Egg freshness detection method based on intelligent sensing system

Abstract

The present application relates to the technical field of intelligent sensing detection and nondestructive detection of agricultural product quality, in particular to a method for detecting the freshness of eggs based on an intelligent sensing system; the method comprises the following steps: a piezoelectric sensing array is laid under an excitation track with a preset spatial periodic sine wave or a tooth-shaped wave, the force change in the vertical direction of the eggs when rolling is collected in real time, and a dynamic contact force time sequence signal is generated; a photoelectric speed measurement module is used to calculate the real-time rolling linear speed, and the dynamic excitation frequency is calculated in combination with the spatial period length; the frequency is taken as a reference, orthogonal demodulation and Hilbert transformation are performed on the time sequence signal, high-frequency carrier feature data and low-frequency envelope feature data are extracted; the two are input into a preset classification model, an internal rheological state evaluation result is output, the freshness grade is determined, and the determination boundary parameters of the preset classification model are dynamically calibrated; the present application solves the problem that the traditional detection method cannot simultaneously consider the detection speed and determination consistency.

Description

Technical Field

[0001] This invention relates to the field of intelligent sensing and detection and non-destructive testing of agricultural product quality, specifically a method for detecting the freshness of eggs based on an intelligent sensing system. Background Technology

[0002] With the automation upgrade of the egg processing industry, the processing scale and speed of sorting lines have been significantly improved; this high-throughput production environment poses a severe challenge to the real-time monitoring of egg quality, especially in the online assessment of the internal freshness of eggs. Currently, the industry mainly assesses egg quality through manual sampling or traditional physical testing methods. Technicians typically use sensory inspection, light penetration, or tapping and listening to make judgments and manually record the results. However, traditional manual monitoring and evaluation methods rely heavily on operator experience, resulting in low testing efficiency and difficulty in performing comprehensive, standardized, and real-time assessments of each egg without interrupting the production line. Furthermore, because eggs have a closed multiphase structure, subtle changes in the degree of protein denaturation and moisture evaporation are easily masked by mechanical noise and individual quality differences during continuous rolling. In addition, the lack of efficient online sensing mechanisms and automated feature extraction methods makes it difficult for the system to detect eggs with abnormal internal rheological states in a timely manner during dynamic production. Therefore, how to overcome environmental interference and accurately capture the quality changes inside eggs in a continuously rotating industrial setting, and achieve automated grading of freshness, has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a method for detecting egg freshness based on an intelligent sensing system. Specifically, the technical solution of this invention includes: Used for online quality assessment of eggs as the target analyte, including: By laying a piezoelectric sensor array under an excitation track with a preset spatial period sine wave or toothed corrugation, the vertical force change of the target object during rolling is collected in real time, and a dynamic contact force timing signal is generated. The photoelectric speed measurement module set on the side of the excitation track captures the movement time of the target object as it passes through the preset detection interval on the excitation track, and calculates the real-time rolling linear velocity data. Based on real-time rolling linear velocity data and the spatial period length of the preset spatial period ripple, the dynamic excitation frequency of the target object is calculated. Using the dynamic excitation frequency as a reference signal, the dynamic contact force time-series signal is frequency-domain isolated and demodulated. High-frequency carrier feature data characterizing the coupling response between the outer shell and the internal medium of the target object, and low-frequency envelope feature data characterizing the internal center of gravity shift of the target object are extracted respectively. The high-frequency carrier feature data includes the effective value of energy and the phase angle, and the low-frequency envelope feature data includes the variance. High-frequency carrier feature data and low-frequency envelope feature data are input into a preset classification model, and the internal rheological state assessment results of the target object are output; based on the internal rheological state assessment results, the freshness level of the target object is determined.

[0004] Preferably, the specific steps for extracting high-frequency carrier feature data and low-frequency envelope feature data include: performing quadrature demodulation on the dynamic contact force timing signal using the dynamic excitation frequency, extracting the effective energy value and phase angle within the target frequency band with the dynamic excitation frequency as the center frequency, and generating the high-frequency carrier feature data; Perform a Hilbert transform on the dynamic contact force time-series signal to obtain the basic envelope signal; calculate the variance of the basic envelope signal to generate the low-frequency envelope feature data.

[0005] Preferably, high-frequency carrier feature data and low-frequency envelope feature data are input into a preset classification model, and the internal rheological state evaluation result of the target object is output, including: extracting the effective energy value from the high-frequency carrier feature data and the variance from the low-frequency envelope feature data; comparing the effective energy value with a preset energy reference threshold and comparing the variance with a preset variance fluctuation threshold; when the effective energy value is greater than or equal to the energy reference threshold and the variance is less than or equal to the variance fluctuation threshold, the target object is determined to be in a first rheological state; when the effective energy value is less than the energy reference threshold or the variance is greater than the variance fluctuation threshold, the target object is determined to be in a second rheological state; and the first or second rheological state is output as the internal rheological state evaluation result.

[0006] Preferably, the method further includes a step of decoupling the baseline offset caused by individual mass differences: before acquiring the dynamic contact force timing signal, static weight data of the target object is collected; based on the static weight data, the corresponding mass compensation coefficient is matched from a preset baseline compensation mapping table; the amplitude of the dynamic contact force timing signal is normalized using the mass compensation coefficient to generate a standardized contact force timing signal that eliminates mass difference interference; the standardized contact force timing signal replaces the original dynamic contact force timing signal and is input into subsequent processing steps.

[0007] Preferably, the method further includes dynamically calibrating the decision boundary parameters of the preset classification model based on the internal rheological state assessment results. Specifically, the dynamic calibration includes: collecting the internal rheological state assessment results and corresponding high-frequency carrier feature data and low-frequency envelope feature data within historical detection periods to construct a historical feature distribution set; calculating the centroid coordinates of the historical feature distribution set in the feature space, and using the distance between the current centroid coordinates and the initial centroid coordinates as the aggregation center drift; when the aggregation center drift exceeds a preset feature space Euclidean distance drift tolerance, using the aggregation center drift to compensate and update the energy baseline threshold and variance fluctuation threshold in the preset classification model in the opposite direction of their offset vector; when the aggregation center drift does not exceed the preset drift tolerance, keeping the current parameters in the preset classification model unchanged.

[0008] Preferably, the dynamic contact force timing signal is quadrature demodulated using the dynamic excitation frequency to extract the effective energy value and phase angle within the target frequency band centered on the dynamic excitation frequency, generating high-frequency carrier characteristic data. This includes: generating a local in-phase reference signal and a local quadrature reference signal based on the dynamic excitation frequency; mixing the dynamic contact force timing signal with the local in-phase reference signal and the local quadrature reference signal respectively to output in-phase and quadrature components; performing low-pass filtering on the in-phase and quadrature components to filter out high-frequency aliasing noise; and synthesizing the effective energy value and phase angle corresponding to the dynamic excitation frequency based on the filtered in-phase and quadrature components.

[0009] Preferably, in the process of frequency domain isolation and demodulation processing of dynamic contact force timing signal using dynamic excitation frequency as reference signal, mechanical noise isolation constraint is introduced; wherein, mechanical noise isolation constraint includes: when extracting features, only the signal energy within a preset bandwidth range centered on dynamic excitation frequency is retained, and broadband mechanical noise energy exceeding the preset bandwidth range is removed, so as to improve the signal-to-noise ratio of extracted features.

[0010] Preferably, the target detection object is an egg; the freshness level is correlated with the degree of protein denaturation and moisture evaporation inside the egg through a preset mapping relationship; the excitation track is a corrugated guide rail on the sorting assembly line.

[0011] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention is based on a corrugated guide rail, a piezoelectric sensor array, and a photoelectric speed measurement module on a sorting assembly line. During the continuous rolling of eggs, dynamic contact force timing signals and real-time rolling linear velocity data are collected simultaneously. The dynamic excitation frequency is calculated from the real-time rolling linear velocity data and the period length of the corrugated space. This can transform the rolling process in the assembly line into a controlled detection process with a known reference frequency. Thus, online detection can be completed without stopping the eggs, tapping, or manual sampling, improving the real-time performance and adaptability of egg freshness detection under continuous conveying conditions. 2. This invention uses the dynamic excitation frequency as a reference signal to perform frequency domain isolation and demodulation processing on the dynamic contact force time-series signal. By extracting the effective energy value and phase angle within the target frequency band through orthogonal demodulation, and by obtaining the basic envelope signal through Hilbert transform, the envelope variance is calculated. This allows for the acquisition of high-frequency carrier characteristic data characterizing the coupling response between the outer shell and the internal medium of the target object, as well as low-frequency envelope characteristic data characterizing the internal centroid shift. This enables the detection results to reflect not only the excitation response of the target object's outer shell, but also the degree of protein denaturation, the state of water evaporation, and internal rheological changes, thereby improving the ability to determine the freshness of the egg's interior. 3. This invention inputs high-frequency carrier feature data and low-frequency envelope feature data into a preset classification model, or uses a dual threshold judgment method of energy effective value and variance to output the internal rheological state evaluation result, and then maps the internal rheological state to freshness level. This can quickly achieve egg freshness grading with low computational load, which takes into account the system's computational efficiency in high-speed production line scenarios, and makes the judgment logic clear and stable. This solves the problem that traditional detection methods cannot simultaneously take into account detection speed and judgment consistency. Attached Figure Description

[0012] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart illustrating the steps of the egg freshness detection method based on an intelligent sensing system provided in an embodiment of this application. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0014] Please see Figure 1 An egg freshness detection method based on an intelligent sensing system is used to conduct online quality assessment of eggs as the target detection object. The method includes: real-time acquisition of the vertical force changes of the target detection object during rolling by a piezoelectric sensing array laid under an excitation track with a preset spatial period sine wave or toothed corrugation, and generating a dynamic contact force timing signal. A photoelectric velocity measurement module positioned on the side of the excitation track captures the movement time of the target object as it passes through a preset detection interval on the excitation track, and calculates the real-time rolling linear velocity data. Based on the real-time rolling linear velocity data and the spatial period length of the preset spatial periodic ripple, the dynamic excitation frequency experienced by the target object is calculated. Using the dynamic excitation frequency as a reference signal, the dynamic contact force timing signal is subjected to frequency domain isolation and demodulation processing. High-frequency carrier characteristic data characterizing the coupling response between the target object's outer shell and internal medium, and low-frequency envelope characteristic data characterizing the shift of the target object's internal center of gravity are extracted. The high-frequency carrier characteristic data includes the effective energy value and phase angle, while the low-frequency envelope characteristic data includes variance. High-frequency carrier feature data and low-frequency envelope feature data are input into a preset classification model, and the internal rheological state assessment results of the target object are output; based on the internal rheological state assessment results, the freshness level of the target object is determined.

[0015] Furthermore, the specific steps for extracting high-frequency carrier feature data and low-frequency envelope feature data include: performing quadrature demodulation on the dynamic contact force time-series signal using the dynamic excitation frequency, extracting the effective energy value and phase angle within the target frequency band centered on the dynamic excitation frequency, and generating high-frequency carrier feature data; performing Hilbert transform on the dynamic contact force time-series signal to obtain the basic envelope signal; and calculating the variance of the basic envelope signal to generate low-frequency envelope feature data.

[0016] This embodiment provides an egg freshness detection mechanism based on an intelligent sensing system. Specifically, the mechanism is deployed on the sorting line of an egg processing plant. After the eggs enter the quality detection section from the feeding section, there is no need to stop the egg conveying or mechanically tap them. Instead, the force collection, speed measurement, feature extraction, state determination and model boundary update are completed during the continuous rolling process, thereby forming a whole-link detection process suitable for online grading. The details are as follows: In the detection section, an excitation track with a preset spatial period sine wave or toothed wave is set. The surface of the excitation track forms equally spaced microwave ripples along the rolling direction, and the spatial period length between two adjacent wave peaks can be preset to 5 mm. A piezoelectric sensor array is laid under the track to collect the dynamic contact force in the vertical direction when the egg rolls through. Adjacent to it, a set of photoelectric speed measurement modules is set. The speed measurement module includes two photoelectric gates, one in front and one behind. The detection interval between the two photoelectric gates is, for example, 30 mm. The photoelectric speed measurement module and the piezoelectric sensor array are spatially aligned to ensure that the captured real-time rolling linear velocity data corresponds to the same physical sampling period for generating the dynamic contact force timing signal. If the time difference recorded by the photoelectric gates before and after an egg passes through is 0.06 seconds, then its real-time rolling linear velocity can be obtained as follows: Meters per second; furthermore, the dynamic excitation frequency is calculated based on the velocity and the spatial period length of the guide rail ripples, for example... To ensure the accuracy of the excitation frequency extraction, the calculation process needs to be updated in real time within a preset sampling sliding window to eliminate frequency offset interference caused by instantaneous speed fluctuations in the pipeline. At this time, the 100 Hz is not a frequency estimated after the fact, but a reference frequency determined by the pipeline spatial structure and the measured speed. During the acquisition phase, the piezoelectric sensing array outputs a dynamic contact force timing signal. For ease of explanation, it is assumed that this timing signal consists of 8 discrete sampling points within a short sampling window, with original amplitudes of 9, 12, 10, 13, 8, 11, 9, and 12, respectively. This signal contains both periodic high-frequency components caused by the contact between the egg and the corrugated rail, and low-frequency fluctuations caused by the shift of the egg's center of gravity. Since the dynamic excitation frequency of 100 Hz is obtained in advance in this embodiment, the subsequent frequency traversal scan is not performed indiscriminately across the entire frequency domain, but rather target frequency band isolation is performed around 100 Hz. Specifically, the dynamic excitation frequency is first used as a local reference to perform quadrature demodulation on the dynamic contact force timing signal. During this process, the system generates a set of in-phase reference sequences and a set of quadrature reference sequences, which are then multiplied point-by-point with the original signal and low-pass processed. For ease of illustration, assuming that the average value of the in-phase component is 3 and the average value of the quadrature component is 4 after mixing and low-passing, the corresponding effective value of the high-frequency carrier energy can be obtained as 5 by combining the two. The phase angle can be characterized as having a certain lag with the in-phase direction. In the specific synthesis, the phase angle... The calculation formula is defined as follows: in, These are the filtered orthogonal components. This is the filtered in-phase component. Indicates basis , The two symbols together determine the two-parameter arctangent operation in the quadrant, thereby ensuring the continuity and monotonicity of phase changes in continuous dynamic detection through the four-quadrant arctangent function; for fresh eggs, due to the relatively viscous egg white, the coupling response between the eggshell and the internal medium is closer to the overall rigidity, which usually shows a higher effective energy value and a smaller phase lag; for aged eggs, the internal fluidity is enhanced, which usually shows a lower effective energy value and a larger phase lag. In the low-frequency feature extraction section, a Hilbert transform is performed on the same segment of dynamic contact force time-series signal to obtain the basic envelope signal. For ease of explanation, it is assumed that the envelope sequence is 10, 10, 11, 9, 12, 8, 11, 9, then its average value is approximately 10, with relatively small fluctuations around the average value and a variance of approximately 2. If the envelope sequence obtained from another egg is 8, 13, 7, 14, 6, 15, 8, 13, then its variance is significantly larger, for example, approximately 10. The former indicates that the internal center of gravity shift is relatively stable during the rolling process, while the latter indicates that the internal yolk is more significantly free and the center of gravity trajectory fluctuates more. Therefore, the system uses the effective energy value and phase angle as high-frequency carrier feature data and the envelope variance as low-frequency envelope feature data. It should be further explained that, in this embodiment, the meaning of the envelope variance used to characterize the internal center of gravity shift is as follows: the dynamic contact force signal can usually be regarded as a controlled excitation response with the dynamic excitation frequency as the main carrier, and the slow migration, shaking and eccentricity of the medium inside the egg relative to the shell will be superimposed on the carrier response in the form of amplitude modulation, so that the Hilbert envelope will show statistical fluctuations in the lower frequency band. To ensure the basic envelope signal more stably corresponds to the aforementioned amplitude modulation effect and to avoid the misleading effect of broadband mechanical disturbances on the envelope, in engineering implementation, it is permissible to perform narrowband pre-filtering or target frequency band isolation on the dynamic contact force timing signal centered on the dynamic excitation frequency before performing the Hilbert transform. The center frequency is derived from the dynamic excitation frequency, and the bandwidth can be preset according to the production line speed fluctuation level. Then, the Hilbert transform is performed on the filtered signal to obtain the basic envelope. Furthermore, a low-pass smoothing or statistical sampling by window segmentation can be performed on the obtained envelope to ensure that the final variance mainly reflects low-frequency fluctuations rather than carrier residual ripple, thereby improving the repeatability of features under short sampling windows and mechanical noise backgrounds. The above supplement is a refinement of the implementation of the processing link of obtaining the envelope and calculating the variance by Hilbert transform, without changing the processing link, and is only used to improve the repeatability of the envelope variance under short sampling windows and mechanical noise backgrounds. After the above high-frequency carrier feature data and low-frequency envelope feature data are processed to unify the dimensions, they are constructed into feature vectors and input into a preset classification model based on dual feature threshold joint judgment. The preset classification model can be a lightweight binary or multi-class judgment model running on the edge device. Its input includes at least the effective value of high-frequency energy and the variance of low-frequency envelope, and the phase angle can also be further introduced as an auxiliary dimension. The model outputs the internal rheological state evaluation results. For example, the first rheological state indicates that the egg white has high viscosity and the internal medium is tightly coupled, while the second rheological state indicates that the egg white is significantly thinned and the internal shaking is enhanced. Then, the rheological state is mapped to the freshness level. For example, the first rheological state corresponds to Grade 1 or Grade 2 fresh eggs, and the second rheological state corresponds to the critical freshness state or the stale state. In the dynamic calibration part of the model, the system does not fix the judgment boundary in the long term, but corrects the judgment boundary parameters in reverse according to the historical evaluation results within the continuous production cycle. For example, when the ambient temperature is lower in the morning shift and higher in the afternoon shift, the sensor baseline and the friction state of the egg surface may drift slowly. At this time, the center of the historical feature distribution will be slightly shifted. The system can calculate the aggregation center of the historical features according to a preset cycle, such as every 1,000 eggs or every 30 minutes, and adjust the threshold or boundary position of the classification model accordingly to ensure that the model remains consistent under seasonal and batch changes. As an anomaly tolerance mechanism, if the photoelectric speed measurement module fails to obtain an effective time difference due to occlusion, the dynamic excitation frequency of the current egg cannot be accurately calculated. At this time, the system does not directly enter the frequency domain demodulation, but marks the egg as a sample with missing speed information, re-triggers the detection logic, or guides it to the re-inspection channel. If the output amplitude of the piezoelectric sensing array is lower than the minimum response threshold, for example, if the egg is not in stable contact with the guide rail or the abnormal posture leads to insufficient contact, it can be directly judged as an abnormal contact sample and the feature is not used to update the model. If the envelope variance is extremely large and the high-frequency energy is extremely low at the same time, exceeding the reasonable physiological range, it can be identified as an abnormal interference sample such as eggshell damage, foreign matter attached to the surface, or double yolk, and will enter the manual review and diversion to avoid abnormal samples from contaminating subsequent model calibration. In addition, if non-periodic sliding friction or jumping impact interference is detected in the time sequence signal, the system will reduce the judgment weight of the feature of that segment, or directly switch to the fault tolerance compensation algorithm to avoid misjudgment caused by the failure of the ideal rolling model. For example, on the same sorting line, eggs A and B enter the detection section sequentially. Egg A passes through the photogate with a time difference of 0.06 seconds, corresponding to a speed of 0.5 meters per second, forming a 100 Hz dynamic excitation on a 5 mm corrugated guide rail. Its piezoelectric timing signal, after orthogonal demodulation, yields an effective energy value of 5, a small phase angle, and an envelope variance of 2. The preset classification model evaluates it as the first rheological state and gives a result of high freshness grade. Egg B also passes through with a time difference of 0.06 seconds, but its demodulated effective energy value is only 2.5, its phase angle is significantly increased, and its envelope variance is 10. The preset classification model evaluates it as the second rheological state and gives a result of decreased freshness grade. The system then incorporates A, B, and historical samples from this period into boundary calibration to update the judgment parameters for subsequent batches. The purpose of this step is to transform the originally disordered rolling excitation in the sorting line into detection excitation with known frequency labels, and to achieve stable determination of the internal rheological state and freshness level of eggs without interrupting the production line through joint analysis of dual features of high-frequency carrier and low-frequency envelope. At the same time, it provides a unified data entry and scenario basis for subsequent extended steps such as threshold determination, quality compensation, and boundary dynamic correction in subordinate embodiments. In this embodiment, the corrugated guide rail is an excitation track with a preset spatial period sine wave or toothed corrugation; unless otherwise specified, the dynamic contact force timing signal refers to the acquired original waveform; the signal after quality compensation is called the standardized contact force timing signal; the effective energy value refers to the characteristic quantity synthesized in the target frequency band corresponding to the dynamic excitation frequency after orthogonal demodulation; the envelope variance is the specific statistical expression of the low-frequency envelope characteristic data; The preset classification model can be either a lightweight model or a threshold-based judgment model; the dual-threshold joint judgment method is a specific implementation of the preset classification model.

[0017] Furthermore, the high-frequency carrier feature data and low-frequency envelope feature data are input into a pre-defined classification model, and the internal rheological state evaluation results of the target object are output, including: Extract the effective energy value from the high-frequency carrier feature data and the variance from the low-frequency envelope feature data; compare the effective energy value with a preset energy reference threshold and compare the variance with a preset variance fluctuation threshold; when the effective energy value is greater than or equal to the energy reference threshold and the variance is less than or equal to the variance fluctuation threshold, the target object is determined to be in the first rheological state; when the effective energy value is less than the energy reference threshold or the variance is greater than the variance fluctuation threshold, the target object is determined to be in the second rheological state; output the first or second rheological state as the internal rheological state evaluation result. This embodiment provides a threshold-based rheological state determination mechanism for real-time sorting in pipelines. Specifically, after extracting high-frequency carrier features and low-frequency envelope features, in order to avoid increased latency caused by running complex models on edge devices, this embodiment adopts a dual-threshold judgment method of effective energy value and envelope variance to quickly output the internal rheological state evaluation result. This dual-threshold judgment method is a specific implementation of the preset classification model. In detail, although the features of the preset classification model have been given, if high-dimensional features and complex classifiers are directly used on a high-speed production line, computational resources may be overloaded and processing delays may occur when multiple eggs pass through concurrently per second. Therefore, in this embodiment, the effective value of energy in the high-frequency carrier features is used as the first criterion, and the variance in the low-frequency envelope features is used as the second criterion, and then compared with preset thresholds respectively. Before the system is run for the first time or before the test batch is changed, initial calibration is required: by putting in standard test eggs with known freshness labels, such as fresh egg groups that are determined to be in the first rheological state, the system collects the high-frequency carrier characteristics and low-frequency envelope characteristics during the rolling process, calculates the characteristic mean and standard deviation of the standard test egg group, and generates preset energy benchmark threshold and variance fluctuation threshold accordingly. Assuming the energy baseline threshold is set to 4.0 and the variance fluctuation threshold is set to 3.5, the following rules are applied to any egg: when the effective energy value is greater than or equal to 4.0 and the variance is less than or equal to 3.5, it is determined to be in the first rheological state; when the effective energy value is less than 4.0 or the variance is greater than 3.5, that is, when any freshness loss characteristic is met, it is determined to be in the second rheological state. To facilitate understanding, specific numerical examples are provided below. If egg C has an effective energy value of 4.8 and an envelope variance of 2.1, it meets both criteria and is classified as the first rheological state. If egg D has an effective energy value of 3.2 and an envelope variance of 2.0, although the low-frequency envelope variance is at a low fluctuation level, its high-frequency carrier attenuation is too strong, indicating increased internal damping, so it is still classified as the second rheological state. If egg E has an effective energy value of 4.5 and an envelope variance of 5.8, although the high-frequency carrier characteristics meet the benchmark, the low-frequency center of gravity sway is obvious, so it should still be classified as the second rheological state. Thus, it can be seen that the logical disjunction condition in this embodiment is used to cover two types of quality degradation modes, namely, weakened overall coupling and increased center of gravity detachment. As an anomaly tolerance mechanism, if the effective energy value is exactly equal to the energy baseline threshold, or the variance is exactly equal to the variance fluctuation threshold, this embodiment prioritizes including such boundary samples in the first rheological state, thereby reducing the probability of misjudging boundary fresh eggs; however, if the same sample is repeatedly detected twice and the results are inconsistent, for example, the first time it is 4.0 and 3.5, and is judged as the first rheological state, and the second time it is 3.9 and 3.6, and is judged as the second rheological state, then the system marks it as a critical sample and enters the re-inspection buffer section; furthermore, if the sampling window is too short and the variance calculation is unstable, for example, the effective sample points are less than the preset number, then the rheological state is not output temporarily, but instead the sampling path of the egg in the detection section is extended or switched to the backup detection channel; For example, on the aforementioned egg processing plant production line, egg A, after processing, yields an effective energy value of 5 and a variance of 2. After being input into the threshold determination module of this embodiment, it is identified as the first rheological state, corresponding to the high freshness channel; egg B yields an effective energy value of 2.5 and a variance of 10, and its high-frequency carrier and low-frequency envelope feature parameters both deviate from the preset determination threshold range, therefore it is identified as the second rheological state, corresponding to the re-inspection or secondary classification channel; egg F has an effective energy value of 4.1 and a variance of 3.4, so it is still judged as the first rheological state; egg G has an effective energy value of 4.2 and a variance of 3.8, so because the variance exceeds the threshold, it is judged as the second rheological state. The purpose of this step is to quickly transform the extracted dual features into internal rheological state conclusions with low computational cost and clear threshold rules, thereby meeting the dual requirements of computational latency and judgment stability for real-time sorting in the pipeline.

[0018] Furthermore, this method also includes a step of decoupling the baseline offset caused by individual mass differences: before acquiring the dynamic contact force timing signal, static weight data of the target object is collected; based on the static weight data, the corresponding mass compensation coefficient is matched from a preset baseline compensation mapping table; the amplitude of the dynamic contact force timing signal is normalized using the mass compensation coefficient to generate a standardized contact force timing signal that eliminates mass difference interference; the standardized contact force timing signal replaces the original dynamic contact force timing signal and is input into subsequent processing steps. This embodiment provides a baseline compensation mechanism for individual egg quality differences. Specifically, in a continuous rolling detection scheme, if the quality of different eggs differs significantly, the dynamic contact force baseline collected by the piezoelectric sensing array will rise or fall with the static weight. If decoupling is not performed, a large-weight aging sample may be misjudged as a fresh sample with a high energy response due to the rise in the contact force baseline, and a small-weight fresh sample may also be misjudged as having freshness characteristics due to a low initial amplitude reference. Therefore, this embodiment performs quality compensation before entering dynamic demodulation. Specifically, before the eggs enter the corrugated guide rail, their weight data can be obtained through the upstream static weighing unit; for example, egg H weighs 50 grams and egg I weighs 65 grams; the system pre-establishes a baseline compensation mapping table, which can be obtained during the calibration stage; for example, when the weight range is 45 to 55 grams, the corresponding mass compensation coefficient is set to 1.10; when the weight range is 56 to 65 grams, the corresponding mass compensation coefficient is set to 0.92; when the weight range is 66 to 75 grams, the corresponding mass compensation coefficient is set to 0.85; the function of this compensation coefficient is to normalize and map the difference in contact force amplitude caused by different weights to a unified reference scale; Continuing the deduction, if the dynamic contact force time-series signal collected after egg H enters the detection section is 8, 10, 9, 11, 8, 10, then based on its 50-gram weight, a compensation coefficient of 1.10 is matched, and after normalization, we get 8.8, 11.0, 9.9, 12.1, 8.8, 11.0; if the original time-series signal of egg I is 10, 12, 11, 13, 10, 12, based on its 65-gram weight, a compensation coefficient of 0.92 is matched, and after normalization, we get 9.2, 11.04, 10.12, 11.96, 9.2, 11.04; through the above normalization process, the static amplitude offset caused by the difference in weight between the two is compressed into the same comparable range, and the subsequent orthogonal demodulation and envelope analysis are no longer dominated by individual weight; As an anomaly tolerance mechanism, if the static weighing unit output is missing or the weighing result exceeds the coverage range of the preset mapping table, such as less than 40 grams or greater than 90 grams, the system can use the compensation coefficient of the nearest neighbor interval for temporary replacement and simultaneously mark the sample as a weight out-of-bounds sample; if the weighing data fluctuates abnormally in a short period of time, such as the same egg experiencing multiple jumps when passing through the weighing platform, the system can take the average of the stable segment as the final weight after removing impulse noise; if the mapping table query fails or the compensation coefficient is zero, the normalization process is prohibited to avoid subsequent feature distortion, and the egg is transferred to the backup detection line or manual re-inspection line; if weight abnormality and contact force abnormality occur simultaneously, they are treated as abnormal samples first and are not used for model dynamic calibration; For example, on the same production line in the aforementioned egg processing plant, although eggs A and B both pass through the same corrugated guide rail, egg A weighs 52 grams and egg B weighs 68 grams. Without the mass compensation in this embodiment, the overall amplitude of the dynamic contact force of egg B may naturally be higher than that of egg A, thereby masking the high-frequency attenuation caused by its internal rheological deterioration. After introducing compensation, egg A is amplified by a factor of 1.10 and egg B is scaled by a factor of 0.85. The two are remapped to a unified reference and enter orthogonal demodulation and Hilbert envelope calculation, ultimately obtaining more comparable energy effective values ​​and variances. The purpose of this step is to decouple and eliminate the baseline offset caused by individual quality differences from the dynamic contact force, so that subsequent features can more accurately reflect the real rheological changes caused by internal protein denaturation, water evaporation and center of gravity displacement, thereby improving the consistency of judgment when detecting mixed lines of eggs of different specifications.

[0019] Furthermore, the method also includes dynamically calibrating the decision boundary parameters of the preset classification model based on the internal rheological state assessment results. The dynamic calibration specifically includes: collecting the internal rheological state assessment results and the corresponding high-frequency carrier feature data and low-frequency envelope feature data within the historical detection period to construct a historical feature distribution set; calculating the centroid coordinates of the historical feature distribution set in the feature space, and using the distance between the current centroid coordinates and the initial centroid coordinates as the aggregation center drift. When the cluster center drift exceeds the preset feature space Euclidean distance drift tolerance, the energy baseline threshold and variance fluctuation threshold in the preset classification model are updated in the opposite direction of their offset vector using the cluster center drift; when the cluster center drift does not exceed the preset drift tolerance, the current parameters in the preset classification model remain unchanged.

[0020] This embodiment provides a dynamic calibration mechanism for the determination boundary of a production line for long-term operation. Specifically, although the rheological state determination can be completed quickly using a dual threshold method, the surface friction of eggs, ambient temperature and humidity, sensor zero point and guide rail wear may change slowly after shifts, seasons or changes in the supply farm. If a fixed threshold is always used, long-term operation will lead to the cumulative deviation of boundary samples. Therefore, this embodiment introduces the drift detection and reverse compensation update of historical feature distribution. Specifically, the system collects historical data at a preset detection cycle; for example, in the most recent 1000 eggs, it records the effective energy value, envelope variance and the corresponding first or second rheological state result of each egg, forming a historical feature distribution set; It should be noted that the dimension of the feature space corresponds to the number of features used in the preset classification model. For ease of explanation, it is assumed that the centroid of the first rheological state feature in the initial calibration stage is located at coordinates (5.0, 2.0), where the horizontal axis represents the mean effective energy value and the vertical axis represents the mean envelope variance. After a period of operation, the centroid of the current sample of the same type becomes coordinates (4.6, 2.4). The distance between the two is approximately 0.57. If the preset drift tolerance is 0.5, it indicates that the current system state has undergone a perceptible drift. The drift amount is determined by using the cluster center of the first rheological state sample in the two-dimensional feature space as a reference. By monitoring the evolution of the Euclidean distance of this reference point relative to the initial calibration point, the automatic identification of sensor aging or environmental temperature drift is achieved. In this case, instead of directly discarding and reconstructing the existing model, this embodiment uses the drift amount to perform reverse compensation updates on the threshold. Reverse compensation means that if the overall energy distribution of the first rheological state shifts downward while the variance distribution shifts upward, the threshold should be appropriately adjusted accordingly to maintain the consistency of the actual physical meaning. Continuing with the above example, the original energy baseline threshold is 4.0 and the variance fluctuation threshold is 3.5. When the centroid drifts to the upper left and exceeds the tolerance, the energy baseline threshold can be lowered to 3.8, while the variance fluctuation threshold can be raised to 3.7. In this way, the critical fresh sample that was originally misjudged as the second rheological state due to the device state drift can fall back into the effective classification boundary. To further illustrate this, the following verification is performed using specific data. Assume there are four samples in the first rheological state within the current period, with coordinates (4.7, 2.5); (4.5, 2.2); (4.6, 2.6); and (4.6, 2.3), respectively. Their centroids are approximately at coordinates (4.6, 2.4). If the initial centroid is at coordinates (5.0, 2.0), it indicates that the system as a whole exhibits lower high-frequency energy and a slightly higher envelope variance. If the original thresholds of 4.0 and 3.5 are maintained, some actual fresh eggs may be misclassified to the second rheological state. Using the threshold update in this embodiment, the boundary can be corrected to a more reasonable position. As an anomaly tolerance mechanism, if the number of effective samples in the historical feature distribution set is insufficient, for example, only 20 samples are collected in a certain period, which is lower than the preset minimum statistical number, then the centroid update will not be performed, and the current parameters will remain unchanged; if the distance between the current centroid and the initial centroid exceeds the threshold, but the drift direction is inconsistent with the device status log, for example, energy and variance show abnormally large changes at the same time, which may indicate sensor failure or guide rail loosening, then the threshold will not be updated immediately, but the device self-check will be triggered; if the historical centroids of the first rheological state and the second rheological state overlap, it indicates that the classification boundary has lost its separation, the dynamic calibration function should be frozen, and the recalibration process should be initiated to avoid further amplification of the deviation by error compensation; For example, on the same production line in the aforementioned egg processing plant, the initial calibration completed in the morning showed the centroid of the fresh egg population to be at coordinates (5.0, 2.0); after entering the afternoon high-temperature period, the sensor noise floor increased slightly and the friction state of the egg surface changed, and the centroid of the fresh egg population statistically recorded by the system changed to coordinates (4.6, 2.4); since the drift amount exceeded 0.5, the system automatically adjusted the energy reference threshold from 4.0 to 3.8 and the variance fluctuation threshold from 3.5 to 3.7; thereafter, the boundary samples of the same batch were no longer concentrated and misclassified into the second rheological state, and the sorting line continued to operate stably; The purpose of this mechanism is to enable the threshold determination to be adaptive, so as to absorb the slow drift of equipment and environment without re-stopping the production line for large-scale calibration, and maintain the consistency and repeatability of freshness test results in long-term continuous production. Furthermore, the historical feature distribution set should maintain a consistent statistical caliber in engineering implementation; preferably, the centroid calculation should be performed using the same type of reference sample subset, for example, continuously using samples that have been determined to be in the first rheological state and have been screened out as abnormal samples as the drift monitoring benchmark, because the physical state of such samples is relatively concentrated and more suitable as a reference for the slow drift of equipment and environment. If a historical feature distribution set is constructed using a full set of samples including the first and second rheological states, the initial centroid and the current centroid should also be calculated under the same full-scale caliber. The centroid of the first rheological state subset and the centroid of the full set of samples should not be used alternately in different periods to avoid the change in the proportion of sample categories being misidentified as equipment drift. In the calculation of the aggregation center drift, the centroid of the first rheological state characteristic in the initial calibration stage and the centroid of the current similar sample are compared under the same statistical caliber; the centroid calibrated in this embodiment is the slow shift of the judgment boundary caused by equipment and environmental factors.

[0021] Furthermore, the dynamic contact force timing signal is quadrature demodulated using the dynamic excitation frequency to extract the effective energy value and phase angle within the target frequency band centered on the dynamic excitation frequency, generating high-frequency carrier characteristic data. This includes: generating a local in-phase reference signal and a local quadrature reference signal based on the dynamic excitation frequency; mixing the dynamic contact force timing signal with the local in-phase reference signal and the local quadrature reference signal respectively to output in-phase and quadrature components; performing low-pass filtering on the in-phase and quadrature components to filter out high-frequency aliasing noise; and synthesizing the effective energy value and phase angle corresponding to the dynamic excitation frequency based on the filtered in-phase and quadrature components. This embodiment provides an orthogonal demodulation extraction mechanism for high-frequency carrier characteristics; specifically, as previously indicated, frequency domain isolation and demodulation of dynamic contact force timing signals are performed using dynamic excitation frequencies; this embodiment further provides a specific implementation of the demodulation process to stably extract the effective energy value and phase angle within a short sampling window; In summary, although this method can generally explain the extraction of high-frequency features around the dynamic excitation frequency, if the ordinary spectral peak method is used directly, the peak value is easily affected by sidelobes, sampling length, and mechanical vibration interference under the short pipeline window. Therefore, this embodiment introduces orthogonal demodulation. Taking the dynamic excitation frequency of an egg as 100 Hz as an example, the system synchronously generates a set of local in-phase reference signals and a set of local quadrature reference signals. If the sampling rate is 1000 Hz, 10 sampling points can be formed within a 10-millisecond segment. For ease of deduction, it is assumed that the in-phase reference sequence is simplified to 1, 0, -1, 0, 1, 0, -1, 0, 1, 0, and the quadrature reference sequence is simplified to 0, 1, 0, -1, 0, 1, 0, -1, 0, 1. It should be noted that the above sequence of 1, 0, -1, 0... is only used for illustrative deduction in the specification to intuitively demonstrate the point-by-point multiplication relationship of in-phase / quadrature two-way mixing; in actual engineering implementation, the local in-phase reference signal and the local quadrature reference signal can adopt a discrete cosine / sine sequence that is locked in phase with the dynamic excitation frequency, or a high-precision digital quadrature reference generated by a numerically controlled oscillator (NCO) to ensure that phase continuity is maintained even with slight speed fluctuations. Assuming the dynamic contact force timing signal after quality compensation is 10, 12, 9, 11, 10, 12, 8, 10, 9, 11, the system multiplies this signal point-by-point with the in-phase reference sequence and the quadrature reference sequence, respectively. After mixing with the in-phase reference sequence, we get 10, 0, -9, 0, 10, 0, -8, 0, 9, 0; after mixing with the quadrature reference sequence, we get 0, 12, 0, -11, 0, 12, 0, -10, 0, 11. Since the mixed signals still contain high-frequency and aliasing components, low-pass filtering is performed on the above two sets of results. To simplify the explanation, the stable output after low-pass can be regarded as in-phase component 3 and quadrature component 4; thus, the effective value of the high-frequency carrier energy can be obtained by combining the in-phase component and the quadrature component 5, and the phase angle is obtained by the relative ratio of the two, which represents the phase delay of the measured response relative to the reference excitation. Furthermore, to avoid phase angle jumps during quadrant switching, in engineering practice, it can be calculated based on in-phase components. Orthogonal components A unified set of rules for synthesis and phase calculation is adopted; for example, the effective energy value can be obtained from... and The phase angle can be obtained by taking the square root of the square. The function is used to obtain the phase angle, thus ensuring a consistent interpretation caliber when crossing the boundaries of positive and negative circumferential angles. In engineering, the in-phase component better reflects the rigid response synchronized with the reference excitation, while the quadrature component more easily reflects internal damping and hysteresis characteristics. For fresher eggs, the filtered result usually shows a higher proportion of in-phase components, a smaller quadrature component, and a smaller phase angle. For older eggs, the internal liquid components exhibit a more significant phase hysteresis relative to the eggshell, the quadrature component increases, the phase angle increases, and the effective energy value of the synthesized result decreases. As an anomaly tolerance mechanism, if the local reference signal is out of sync with the dynamic excitation frequency calculated from the measured speed, for example, if the speed sampling delay causes the reference frequency deviation to exceed the preset tolerance, the demodulation result may be distorted. In this case, the system can prioritize using the most recent stable speed value for compensation. If synchronization still cannot be restored, the sample is marked as a demodulation out-of-sync sample. If both the in-phase and quadrature components are close to zero after low-pass filtering, it indicates that there is no effective excitation response in the current frequency band. This may be caused by poor contact, false triggering under no-load conditions, or foreign objects on the guide rail. Feature synthesis should be stopped and a cleaning or re-inspection process should be initiated. If the in-phase component is zero while the quadrature component is non-zero, or vice versa, the phase angle calculation will fall on a special boundary. In this case, it can be corrected according to the preset quadrant rule to avoid phase jumps. For example, in the aforementioned sorting line, the dynamic excitation frequency of egg A is 100 Hz. After quadrature demodulation, it yields an in-phase component of 4.2 and a quadrature component of 1.6, resulting in a high effective value of synthesized energy and a small phase angle. The in-phase component of egg B decreases to 1.8, while the quadrature component increases to 2.4, resulting in a significant decrease in the effective value of synthesized energy and an increase in the phase angle. Even though both eggs are excited by the same corrugated guide rail during rolling, this embodiment can still separate the high-frequency response changes caused by internal rheological differences through synchronous reference demodulation. The purpose of this step is to implement the target frequency band extraction into an executable synchronous demodulation process, so that the system can still stably obtain high-frequency carrier characteristics that characterize the coupling degree between the eggshell and the internal medium in a short time window and high noise environment. In this embodiment, the low-pass filtering and mechanical noise isolation constraints work together in terms of processing location and function. The mechanical noise isolation constraints emphasize pre-defining the target frequency band around the dynamic excitation frequency to eliminate broadband mechanical noise. The low-pass filtering, after mixing, performs baseband smoothing on the high-frequency and out-of-beat components in the result. Under production line conditions with strong mechanical disturbances, a bandwidth constraint centered on the dynamic excitation frequency can be applied to the dynamic contact force timing signal before entering in-phase and quadrature mixing and low-pass processing. Alternatively, the bandwidth constraint can be embedded into the demodulation link to achieve synchronous narrowband retention, so that the finally extracted feature parameters maintain a unified physical meaning.

[0022] Furthermore, in the process of frequency domain isolation and demodulation of the dynamic contact force timing signal using the dynamic excitation frequency as a reference signal, a mechanical noise isolation constraint is introduced. The mechanical noise isolation constraint includes: when extracting features, only the signal energy within a preset bandwidth centered on the dynamic excitation frequency is retained, and broadband mechanical noise energy exceeding the preset bandwidth is removed, so as to improve the signal-to-noise ratio of the extracted features.

[0023] This embodiment provides a mechanical noise isolation constraint mechanism. Specifically, a traceable reference frequency is established through the dynamic excitation frequency, further illustrating the orthogonal demodulation process. However, in a real production line environment, motor vibration, chain impact, resonance of adjacent lanes, and loose guide rail connectors can all introduce energy over a wide frequency range. The preset bandwidth range is typically set to 10% to 20% of the dynamic excitation frequency to ensure that while filtering out mechanical background noise, the frequency modulation component caused by egg white rheology is fully preserved. If bandwidth constraints are not applied, the demodulation result may still be mixed with non-target mechanical energy. Therefore, this embodiment limits the retention of energy within the preset bandwidth surrounding the dynamic excitation frequency during the frequency domain isolation process. Specifically, assuming the dynamic excitation frequency of an egg is 100 Hz, and the system has a preset bandwidth range of 10 Hz above and below the center frequency, that is, only the signal energy within 90 to 110 Hz is retained; for ease of deduction, assume the frequency domain energy distribution obtained from a certain sampling is: energy 2 at 80 Hz, energy 6 at 95 Hz, energy 8 at 100 Hz, energy 5 at 105 Hz, and energy 7 at 130 Hz; according to the mechanical noise isolation constraint of this embodiment, only the energy at 95, 100, and 105 Hz is retained, while the energy at 80 Hz and 130 Hz is regarded as broadband mechanical noise or non-target vibration components and is discarded; thus, the data used by the system for subsequent energy effective value and phase angle estimation are more focused on the known excitation response, rather than stray vibrations; This constraint is particularly suitable for multi-lane parallel sorting scenarios. Because the rolling of eggs in adjacent lanes will generate asynchronous vibrations through rack coupling, if these vibrations fall in a far frequency band, the feature discrimination of conventional time-domain algorithms is low. However, this embodiment can directly eliminate most of the external interference by retaining the constraint rule of only the center frequency band. In other words, this embodiment does not perform broad global noise reduction processing, but uses a priori reference frequency and narrowband retention mechanism to achieve effective feature-level isolation between useful signals and background disturbances at the algorithm entry point. As an anomaly tolerance mechanism, if the preset bandwidth is set too narrow, it may truncate part of the effective response when the speed fluctuates slightly, resulting in a low effective energy value. If the bandwidth is set too wide, too much non-target energy will be introduced; therefore, this embodiment can adaptively adjust the bandwidth according to the speed stability; for example, when the speed fluctuation rate is less than 2%, the bandwidth is set to ±8 Hz, and when the speed fluctuation rate is greater than 5%, the bandwidth is widened to ±15 Hz. If the cumulative energy in the current frequency band is lower than the minimum confidence threshold, it indicates that the actual excitation is insufficient or the reference frequency estimation is inaccurate, and the current high-frequency feature should be rejected. If the proportion of ultra-bandwidth energy to total energy is consistently higher than the preset upper limit, for example, if 50 consecutive samples exceed 60%, it indicates that there is an abnormal mechanical source in the equipment, and it needs to be shut down for inspection. It is not possible to rely solely on algorithms for feature masking or filtering suppression. For example, during the afternoon shift at the aforementioned egg processing plant, the conveyor tension wheel experienced a slight sway at a certain time, causing the frame to continuously generate mechanical noise around 130 Hz. When egg A was tested, its target dynamic excitation frequency was still 100 Hz. Without the bandwidth constraint of this embodiment, the 130 Hz noise would raise the overall high-frequency energy, affecting the freshness determination. After adopting the mechanical noise isolation constraint, the system only retains the energy in the range of 90 to 110 Hz, and the 130 Hz noise is eliminated. Egg A can still obtain an effective energy value and phase angle consistent with the actual internal rheological state. The purpose of this mechanism is to improve the signal-to-noise ratio of high-frequency feature extraction by using target frequency band preservation and broadband noise removal strategies, so that changes in the mechanical environment of the production line will not significantly reduce the stability of egg freshness detection.

[0024] Furthermore, the target detection object is an egg; the freshness level is correlated with the degree of protein denaturation and moisture evaporation inside the egg through a preset mapping relationship; the excitation track is a corrugated guide rail on the sorting assembly line.

[0025] This embodiment provides an object definition and grade mapping mechanism adapted to the egg industry site; specifically, the detection process can form a complete data link, but in order to make the detection results consistent with the actual quality control target, this embodiment further clarifies the internal biophysical meaning of the target detection object, track morphology and freshness grade; Specifically, in this embodiment, the target detection object is limited to eggs, rather than other spherical or ellipsoidal agricultural products. The reason is that eggs have a closed multiphase structure consisting of a shell, albumen, yolk, and air cell. The degree of protein denaturation and the state of moisture evaporation during storage will jointly affect the rheological properties and center of gravity stability. Protein denaturation and albumen thinning usually reduce the high-frequency coupling rigid response, resulting in a decrease in the effective energy value and an increase in phase lag. Moisture evaporation and changes in the air cell will change the internal mass distribution, resulting in an increase in the low-frequency envelope variance. Therefore, the freshness level in this embodiment is not an abstract label, but a comprehensive internal rheological characterization corresponding to the degree of protein denaturation and the state of moisture evaporation inside the egg. Furthermore, the excitation track is defined as a corrugated guide rail on the sorting assembly line. This guide rail serves two purposes: firstly, it conveys materials, and secondly, it continuously applies controlled excitation to the rolling eggs through a spatial periodic structure on its surface. This allows the entire detection scheme to be completed without interrupting the sorting cycle. The corrugations of the guide rail can be sinusoidal, circular arc-connected, or approximately periodic toothed. Its physical structure must meet the condition of forming a repeatable spatial periodic excitation when the eggs are rolling. The length of its spatial period can be selected based on the common rolling speed of the eggs and the desired detection frequency band, for example, so that the dynamic excitation frequency falls within the sensor's sensitivity range. For the mapping of freshness grades, a two-level, three-level, or more-level grading method can be adopted. Taking a three-level grading method as an example, when an egg is assessed as being in the first rheological state and its effective energy value is significantly higher than the threshold and its envelope variance is significantly lower than the threshold, it can be defined as high freshness. When the indicators are close to the boundary but still in the first rheological state, it can be defined as qualified fresh eggs. When it enters the second rheological state, it can be defined as low freshness or a state awaiting re-inspection. When more refined management is required, the second rheological state can be further subdivided into sub-levels such as critically stale and obviously stale, by combining phase angle, historical batch information, and storage and transportation time. As an anomaly tolerance mechanism, if the object to be detected is not an egg, but is mixed with duck eggs, quail eggs, or broken eggs, the threshold and mapping of this embodiment may produce deviations due to the differences in their geometric dimensions, shell rigidity, and internal medium ratios. Therefore, the system can first confirm the object upstream through shape or image recognition. If it is confirmed that the target detection object is not an egg, switch to the parameter group of the corresponding type, or directly separate the detection process and do not share the same model with the egg. If the wear of the corrugated guide rail causes a significant deviation in the spatial period, it will affect the excitation frequency calculation. In this case, the guide rail period must be recalibrated and the system parameters updated. If the freshness grade continues to deviate from the results of manual sampling, the compensation, demodulation, and calibration parameters should be checked retrospectively. System-level errors should not be masked simply by manually changing the classification mapping label. For example, in the aforementioned egg processing plant, the corrugated guide rails on the sorting line continuously apply spatial periodic excitation to eggs A and B; after testing, egg A shows a high effective value of high-frequency energy and a low variance of low-frequency envelope, indicating that its protein denaturation is low and its moisture evaporation is minimal, and it is sorted into the high-freshness packaging line; egg B shows a decrease in high-frequency energy and an increase in low-frequency envelope variance, indicating that its internal egg white thinning and air cell changes are more obvious, and it is sorted into the freshness deterioration re-inspection line; the entire production line can complete online grading consistent with the internal physiological state of the eggs without manual tapping or stopping the eggs; The purpose of this mechanism is to apply the aforementioned technical features to the specific test object of eggs and its industrial sorting scenario, and to correlate the algorithm output with understandable quality indicators such as the degree of protein denaturation and the state of moisture evaporation, thereby realizing an executable freshness grading for actual egg processing operations.

[0026] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for detecting egg freshness based on an intelligent sensing system, characterized in that, Used for online quality assessment of eggs as the target analyte, including: By laying a piezoelectric sensing array under an excitation track with a preset spatial period sine wave or toothed corrugation, the vertical force change of the target object during rolling is collected in real time, and a dynamic contact force timing signal is generated. The photoelectric speed measurement module set on the side of the excitation track captures the movement time of the target object through the preset detection interval on the excitation track, and then calculates the real-time rolling linear velocity data. Based on the real-time rolling linear velocity data and the spatial period length of the preset spatial period ripple, the dynamic excitation frequency of the target object is calculated. Using the dynamic excitation frequency as a reference signal, the dynamic contact force timing signal is subjected to frequency domain isolation and demodulation processing. High-frequency carrier feature data characterizing the coupling response between the outer shell and the internal medium of the target object, and low-frequency envelope feature data characterizing the internal center of gravity shift of the target object are extracted respectively. The high-frequency carrier feature data includes the effective value of energy and the phase angle, and the low-frequency envelope feature data includes the variance. The high-frequency carrier feature data and the low-frequency envelope feature data are input into a preset classification model to output the internal rheological state evaluation result of the target detection object; based on the internal rheological state evaluation result, the freshness level of the target detection object is determined.

2. The egg freshness detection method based on an intelligent sensing system according to claim 1, characterized in that, The specific steps for extracting high-frequency carrier feature data and low-frequency envelope feature data include: performing quadrature demodulation on the dynamic contact force timing signal using the dynamic excitation frequency, extracting the effective energy value and phase angle in the target frequency band with the dynamic excitation frequency as the center frequency, and generating the high-frequency carrier feature data; Perform a Hilbert transform on the dynamic contact force time-series signal to obtain the basic envelope signal; calculate the variance of the basic envelope signal to generate the low-frequency envelope feature data.

3. The egg freshness detection method based on an intelligent sensing system according to claim 1, characterized in that, The step of inputting the high-frequency carrier feature data and the low-frequency envelope feature data into a preset classification model and outputting the internal rheological state evaluation result of the target object includes: Extract the effective energy value from the high-frequency carrier feature data and the variance from the low-frequency envelope feature data; The effective energy value is compared with a preset energy benchmark threshold, and the variance is compared with a preset variance fluctuation threshold. When the effective energy value is greater than or equal to the energy reference threshold and the variance is less than or equal to the variance fluctuation threshold, the target detection object is determined to be in the first rheological state. When the effective energy value is less than the energy reference threshold, or the variance is greater than the variance fluctuation threshold, the target detection object is determined to be in the second rheological state. The first rheological state or the second rheological state is output as the internal rheological state evaluation result.

4. The egg freshness detection method based on an intelligent sensing system according to claim 1, characterized in that, The method also includes a step of decoupling baseline offsets caused by individual quality differences: Before acquiring the dynamic contact force timing signal, static weight data of the target object is collected; based on the static weight data, the corresponding mass compensation coefficient is matched from a preset baseline compensation mapping table; The amplitude of the dynamic contact force timing signal is normalized using the mass compensation coefficient to generate a standardized contact force timing signal that eliminates interference from mass differences. The standardized contact force timing signal replaces the original dynamic contact force timing signal in subsequent processing steps.

5. The egg freshness detection method based on an intelligent sensing system according to claim 3, characterized in that, The method further includes dynamically calibrating the decision boundary parameters of the preset classification model based on the internal rheological state evaluation results, and the dynamic calibration specifically includes: Collect the internal rheological state assessment results and the corresponding high-frequency carrier feature data and low-frequency envelope feature data within the historical detection period to construct a historical feature distribution set; Calculate the centroid coordinates of the historical feature distribution set in the feature space, and use the distance between the current centroid coordinates and the initial centroid coordinates as the cluster center drift. When the cluster center drift exceeds the preset feature space Euclidean distance drift tolerance, the energy benchmark threshold and variance fluctuation threshold in the preset classification model are updated in the opposite direction of their offset vector using the cluster center drift. When the drift of the aggregation center does not exceed the preset drift tolerance, the current parameters in the preset classification model remain unchanged.

6. The egg freshness detection method based on an intelligent sensing system according to claim 2, characterized in that, The process of performing quadrature demodulation on the dynamic contact force timing signal using the dynamic excitation frequency, extracting the effective energy value and phase angle within the target frequency band centered on the dynamic excitation frequency, and generating the high-frequency carrier characteristic data includes: A local in-phase reference signal and a local quadrature reference signal are generated based on the dynamic excitation frequency; The dynamic contact force timing signal is mixed with the local in-phase reference signal and the local quadrature reference signal respectively to output in-phase and quadrature components. Low-pass filtering is performed on the in-phase component and the quadrature component to filter out high-frequency aliasing noise; Based on the filtered in-phase component and the quadrature component, the effective energy value and the phase angle corresponding to the dynamic excitation frequency are synthesized.

7. The method for detecting egg freshness based on an intelligent sensing system according to claim 1, characterized in that, In the process of using the dynamic excitation frequency as a reference signal to perform frequency domain isolation and demodulation processing on the dynamic contact force timing signal, a mechanical noise isolation constraint is introduced; wherein, the mechanical noise isolation constraint includes: When extracting features, only the signal energy within a preset bandwidth centered on the dynamic excitation frequency is retained, while broadband mechanical noise energy exceeding the preset bandwidth is removed to improve the signal-to-noise ratio of the extracted features.

8. The method for detecting egg freshness based on an intelligent sensing system according to any one of claims 1 to 7, characterized in that, The target detection object is an egg; the freshness level is associated with the degree of protein denaturation and moisture evaporation inside the egg through a preset mapping relationship; the excitation track is a corrugated guide rail on the sorting assembly line.

Images