of quality detection and the reliability of early warnings during the storage of cucurbit vegetables.

[0144] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0145] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0146] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0147] The above description is merely a specific embodiment 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.

[0148] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for testing physical parameters for detecting quality changes during storage of melon vegetables, characterized by: The method comprises the following steps: Representative melon vegetable samples are selected and fixed detection points are set, a stable multi-band light source, a spectrometer and an imaging unit are arranged in a transparent measurement cabin, dark field and white board calibration are completed and incident and collection geometric parameters are locked; Temperature, humidity and dew point information of the storage environment are recorded synchronously, whether a water film exists on the surface of the sample is determined based on the polarization contrast and the diffuse reflectance ratio, and a surface water film index is generated; Two-state rapid collection is performed on the same detection point, state one is the original state and state two is the weakened surface water film state under the action of a micro-heat pulse, multi-band reflectance spectra and color images are obtained respectively in the two states, and difference spectra and difference images are formed; Based on a two-layer optical model of the surface film and the internal tissue, the difference data are inverted to obtain the surface film thickness, the tissue absorption parameters and the tissue scattering parameters, the internal water content indicator and the cell integrity indicator are extracted, and the adaptive correction is performed in combination with the surface water film index and the environmental parameters; A quality mapping model based on the internal water content indicator, the tissue scattering parameters and the colorimetric parameters is established, the quality grade and the change trend are output during storage, and a warning is triggered and storage control suggestions are provided when the quality index exceeds the preset quality threshold; The melon vegetable samples mainly include bitter gourd and include wax gourd and silk gourd, the detection points are fixed on the representative areas of the sample surface by marking or coordinate indexing, and the consistency and traceability of repeated measurements are maintained.

2. The method for testing physical parameters for detecting quality changes during storage of melon vegetables according to claim 1, characterized in that: The stable multi-band light source in the transparent measurement cabin is composed of visible light and near-infrared light, the spectrometer and the imaging unit are arranged in a common field of view and fixed, and the incident and collection geometry with the sample surface is locked; a polarization component, a white board and a dark field calibration board and a mechanical positioning clamp are arranged in the measurement cabin to complete the dark field and white board calibration and ensure the repeated positioning of the detection points.

3. The method for testing physical parameters for detecting quality changes of melon vegetables during storage according to claim 2, characterized in that: The temperature, humidity and dew point information is collected by arranging temperature sensors, humidity sensors and dew point sensors in the transparent measurement cabin. The collected temperature, humidity and dew point information, time stamp and detection point number are associatedly stored to generate environmental reference data, which is used for surface water film index discrimination and subsequent adaptive correction.

4. The method for testing physical parameters for detecting quality changes during storage of melon vegetables according to claim 3, characterized in that: Whether a water film exists on the surface of the sample is determined based on the polarization contrast and the diffuse reflectance ratio, and a surface water film index is generated, comprising the following steps: The incident light and the analyzer are set to parallel polarization and orthogonal polarization respectively, the parallel polarization reflectance intensity and the orthogonal polarization reflectance intensity of the same detection point are collected in turn, and the two are normalized to obtain the polarization contrast; The cross-polarization or off-axis incident and collection geometry is used to suppress the specular reflection, the diffuse reflectance intensity of the sample is obtained, and the diffuse reflectance ratio of the sample relative to the reference is calculated by taking the white board or the built-in diffuse reflectance standard as the reference; The polarization contrast and the diffuse reflectance ratio are normalized and combined according to the fusion rule obtained by calibration to generate the surface water film index; When the surface water film index exceeds the preset water film threshold, it is determined that a surface water film exists, and when it is lower than the preset water film threshold, it is determined that a surface water film does not exist.

5. The method for testing physical parameters for detecting quality changes during storage of melon vegetables according to claim 4, characterized in that: The process of weakening the surface water film comprises the following steps: Under the condition of keeping the sample positioning and the incident and collection geometry unchanged, a micro-heat pulse unit is started to apply one or more short-time heating pulses to the detection point, and the pulse action area covers the detection point and expands no more than the preset boundary to avoid heat diffusion to the non-measurement area; The micro-heat pulse is generated by an infrared micro-heating element or a micro-PTC heater, and the pulse length and power are controlled to limit the temperature rise of the sample surface within a range that does not cause internal water migration and tissue degeneration, and a stabilization time is set after the pulse to eliminate transient thermal disturbance; The surface water film index is calculated before and after the pulse, and when the surface water film index decreases to a preset water film threshold, it is determined that the water film is effectively weakened; if the preset water film threshold is not reached, a micro-heat pulse is repeated under the condition of not changing the incident and collection geometry; After the film weakening process is completed, data acquisition in state two is performed within a set waiting time window, and the multi-band reflectance spectra and color images corresponding to state one are obtained to construct a difference data set.

6. The method for testing physical parameters for detecting quality changes during storage of melon vegetables according to claim 5, characterized in that: The two states respectively acquire multi-band reflectance spectra and color images, and form difference spectra and difference images, including the following steps: In state one, multi-band reflectance spectra and color images are synchronously acquired at the detection point, and the original data are respectively dark field and white board corrected, and the exposure time, light source power, time stamp and detection point number are recorded; After the surface water film is weakened, the data of state two are acquired under the condition of keeping the incident and collection geometry, focal length and exposure parameters consistent, and the same dark field and white board correction as state one is performed; The two-state spectra are resampled and corrected according to the common wavelength grid, and the two-state color images are geometrically registered to establish a one-to-one correspondence relationship; The difference spectra are constructed by subtracting the data of state two from state one wavelength by wavelength, and the difference images are constructed by subtracting the pixels one by one.

7. The method for testing physical parameters for detecting quality changes during storage of melon vegetables according to claim 6, characterized in that: Based on the two-layer optical model of the surface film and the internal tissue, the difference data are inverted to obtain the surface film thickness, tissue absorption parameter and tissue scattering parameter, including the following steps: A two-layer optical model is established, the sample surface is modeled as an equivalent liquid water film, the lower layer is modeled as a homogeneous scattering and absorbing tissue medium, the parameters to be solved are set as the surface film thickness, tissue absorption parameter and tissue scattering parameter, and the device incident and collection geometry, light source response and white board calibration constant are taken as known quantities; The difference spectra are taken as the main constraint data, and the brightness and chroma changes in the difference images are taken as auxiliary constraints; when there are shallow and deep geometric collection data, they are integrated into the joint target to enhance the sensitivity to the tissue layer; The initial value and value boundary of the surface film thickness are determined according to the surface water film index, and the initial values of the tissue absorption parameter and the tissue scattering parameter are determined by the device baseline obtained from the film-free reference sample or the built-in standard sample; A nonlinear minimization fitting algorithm is used to solve the two-layer model, and a regularization term is introduced to suppress parameter correlation; When the joint target is used, the spectral domain residual and the image domain residual are weighted to solve; When the quality control criterion meets the preset quality threshold, the inversion result is output, and the surface film thickness, tissue absorption parameter and tissue scattering parameter are obtained.

8. The method for testing physical parameters for detecting quality changes during storage of melon vegetables according to claim 7, characterized in that: The internal water content indicator and the cell integrity indicator are extracted, and the surface water film index and the environmental parameters are adaptively corrected, including the following steps: Based on the band intensity and band shape of the tissue absorption parameters in the near-infrared sensitive range related to water, the baseline correction and band consistency correction after dark field and white plate calibration are performed to generate the internal water content indicator, which is normalized to a fixed interval for cross-batch comparison; Based on the amplitude, wavelength variation trend and scattering slope characteristics of the tissue scattering parameters, combined with the local texture and edge integrity characteristics of the difference image, the cell integrity indicator is generated and normalized; The surface water film index is used as a weight coefficient of the surface residual contribution to deduct the residual surface effect of the internal water content indicator and the cell integrity indicator; The temperature, humidity and dew point in the environmental reference data are used as adaptive correction factors to jointly correct the baseline shift and range drift of the two indicators with temperature, humidity and dew point; After adaptive correction, the final internal water content indicator and cell integrity indicator are output, and are associated with the detection point number and time stamp for quality mapping and trend analysis.

9. The method for testing physical parameters for detecting quality changes during storage of melon vegetables according to claim 8, characterized in that: A quality mapping model based on internal water content indicator, tissue scattering parameter and color parameter is established to output quality grade and change trend during storage, and to trigger an early warning and provide storage control suggestions when the quality index exceeds the preset quality threshold, including the following steps: The color parameter is calculated according to the difference image, the image is converted to the standard color space to obtain the brightness component and two color components, and the internal water content indicator and the tissue scattering parameter are combined to form a feature vector; The reference hardness and color difference grade labels are obtained by selecting calibration samples, and the feature vector is paired with the reference hardness and reference color difference grade label to train the quality mapping model; In online detection, the feature vector is input, and the hardness estimate and color difference grade are output, and the change trend is calculated based on a sliding time window, which represents the quality evolution speed; Set the early warning judgment rule, when the hardness estimate is lower than the hardness threshold, the change trend exceeds the change threshold, or the color difference grade reaches the critical level, trigger the early warning event; According to the environmental reference data and the type of early warning, storage control suggestions are given, and the early warning and suggestions are associated with the detection point number and time stamp for archiving.

Images