Photoelectric acquisition and multi-index fused egg yolk oil yield rapid detection method

Abstract

The invention discloses an egg yolk oil release rapid detection method based on photoelectric acquisition and multi-index fusion. According to the method, various characteristic indexes of the yolk in the pickling process are collected through spectral absorption signals and electromagnetic relaxation characteristics, a multi-dimensional characteristic system of an oxidative damage index, a structural collapse index and a water migration index is constructed, a three-layer structured prediction model is provided, and quantitative correlation is established between photoelectric characteristics and yolk oil release behaviors; and rapid prediction of egg yolk oil release is realized. Meanwhile, by combining a multi-index fusion strategy and model interpretability analysis, the stability and reliability of a prediction result are improved. The method can be used for quality control and product grading in the salted egg yolk production process, and has remarkable industrial application value. The method is suitable for online monitoring and product quality grading in the salted egg yolk processing process, and has remarkable industrial application value.

Description

Technical Field

[0001] This invention belongs to the fields of photoelectric detection, food spectral analysis, and intelligent sensing systems, and relates to methods and technologies for rapid evaluation of food quality. Specifically, it relates to a rapid detection method and photoelectric detection system for the oil yield of salted egg yolks based on photoelectric acquisition and multi-index fusion, which can be applied to quality monitoring and online monitoring during the processing of salted egg yolks. Background Technology Egg yolks are prone to oxidation, structural collapse, and oil migration during pickling and heat processing, exhibiting a typical "oil release" phenomenon. The degree of oil release not only affects the taste, flavor, and texture of salted egg yolks but is also an important quality indicator for processed egg products, snack foods, and baked goods fillings. Traditional methods for determining oil yield mainly rely on heating centrifugation or solvent extraction, which involves separating and weighing the oil through heating, centrifugation, or extraction. However, these methods have limitations such as long operation time, strong instrument dependence, and significant sample destructiveness, making them unsuitable for online detection and process quality control, and failing to meet the demands of the modern food industry for rapid and predictable detection technologies. In recent years, some studies have attempted to use single physicochemical indicators, such as thiol content (SH), carbonyl content (Carbonyl), malondialdehyde (MDA), particle size distribution, or low-field NMR T2 parameters, to assess the oxidation state or structural changes of egg yolks. While these indicators can reflect certain characteristics of yolk oxidation or water migration, the yolk system is extremely complex. Oil release is influenced by multiple factors, including protein oxidation, gel network collapse, water migration, and lipid phase changes. A single indicator cannot accurately characterize the entire oil release process, resulting in limited predictive accuracy. On the other hand, although some researchers have attempted to introduce data-driven models to establish regression relationships between physicochemical indicators and quality attributes, most remain at the stage of single-layer linear models or conventional machine learning models, such as multiple linear regression, simple decision trees, or basic random forests. However, firstly, these models have not yet been applied to monitoring oil yield during the yolk curing process. Secondly, these models often ignore the clear causal chain between yolk oxidation, structure, and oil release, lacking physical constraints on the laws of biochemical reactions. Relying solely on data fitting, the models lack interpretability and are difficult to generalize to samples under different curing conditions and varying free radical intensities. More importantly, existing research has not constructed a systematic multi-indicator fusion framework, nor does it lack a method to integrate oxidation characteristics, structural collapse characteristics, and water migration behavior into the prediction system, nor has it established a mechanism for judging the reliability of information fusion. Furthermore, the source of free radicals (such as AAPH and H2O2), the intensity of free radicals, and the concentration of sodium chloride have a significant impact on the degree of egg yolk oxidation and oil release behavior. Existing technologies do not parameterize the intensity of free radicals in different treatment systems such as AAPH, H2O2, and BHT, nor do they provide corresponding structured models to characterize the chain of influence of these systems on the egg yolk oxidation-structure-oil release process. In contrast, this invention introduces a free radical-salt synergistic factor (RSF) to achieve quantitative modeling of oxidation intensity, significantly improving model interpretability and prediction accuracy. Therefore, this patented method is a rapid, accurate, interpretable, and scalable detection method for the degree of egg yolk oil release, overcoming the limitations of traditional detection techniques and existing data models. Summary of the Invention

[0002] The purpose of this invention is to provide a rapid detection method for egg yolk oil yield by combining photoelectric acquisition with multi-index fusion. It collects oxidative damage indicators, structural characteristic parameters and water migration behavior through photoelectric detection methods such as spectral absorption and electromagnetic relaxation. Through multi-layer prediction models and DS evidence theory, it achieves rapid assessment and risk classification of the degree of egg yolk oil release.

[0003] The technical solution of this invention: A rapid detection method for egg yolk oil yield based on photoelectric acquisition and multi-index fusion, the method comprising the following steps: S1: Sample Preparation and Construction of Radical-Salt Factor (RSF): Fresh eggs were separated into egg white and yolk using an egg separator. Excess egg white and chalazae were absorbed from the yolk with filter paper. The yolk membrane was punctured, and the resulting yolk liquid was placed in a pre-cooled beaker and stored at 4°C for later use. Depending on the experimental requirements, the yolk liquid underwent the following four treatments: Sodium chloride treatment: The yolk liquid was mixed with 1% sodium chloride and stirred until completely dissolved. Sodium chloride + AAPH (2,2'-azobis(2-methylpropanediamine) dihydrochloride) treatment: The yolk liquid was mixed with 1% sodium chloride and 5 mM AAPH and stirred until homogeneous. Sodium chloride + BHT (di-tert-butyl-p-cresol) treatment: The yolk liquid was mixed with 1% sodium chloride and 0.05% BHT and stirred until homogeneous. Sodium chloride + H2O2 treatment: Egg yolk liquid was mixed with 1% sodium chloride, 0.01 mM FeCl3, 0.1 mM ascorbic acid, and 10 mM H2O2 and stirred thoroughly. All samples were treated at 25℃ for 0-96 h before physicochemical analysis. Free radical types were recorded, and a free radical-salt synergistic factor (RSF) was constructed: NaCl = 0, BHT = -1 (antioxidant), AAPH = +1 (mild oxidation), and H2O2 = +2 (strong oxidation). The RSF provides a unified numerical expression for different free radical systems (e.g., antioxidant, weak oxidation, strong oxidation), enabling the model to automatically learn the influence of oxidation systems on structural collapse and water migration, achieving generalized predictive capabilities for different treatment conditions. The RSF provides information on the intensity of free radical attack for subsequent models.

[0004] S2: Multidimensional physicochemical indicators were acquired through photoelectric detection and electromagnetic relaxation signal acquisition, including optical absorption signals (412 nm, 370 nm, 532 nm) and electromagnetic relaxation signals (T2 relaxation time and peak area). Based on photoelectric signal conversion, thiol content, carbonyl content, MDA content, particle size characteristics, and moisture migration characteristics were obtained. Specific indicators and methods are as follows: Oxidation indicators: free thiol (DTNB method, 412 nm), carbonyl (DNPH method, 370 nm), MDA (TBA method, 532 nm); Structural characteristics: sample particle size distribution parameters D were acquired using a laser particle size analyzer.10 D 50 D 90 Moisture migration characteristics: Low-field NMR T 21 T 22 T 23 and the corresponding peak area A 21 A 22 A 23 (Echo count 8000–12000). Each indicator was measured three times and the average value was taken.

[0005] S3: Data Preprocessing and Construction of Multi-Indicator Feature System: Feature construction analysis was performed on indicators such as SH, Carbonyl, MDA, particle size, and T2 to construct a comprehensive feature index with clear mechanistic significance. Based on the egg yolk oxidation-structure-lipid release mechanism, this invention proposes the following four comprehensive feature indices for the first time: (1) Construction of Oxidative Damage Index (ODI): , in The reaction reflects the degree of oxidative damage to egg yolk and the intensity of free radical attack, where SH0 is the initial value of the sulfhydryl content of egg yolk, and SH... t ΔSH represents the thiol content at time t. A larger ΔSH indicates more significant oxidative consumption of protein thiol groups. ΔCarbonyl represents the relative increase in carbonyl content, reflecting the degree of protein oxidative breakage and carbonylation, and is calculated as follows: Where Carbonyl0 represents the initial carbonyl level, and Carbonyl... t Let t represent the carbonyl content. ΔMDA represents the relative increase in malondialdehyde (MDA), a lipid oxidation product, and is calculated as follows: The ODI value reflects the intensity of the lipid oxidation chain reaction. α1, α2, and α3 are the weighting coefficients of each indicator, while p, q, and r can be set to real numbers greater than or equal to 1 to adjust the contribution of different oxidative characteristics to the ODI. The higher the ODI value, the higher the degree of oxidative damage to the lipid-protein system in the egg yolk, reflecting the intensity of free radical action and the accumulation of oxidative damage.

[0006] (2) Structural Collapse Index (SCI): This reflects the degree of collapse, aggregation, and fragmentation of the egg yolk gel network. 10 The particle size corresponding to the 10% volume distribution in the particle size distribution reflects the change in the quantity of small-sized particles. (D) 50 Median particle size (median particle size) is used to characterize the average volume variation of the overall particle structure. 90The particle size corresponding to 90% volume distribution reflects the degree of formation of large particle aggregates. These indicators are all derived from the volume distribution results of a laser particle size analyzer and are used to comprehensively describe the collapse, aggregation, and membrane rupture behavior of the yolk structure during oxidation. 50 / D 10 This ratio measures the expansion factor of the median particle size relative to the smallest particle size in a sample. A higher ratio indicates a greater degree of particle expansion, cross-linking, or loosening of the structure. (D) 90 D 10 This represents the particle size distribution width, used to reflect whether the difference in particle size in the system has increased. The larger the value, the more uneven the particle distribution, the more obvious the charge shielding failure, and the more significant the structural collapse. D 500 D is the initial median particle size. 50t Let ΔD be the median particle size at time t. 50 The relative rate of particle expansion can be characterized to reflect the temporal characteristics of structural damage. β1, β2, and β3 are weighting coefficients used to control the contribution of different structural change characteristics in SCI. m, n, and k represent structural sensitivity. Each parameter can be determined by fitting experimental data using statistical learning methods (such as linear regression, least squares, partial least squares, etc.). Alternatively, it can be preset according to specific application needs to adapt to the sensitivity of yolk structural changes under different processing conditions.

[0007] (3) Water Mobility Index. This reflects a decrease in the water-binding capacity of the egg yolk and an increase in lipid migration. 21 Relaxation time of bound water. This reflects changes in bound water flow caused by protein structural collapse. 22 Relaxation time of weakly bound water. This reflects the water spread resulting from the relaxation of the network structure. 23 The relaxation time of free water. T 23 An increase in T usually indicates an increase in free water and improved water phase mobility, which is a precursor to lipid migration. 21,0 T 22,0 T 23,0 T is the initial relaxation time of the egg yolk. 21,t T 22,t T 23,t Let t be the relaxation time. A 23 T represents 23The corresponding peak area percentage represents the free water proportion. This area proportion term reflects the distribution and migration behavior of different water structures. An increase in the free water proportion is often accompanied by lipoprotein network disruption and an increased risk of oil release. γ1, γ2, γ3, and γ4 are weighting coefficients used to adjust the contribution of different water state changes to the WMI. The existence of these coefficients allows the WMI to adapt to water migration behavior under different pickling conditions, different raw material differences, and different oxidation intensities.

[0008] S4: Three-Layer Structured Prediction Model: This invention breaks down the egg yolk oil release mechanism into a three-layer prediction structure: "oxidation → structure → oil release". The first layer predicts oxidation kinetics changes based on RSF and time; the second layer predicts gel structure collapse and water migration behavior based on ODI; the third layer uses SCI and WMI as inputs to assess the amount of egg yolk oil released and outputs low / medium / high risk levels. Specifically, the first-layer oxidation kinetics prediction model inputs: Day, RSF, output: predicted ΔSH, ΔCarbonyl, ΔMDA. Constraints: SH can only decrease over time, and the rate of increase of MDA is limited by the reaction rate. The second-layer structure collapse prediction model inputs: ODI, output: SCI and WMI (particle size diffusion, T2 peak shift). Constraints: irreversible particle size increase, and the T2 peak shift direction is unidirectional. The third-layer oil release prediction model inputs: SCI, WMI, output: predicted oil yield and risk level (low / medium / high), ultimately forming an explainable oil exudation mechanism.

[0009] S5: Multi-indicator fusion and decision-making (DS evidence theory + risk level determination) S5.1 Single-Indicator Model and BPA Construction: Using SH, Carbonyl, MDA, and D respectively 50 T 23 Establish a single-index discrimination model and output the probability of belonging to "low oil production (H0)" or "high oil production (H1)": m ij i = model number, j = category; S5.2 DS fusion rules: m ij The i-th single-index model for hypothesis H j The reliability value is given by i, where i represents the model number and j represents the category number. H0: low oil yield, H1: high oil yield. i0 =The probability that model i is determined to have low oil production, m i1 =Probability that model i determines high oil output. Conflict coefficient between the two models: K represents the degree of conflict between different models; the larger K is, the more inconsistent the decisions between the models. i and k are the model numbers. Multiple models against H... j Joint reliability: , Molecules: All participants in the model for H jThe confidence scores are multiplied, and the denominator is the normalization factor (1–K), used to eliminate the uncertainty caused by the conflict; j ∈{0, 1}; S5.3 Decision Rule - Automatic Classification of Low / Medium / High Risk: Let M i M represents the category (predicted category) corresponding to the highest confidence level. j For the other category of reliability (j≠i), T1 (high-risk threshold) is used to determine whether the joint reliability of the model for a certain category is sufficient to trigger a high-risk level determination. T2 (difference threshold) is used to determine whether the difference between the reliability of the two categories is large enough to distinguish between medium-risk and low-risk situations. Condition 1 is satisfied: M i >T1, then it is judged as high risk; condition 2 is met: M i -M j If the value is >T2, it is classified as medium risk; otherwise, it is classified as low risk. S6: Rapid Detection and Interpretability Analysis of Oil Release: A three-layer structured model is used to rapidly predict oil release from unknown egg yolk samples, outputting a risk level. An attribution algorithm based on Shapley Additive Explanation (SHAP) values ​​is employed to calculate the contribution of different features to the prediction results. SHAP values ​​quantify the marginal contribution of each input feature to the oil release prediction results, decomposing the model's final output into SH, MDA, and D... 50 The additive contribution from features such as T2 is considered. By calculating the magnitude of the positive or negative influence of each feature, the main driving factors of the model's judgment can be displayed intuitively, improving the transparency, interpretability, and traceability of the prediction results, and making the model more consistent with the actual mechanism of food material changes.

[0010] The beneficial effects of this invention are as follows: This invention proposes a rapid detection method for egg yolk oil yield based on photoelectric acquisition and multi-index fusion. It obtains oxidative damage indicators such as thiol, carbonyl, and MDA through spectral absorption signals, and obtains water migration behavior through electromagnetic relaxation signals. For the first time, a multi-index fusion detection system integrating "oxidation-structure-water migration-oil release" is constructed. The method achieves quantifiable, multi-dimensional, and precise characterization of egg yolk oil release behavior through RSF free radical-salt synergistic factor, ODI oxidative damage index, SCI structural collapse index, and WMI water migration index, overcoming the technical bottlenecks of existing technologies that rely on single indicators, have unstable judgments, and cannot reflect free radical-induced structural changes. Compared with traditional oil yield evaluation methods that rely on long-term pickling or centrifugation, this invention significantly improves detection speed, requiring only the acquisition of multi-wavelength absorbance (412 nm, 370 nm, 532 nm) and rapid photoelectric indicators such as electromagnetic relaxation signals (T2) and particle size to predict the degree of oil release. Based on a three-layer structured prediction model, the formation mechanism of egg yolk oil is revealed, achieving causal chain inference from the coupling relationship between photoelectric signals and physicochemical indicators. By integrating the DS evidence theory, the detection instability caused by different free radical systems and different pickling stages is significantly reduced. The model of this invention follows biochemical laws such as protein oxidation, particle size diffusion, and T2 peak migration, avoiding the common black-box model predictions lacking physical meaning or mechanistic basis, making the output results more reliable, interpretable, and consistent with the actual changes in food materials. By combining SHAP contribution analysis, the impact of different indicators on the prediction results can be clearly defined, improving the transparency, traceability, and industrial acceptability of the detection system. The established "low-medium-high oil extraction risk classification model" can be directly used for online monitoring of salted egg production lines, optimization of pickling processes, quality consistency control, and early screening of abnormal batches, demonstrating significant industrial application value. Furthermore, the method of this invention is not only applicable to NaCl pickling systems, AAPH free radical systems, Fenton oxidation systems, and antioxidant systems, but can also be extended to other egg types such as duck eggs and goose eggs, baked egg yolks, powdered egg yolks, quick-frozen egg yolks, and novel salted egg processes (sugar-salt pickling, low-temperature pickling, etc.). It has high applicability and scalability. Attached Figure Description

[0011] Figure 1 Schematic diagram of egg yolk sample preparation process; Figure 2 Overall process framework diagram of the invention; Figure 3 Schematic diagram of the construction of a multi-indicator feature integration system; Figure 4 Curves showing changes in various physicochemical properties; (A) MDA (B) Free thiol group (C) Carbonyl group; Figure 5 Particle size variation; Figure 6Three-layer structured prediction model structure diagram; Figure 7 DS fusion decision model process; Figure 8 Oil yield; Figure 9 Single indicator recognition rate comparison chart (A); Recognition rate comparison before fusion vs. after fusion (B); Three-layer structured prediction model vs. single-layer simple model (performance comparison); ROC curve (D); Figure 10 Comparison of errors between different models; Figure 11 Comparison of predicted oil yield between different input indicators (actual measured value is the experimentally measured oil yield; single indicator is any one of the experimentally measured indicators; multiple indicators are all of the experimentally measured indicators). Detailed Implementation

[0012] The technical solution of the present invention is further illustrated below through specific embodiments, but these are not intended to limit the scope of protection of the present invention.

[0013] Example 1 1. Preparation of egg yolk samples Figure 1 illustrates the standardized preparation process of egg yolk samples in this invention. This process aims to ensure the stability and comparability of subsequent oxidation indicators, structural characteristics, and oil release data. First, fresh eggs are separated into yolk and egg white using an egg separator. Excess egg white and chalaza are removed from the yolk using filter paper. The intact yolk membrane is then gently punctured to collect the yolk liquid in a pre-cooled beaker, which is then cooled at low temperature to reduce interference from enzymatic reactions and auto-oxidation on the experimental results. Next, the yolk liquid is divided into four treatment groups based on different systems: NaCl-only pickling group, NaCl + AAPH peroxide radical group, NaCl + H2O2 hydroxyl radical group, and NaCl + BHT antioxidant group. AAPH and H2O2 simulate mild and strong free radical attacks, respectively, while the BHT group simulates antioxidant conditions for scavenging free radicals. After thorough mixing, each treated sample enters a constant-temperature pickling stage for subsequent collection of multidimensional physicochemical indicators such as SH, Carbonyl, MDA, particle size distribution, and T2 moisture migration.

[0014] 2. Overall process of the invention This invention proposes a rapid detection method for egg yolk oil extraction rate based on photoelectric acquisition and multi-index fusion. The technical process, as shown in Figure 2, mainly includes five parts: sample preparation, multi-dimensional physicochemical index acquisition, construction of a multi-index feature system, establishment of a three-layer structured prediction model, and D-S evidence fusion and risk assessment. The overall framework aims to combine the physicochemical changes of the free radical-salt synergistic oxidation model with a statistical prediction model through a three-stage logic of oxidation kinetics-structural collapse-oil release, thereby achieving rapid identification and classification of egg yolk oil extraction risk.

[0015] S1: Sample Preparation: The main steps include: Egg white and yolk separation: Fresh eggs are selected, and the egg white and yolk are separated to ensure batch consistency. Free radical-salt synergistic oxidation treatment: Egg yolks are placed in different treatment systems, including NaCl, NaCl+AAPH (ROO• model), NaCl+BHT (antioxidant control), and NaCl+H2O2 (•OH model), to simulate the oxidation environment in the actual pickling process. RSF generation: The combined oxidation conditions formed by different treatment systems and treatment intensities (time, free radical concentration) are recorded and encoded as RSFs for subsequent feature system construction.

[0016] S2: Multidimensional Physicochemical Index Collection: Systematic multidimensional parameter detection was performed on the processed egg yolk samples, divided into three main categories: ① Oxidation indicators, used to characterize the degree of oxidative damage to proteins and lipids, including: SH content, MDA, and Carbonyl. ② Particle structure characteristics, characterizing lipoprotein aggregation and particle size changes in the egg yolk, including: D... 10 D 50 D 90 D 90 –D 10 With ΔD 50 ③ Moisture migration characteristics, characterized by low-field nuclear magnetic resonance (NMR) to assess the internal moisture state of the yolk, including: T 21 T 22 T 23 A 21 A 22 A 23 The above multidimensional detection method obtains complete physicochemical information of egg yolk, providing basic data for the construction of the feature system.

[0017] S3: Construction of a Multi-Indicator Feature System: The collected physicochemical indicators are summarized into four modules: ODI oxidation damage index, ΔSH, ΔMDA, ΔCarbonyl, etc. SCI structural collapse index: D 50 / D 10 D 90 –D 10 ΔD 50 etc. WMI water migration dynamics index: ΔT 21 / ΔT22 / ΔT 23 A 21 / A 22 / A 23 The percentage change.

[0018] S4: Three-Layer Structured Prediction Model Establishment: Based on the egg yolk oil release mechanism, this invention breaks down the prediction task into a three-layer structured model: First layer: Oxidation kinetics prediction layer, using the ODI index to predict the degree of egg yolk oxidation, providing an oxidation basis for downstream models. Second layer: Structural collapse prediction layer, based on indicators such as SCI and RSF to predict the rupture and aggregation trends of the protein-lipoprotein complex, reflecting key structural events before oil release. Third layer: Oil release prediction layer, integrating the outputs of the first two layers, WMI change characteristics, and RSF conditions to predict the egg yolk oil release rate and oil release risk level. The three-layer structure forms a hierarchical logic, making the model more closely resemble the actual physicochemical process of egg yolk, improving prediction stability and interpretability. D-S Evidence Fusion and Risk Assessment: To improve the reliability of multi-model decision-making, this invention introduces D-S evidence theory, fusing the credibility of each single-indicator model output to obtain the final BPA. The fusion result is used to output the final risk level, including: low oil leakage risk, moderate oil leakage risk, and high oil leakage risk. This fusion mechanism solves the problems of high uncertainty in multiple features and strong heterogeneity of indicators, making the final oil production prediction more robust.

[0019] 3. Construction of a Multi-Indicator Feature Integration System Figure 3 This invention demonstrates the construction method of the multi-index fusion feature system (S3) proposed in this paper. This system integrates four modules: oxidative damage characteristics, structural collapse characteristics, water migration characteristics, and processing condition encoding of egg yolk under free radical-salt synergistic oxidation, which are used as the core input variables of the prediction model. I. Oxidative Damage Index (ODI): This module reflects the key chemical changes in the oxidation process of egg yolk proteins and lipids, including: ΔSH(SH0–SH0). t Δ: Characterizes the degree of reduction in protein thiol groups caused by free radical attack. ΔCarbonyl: Characterizes the change in carbonyl group formation after protein oxidation and is a typical biomarker of protein oxidative damage. ΔMDA: Changes in malondialdehyde content produced by lipid peroxidation, used to reflect the degree of lipid oxidation. The ODI module is the core input of the first oxidation kinetics prediction layer in the three-layer structure model. II. SCI Structure Collapse Index This module reflects the degree of aggregation, rupture, and structural loosening of lipoprotein complexes. It is a key physical event before the leakage of yolk lipids, including: D 50 / D 10 (Particle size ratio) characterizes the shift in particle size distribution and reflects the degree of aggregation of protein-lipoprotein complexes. D 90 –D 10(Distribution width) is an index of particle polydispersity; a larger value indicates a more heterogeneous structure and a closer proximity to the collapse stage. ΔD 50 (Variability) characterizes D 50 The change relative to the initial value is used to monitor the expansion / shrinkage trend of the structural volume. The SCI module mainly serves the second layer of the model for predicting structural collapse. Third, the WMI moisture migration index, based on low-field NMR, reflects the dynamic changes in the migration of water within the yolk between bound, weakly bound, and free states. The index includes: ΔT 21 ΔT 22 ΔT 23 These represent the relaxation time changes of bound water, weakly bound water, and free water, respectively, which can indicate the degree of looseness of the egg yolk microstructure. A 21 %, A 22 %, A 23 The percentage (%) represents the change in the proportion of water in different states and is an important signal of changes in the water-oil interface. Changes in the T2 peak position are used to identify overall relaxation migration caused by the disruption of the water-lipid interface. The WMI module mainly provides dynamic water migration information for the third-layer oil release prediction layer. IV. RSF Radical-Salt Synergistic Factor: This module encodes the radical system and salt concentration conditions to quantify differences in treatment intensity, including: NaCl encoding: salt-only conditions, used to simulate conventional pickling. NaCl+AAPH encoding: a slow oxidation environment dominated by ROO• (peroxy radical). NaCl+H2O2 encoding: representing a rapid oxidation environment dominated by •OH (strongly oxidizing hydroxyl radical). The RSF module is used to adapt the model to different oxidation systems, giving the prediction system stronger generalization ability. Figure 3 As shown, this invention logically categorizes four types of features—ODI, SCI, WMI, and RSF—to form a complete multidimensional feature matrix. This system enables the representation of the entire process from chemical oxidation to physical structural changes to water migration dynamics, and these indicators can be directly used for model training and D-S evidence fusion, improving prediction reliability. This fusion system is the core technological foundation of the prediction method of this invention.

[0020] 4. Curves showing changes in various physicochemical indicators Figure 4This study demonstrates the trends of key physicochemical indicators of egg yolks over treatment time under different free radical-salt treatment systems (NaCl, NaCl+BHT, NaCl+AAPH, NaCl+H2O2), including: (A) Analysis of TBARS (MDA) content changes. As shown in Figure 4(A), the MDA content in the NaCl+H2O2 group increased the fastest over time, reaching its highest value at 4 days, indicating that hydroxyl radicals (•OH) have a stronger lipid peroxidation capacity. The NaCl+AAPH group was the second fastest, indicating that ROO• (peroxy radical) led to moderate oxidative accumulation. The NaCl+BHT group showed the smallest increase in MDA, confirming that BHT has an effective inhibitory effect on lipid peroxidation. The MDA level in the NaCl-only treatment was lower than that in the two free radical groups, indicating that sodium chloride alone does not significantly induce lipid peroxidation. The significant differences in MDA under different oxidation systems indicate that this invention can rapidly distinguish the degree of lipid oxidation through MDA changes, providing a reliable oxidative kinetic basis for predicting subsequent lipid release. (B) Analysis of changes in free thiol (SH) content: According to Figure 4(B), the SH content decreased most significantly in the NaCl+H2O2 group, indicating that •OH has the strongest destructive effect on protein thiol groups. The decrease in the NaCl+AAPH group was also greater than that in the NaCl group, reflecting that ROO• can also cause protein oxidative damage. The NaCl+BHT group had the highest SH retention rate, indicating that the antioxidant effectively slowed down the protein oxidation reaction. Over time, all four groups of samples showed a decreasing trend in SH to varying degrees. SH can serve as an important indicator of protein oxidation characteristics. This invention incorporates it into the ODI characteristic system, which helps to improve the accuracy of protein oxidation determination. (C) Analysis of changes in carbonyl content: Figure 4(C) shows that carbonyl gradually increased over time, and the NaCl+H2O2 group always maintained the highest level, indicating that protein carbonylation is more obvious in a strong oxidizing environment. The NaCl+AAPH group also showed a stable upward trend, higher than the NaCl group. The NaCl+BHT group had the least carbonyl generation, further verifying the inhibitory effect of BHT. An increase in carbonyl content reflects the cumulative effect of protein oxidation; incorporating it into the ODI module enhances the model's ability to discriminate the extent of protein oxidative damage. Based on... Figure 4 The combined changes in the three types of indicators show that the type and intensity of free radicals have a significant impact on lipid oxidation, protein oxidation, and structural damage in egg yolks. • OH (NaCl + H₂O₂) > ROO • (NaCl + AAPH) > NaCl > NaCl + BHT. The antioxidant (BHT) significantly inhibited MDA growth, decreased thiol groups, and increased carbonyl groups, proving that the oxidation model of this invention can correctly characterize the differences in oxidation intensity. The above physicochemical trends are consistent with lipid release behavior, providing a data foundation for the subsequent construction of a three-layer model. ODI coefficient fitting... First, Δ, ΔΔ, Abonyl[and Δ, Abo are respectively determined by Figure 4 The corresponding indicators are calculated based on the measured values ​​at the initial treatment time and treatment time t. For example, on the fourth day, ΔSH is the original thiol value minus the thiol value on the fourth day, then divided by the original thiol value. ΔCarbonyl and ΔMDA are the values ​​on the fourth day minus the original values, then divided by the original values. α1, α2, and α3 are first normalized, α1 + α2 + α3 = 1. Then, initial empirical ratios are set, such as α1 = 0.35, α2 = 0.25, and α3 = 0.4. First, p, q, and r are set to 1. Then, the data ΔDI ratio and ΔMDA are substituted into the input. Based on the results, r = 1.5 and p = 1.2 are adjusted. This ensures that ODI is between 0 and 1, increasing with the degree of oxidation.

[0021] 5. Particle size variation As shown in Figure 5, different free radical-salt treatment systems significantly affected the particle size of egg yolk particles, reflecting the loosening, aggregation, and collapse behavior of lipoprotein complexes under oxidative conditions. The change in average particle size over time showed an increasing trend in all groups, but significant differences existed in the final particle size and the degree of structural damage. The NaCl group showed a slow increase in particle size from 2028.79 nm to 3323.07 nm, exhibiting moderate particle loosening. While the NaCl+BHT group also increased over time (1462.10 nm - 3026.12 nm), its overall particle size remained lower than other treatment groups, indicating that antioxidants had a stabilizing effect on the egg yolk particle structure. The NaCl+AAPH group showed significant expansion in the middle and later stages (2476.83 nm to 3556.80 nm), indicating that continuous oxidation induced by superoxide free radicals accelerated particle aggregation. According to D... 50 The variation range of particle size distribution reveals certain differences between different treatment systems. The overall variation range can be ranked according to the trend: NaCl + H₂O₂(•OH) > NaCl + AAPH(ROO•) > NaCl > NaCl + BHT. Combined with the broadening and rightward shift of the particle size distribution curve, it can be inferred that the egg yolk microstructure gradually undergoes loosening, cross-linking, and collapse under oxidation. These results indicate that particle size variation can sensitively characterize the degree of oxidative damage to the egg yolk structure and is an important precursor signal for predicting subsequent lipid migration and release behavior. Based on this, this invention quantifies D... 50 D 90 –D 10 and ΔD 50A structural collapse index was constructed using equal particle size characteristics, and further combined with the oxidative damage index and moisture migration index. This significantly enhanced the model's ability to characterize the risk of fat release from egg yolks, making the final prediction results more stable, accurate, and interpretable. The SCI formula was simulated using experimental data, fitted with the particle size on day four. D 10 D represents the particle size corresponding to 10% of the volumetric distribution in the particle size distribution on day four. 50 This represents the median particle size on day four. (D) 90 The particle size corresponding to 90% volume distribution on day four. (D) 50 / D 10 Divide the median particle size on the fourth day by the particle size corresponding to the 10% volume distribution in the particle size distribution. D 90 D 10 The particle size corresponding to 90% of the volume distribution on day four is the particle size corresponding to 10% of the volume distribution. ΔD 50 The median particle size on day four is subtracted from the original median particle size, and then divided by the original median particle size. First, under the assumption of a linear response, m=n=k=1. By adjusting the relative proportions of β1, β2, and β3, SCI is made to monotonically increase with increasing oxidation intensity, correctly distinguishing the degree of yolk structure collapse under NaCl, NaCl+AAPH, NaCl+H2O2, and NaCl+BHT treatment systems. Based on this, nonlinear exponential parameters m, n, and k are introduced to enhance the amplification effect of particle size changes in the later stages of structural damage.

[0022] 6. Low-field NMR As shown in Table 1, different free radical-salt treatment systems significantly affected the T2 relaxation time distribution of egg yolks, reflecting the dynamic changes in the water binding state and the looseness of the microstructure within the yolk. Overall, the T2 signal mainly includes T... 21 (Bound water), T 22 (Weakly bound water) and T 23 (Free water) Three intervals, the relaxation time and peak area of ​​each interval show different patterns with treatment intensity. T in the NaCl group 21 T 22 With T 23 The small variation indicates that simple salt permeation only leads to limited microstructure adjustments; the T in the NaCl+BHT group 21 T 22 T 23 All groups showed minimal change, indicating that antioxidants effectively maintain the stability of the egg yolk microstructure. In contrast, the T values ​​in the NaCl+AAPH group and the NaCl+H2O2 group were significantly lower. 21 T 22All showed a significant increase in time, especially the T value of the NaCl+H2O2 group. 22 The time increased from 20.73 ms to 33.70 ms, T 23 The peak time (T2) increased from approximately 290.29 ms to 439.76 ms, and the peak area (A23) also increased significantly over time. These changes indicate that free radical oxidation gradually disrupts the protein-lipoprotein network, causing bound water to migrate towards weakly bound water and free water, resulting in a looser microstructure, increased pore size, and a higher proportion of free water. The intensity of changes in different systems follows the pattern of NaCl + H2O2(•OH) > NaCl + AAPH(ROO•) > NaCl > NaCl + BHT, consistent with the aforementioned trends in oxidation indicators and particle size changes. These results demonstrate that the T2 relaxation characteristic can sensitively characterize the degree of water migration and microstructure damage in egg yolks, and is a key precursor signal for predicting lipid migration and exudation. Based on this, this invention constructs a water migration index (WMI) to quantify ΔT... 21 ΔT 22 ΔT 23 With A 21 %, A 22 %, A 23 By combining characteristic parameters such as percentage (%) with structural collapse index and oxidative damage index, the characterization depth and prediction accuracy of oil release risk can be significantly improved, providing a reliable data foundation for the subsequent hierarchical output and accurate fusion with D-axis. Fitting the moisture migration index, Taking the fourth day as an example for fitting, ΔT 21 Day 4 T 21 Reduce original T 21， Divide by the original T 21 ΔT 22 Equal to the fourth day T 22 Reduce original T 22 Then divide by the original T 22 ΔT 23 Equal to the fourth day T 23 Reduce original T 23 Divide by the original T 23 ΔA 23 The peak area ratio at time T23 on day 4 is subtracted from the peak area ratio at the initial time T23. Based on the low-field NMR mechanism, the change in free water is greater than the change in weakly bound water, which is greater than the change in tightly bound water, and γ4 ≥ γ3 ≥ γ2 ≥ γ1. 2、 γ3, γ4>0, γ1+γ2+γ3+γ4=1, set initial γ4=0.35, γ3=0.3, γ2=0.25, γ1=0.1, substitute the data, perform parameter identification and correction on the coefficients, so that the obtained WMI can monotonically increase with the treatment time and oxidation intensity, and be consistent with the water migration trend and oil release behavior of the egg yolk system. Table 1. T2 relaxation time of egg yolks in different treatment groups

[0023] Note: Within the same treatment group, different letters indicate significant differences (P<0.05) between different treatment times. 7. Three-layer structured prediction model structure diagram like Figure 6 As shown, the three-layer structured prediction model proposed in this invention is based on a multi-index fusion feature system (S3). It achieves high-precision prediction of egg yolk oil exudation behavior through a three-level progressive modeling process: oxidation kinetics prediction, structural collapse prediction, and oil release prediction. The first layer is the oxidation kinetics prediction layer, which takes the treatment days and free radical-salt co-factor (RSF) as inputs and outputs key feature parameters related to the degree of yolk oxidative damage, including ΔSH, ΔCarbonyl, and ΔMDA, to describe the oxidative damage process of proteins and lipids. The second layer is the structural collapse prediction layer, which takes the ODI oxidative damage index as input and further predicts the evolution trend of the egg yolk microstructure, outputting the structural collapse index (SCI) and water migration index (WMI) (including features such as particle size change and T2 peak position change), reflecting changes in the loosening of the lipoprotein network, increased porosity, and water migration state. The third layer is the oil release prediction layer, which takes the SCI and WMI generated in the previous layer as the main inputs and outputs the oil release rate and oil exudation risk level (low / medium / high) to achieve the final determination of egg yolk oil release behavior. This three-layer structured prediction model organically combines the physicochemical laws of egg yolk quality changes with machine learning models through stepwise decomposition, avoiding the problem of poor prediction stability of traditional single-index or single-layer models. It is an important technical basis for the present invention to achieve rapid and accurate assessment of the risk of egg yolk fat release.

[0024] 8. DS Fusion Decision Model Process like Figure 7 As shown, this invention constructs a multi-indicator fusion judgment model based on the D-S evidence theory to comprehensively assess the risk of egg yolk fat release. First, the single-indicator model assesses SH, MDA, and D respectively. 50 and T 23Key features are predicted, and basic probability assignment functions m1, m2, m3, and m4 are constructed based on numerical changes to represent the confidence level of whether a sample belongs to low oil release (H0) or high oil release (H1). Subsequently, the BPA of the above single indicators is input into the D-S fusion module. The consistency between different pieces of evidence is corrected by calculating the conflict coefficient K, and the fused joint confidence levels M0 (low oil release) and M1 (high oil release) are further calculated, enabling the model to comprehensively consider the correlation and uncertainty between multiple physicochemical features. Finally, the oil release risk level is determined based on the relationship between M1 and M0. This invention adopts a three-level automatic grading rule based on joint confidence: Let M... i M represents the category with the highest reliability (predicted category). j For the other category of reliability (j≠i). When condition 1 is met, when a certain single indicator M i When the risk exceeds the set threshold T1, it is considered high-risk. When condition 2 is met, M... i –M j If the risk exceeds the threshold T2, it is classified as medium risk; otherwise, it is classified as low risk of oil release (H0). Because D-S evidence fusion can effectively handle conflicts, redundancy, and uncertainties among multiple indicators, the judgment model of this invention has higher robustness and accuracy compared to the traditional single threshold method. When SH, MDA, particle size, and T2 relaxation time exhibit different variations under different radical-salt treatment systems, the fusion model can automatically integrate multi-source feature information and output a stable and reliable final risk level.

[0025] 9. Oil yield variation curve like Figure 8As shown, different free radical-salt treatment systems significantly affected the degree of oil release from egg yolks. Overall, the oil yield of each treatment group showed a continuous upward trend over time, but the rate of increase differed significantly from the final level. The oil yield of the NaCl group gradually increased with treatment time, reaching approximately 35% at 4 days, reflecting that salt osmosis can induce a certain degree of oil migration. The oil yield of the NaCl+BHT group remained at the lowest level, still less than 15% at 4 days, indicating that antioxidants can effectively delay microstructural damage and oil exudation. The oil yield increase of the NaCl+AAPH group was greater than that of the NaCl and NaCl+BHT groups, reaching approximately 40% at 4 days, indicating that peroxide radicals (ROO•) can accelerate the destruction of lipids and structural systems in egg yolks, promoting the gradual release of oil. The oil yield of the NaCl+H2O2 group was the highest, reaching approximately 45% at 4 days, much higher than other groups, indicating that hydroxyl radicals (•OH) can lead to stronger oxidative damage and structural collapse, thereby significantly increasing the level of oil exudation. The oil yield growth trend of different systems shows the pattern of NaCl+H2O2(•OH)>NaCl+AAPH(ROO•)>NaCl>NaCl+BHT, which is consistent with the aforementioned results of MDA, SH, Carbonyl, particle size change and T2 relaxation characteristics. This demonstrates that the free radical-salt synergistic oxidation model constructed in this invention can accurately distinguish the microstructure damage and oil migration behavior of egg yolk under different oxidation intensities.

[0026] 10 Comparison of recognition rates of different models under different processing times As shown in Figure 9, this invention systematically evaluated the classification performance of the model from four aspects: single-indicator identification capability, multi-indicator fusion performance, three-layer structured model performance, and ROC curve. Figure 9(A) shows the accuracy of single physicochemical indicators in identifying high and low oil release risks, where MDA, D... 50 With T 23 The recognition rate of the model is significantly higher than that of SH and Carbonyl, indicating that different indicators have different sensitivities to lipid release behavior. Figure 9(B) compares the average recognition rate of multiple single-indicator models with that of the D-S fusion model. The results show that the D-S fusion method can significantly improve the comprehensive recognition ability, effectively integrate multi-source evidence information, and improve the robustness of the model under complex conditions. Figure 9(C) shows the performance comparison between the three-layer structured prediction model and the single-layer model. The results show that the three-layer model is significantly better than the single-layer model in terms of overall accuracy, especially the third layer (comprehensive layer) performs the best, further verifying the effectiveness of the progressive modeling strategy of oxidative damage-structural collapse-lipid release. The ROC curve comparison in Figure 9(D) further demonstrates the classification performance of each indicator and the fusion model from the perspective of AUC value, where D 50 T 23The AUC of both the fusion model and the model are close to 1.0, indicating that they have higher discrimination ability and reliability. Based on the above evaluation results, this invention, by fusing oxidation features, structural features, and water migration features at multiple levels and combining them with D-S evidence theory for risk assessment, can effectively improve the identification accuracy of egg yolk oil release risk. This is significantly better than traditional single-indicator or simple models, and has higher stability, interpretability, and application value.

[0027] Comparison of errors of 11 different models The three-layer structured prediction model proposed in this invention strictly follows the natural causal chain of egg yolk oil release: free radical-induced oxidation – structural collapse and water migration – oil exudation. It decouples the oxidation layer, structural layer, and oil-exuding layer through a mechanism, forming a directional and interpretable prediction framework. During the prediction process, the model can rely on single or a few upstream indicators (such as Day, RSF, D...). 50 Alternatively, Model 1 (or T2) automatically infers the Structural Collapse Index (SCI) and Water Migration Index (WMI) of the downstream level, ultimately achieving high-precision prediction and risk classification of oil release. Therefore, this invention's model not only supports rapid detection of single indicators but also possesses high accuracy when multiple indicators are fused, providing a feasible technical path for rapid industrial monitoring. In contrast, Model 1 mixes multiple structural and water migration indicators in the first layer, disrupting the natural mechanism that protein oxidation precedes structural collapse. This results in the model being unable to perform effective hierarchical inference and requiring a large amount of raw data input to maintain predictive ability. Furthermore, due to the incorrect causal direction, its oxidation kinetic prediction is unstable, ultimately leading to a significant decrease in the accuracy of oil release prediction. Model 2 directly mixes all raw physicochemical indicators as input, without any hierarchical division, making it a typical black-box single-layer model. Such models cannot understand the physical relationships between variables, nor can they automatically complete information when data is missing. Therefore, they not only require the measurement of more indicators, resulting in high detection costs and slow speed, but also have the largest prediction error, making rapid on-site detection impractical. Figure 10 The prediction error histogram further validates the above conclusions: the prediction error of the model in this invention is the lowest (approximately 0.08), significantly better than the faulty model 1 (approximately 0.15) and the mixed model 2 (approximately 0.22). This result indicates that only a three-layer structured model based on the real mechanism of egg yolk oil release can simultaneously achieve rapid detection and high-precision prediction of a single indicator, while the faulty stratified model and the black-box mixed model both require the measurement of more indicators to obtain poorer prediction results, which cannot meet the needs of rapid detection and industrial applications.

[0028] Comparison of Single-Indicator Forecasting and Multi-Indicator Forecasting Figure 11The study presents a comparison of the actual oil yield, single-index input predicted values, and multi-index fusion input predicted values ​​for six groups of egg yolk samples. Figure 11 It can be seen that the three-layer structured model constructed in this invention can output effective predictions regardless of the number of input features. Specifically, in the single-indicator input mode (e.g., only SH, MDA, and D are input),... 50 The model (or any one of T2) can still use the three-layer causal chain of oxidative damage → structural collapse → moisture migration → oil release for inference, thus the predicted trend is consistent with the actual measured value, enabling rapid detection and early screening. However, due to the limited information carried by a single feature, the predicted value of some samples deviates from the actual value. In contrast, under the multi-index fusion input mode (simultaneous input of oxidative damage, structural features, and moisture migration features), the model can directly obtain more complete information on yolk quality changes. The prediction results under the multi-index mode (marked by purple triangles in the figure) are generally closer to the actual measured value, with a significantly smaller error than the single-index mode, showing that the sample points almost coincide with the true values. This result verifies that the three-layer structured model proposed in this invention has higher stability and prediction accuracy under complete input conditions. In summary, Figure 11 The results fully demonstrate that the model of this invention can achieve rapid prediction under single index input and obtain high-precision prediction results under multi-index fusion input, combining speed and accuracy to meet the needs of different detection scenarios.

[0029] The embodiments described above are merely technical solutions used to illustrate the feasibility of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent implementation or modification made by those skilled in the art within the technical scope disclosed in the present invention, based on the technical solution of the present invention, without departing from the spirit of the present invention, should be included within the patent scope of the present invention.

Claims

1. A rapid detection method for egg yolk oil yield based on photoelectric acquisition and multi-index fusion, characterized in that, The method includes the following steps: S1: Sample Preparation and Construction of Free Radical-Salt Cofactor: Fresh eggs were separated into egg white and yolk using an egg separator. Excess egg white and chalazae on the yolk were absorbed with filter paper, and the yolk membrane was broken. The resulting yolk liquid was placed in a pre-cooled container. The yolk liquid was treated with sodium chloride, sodium chloride + AAPH, sodium chloride + BHT, and sodium chloride + H2O2, respectively. All samples were treated at 25℃ for 0-96 h before being used for physicochemical analysis. The types of free radicals were recorded, and the free radical-salt cofactor RSF was constructed. S2: Multidimensional physicochemical indicators are collected through photoelectric detection and electromagnetic relaxation signals: including optical absorption signals and electromagnetic relaxation signals, and the content of thiol group, carbonyl group, MDA content, particle size characteristics and moisture migration characteristics are obtained based on photoelectric signal conversion; S3: Data preprocessing and construction of a multi-index feature system: Establish four comprehensive feature indices, including oxidation features, structural collapse index, water migration index, and free radical-salt synergistic factor; S4: Three-layer structured prediction model: The egg yolk oil release mechanism is decomposed into a three-layer prediction structure of oxidation-structure-oil release; S5: Rapid Detection and Interpretability Analysis of Oil Release: A three-layer structured model was established to rapidly predict oil release from unknown egg yolk samples, outputting the risk level. The contribution of different features to the prediction results was calculated by using an attribution algorithm based on Shapley additive explanatory values; thus, rapid detection of egg yolk oil release rate was achieved.

2. The method according to claim 1, characterized in that, In step S1, the conditions for sodium chloride treatment are as follows: mix egg yolk liquid with sodium chloride and stir until the sodium chloride is completely dissolved; the conditions for sodium chloride + AAPH treatment are as follows: mix egg yolk liquid with sodium chloride and AAPH and stir until the mixture is uniform; the conditions for sodium chloride + BHT treatment are as follows: mix egg yolk liquid with sodium chloride and BHT and stir until the mixture is uniform; the conditions for sodium chloride + H2O2 treatment are as follows: mix egg yolk liquid with sodium chloride, FeCl3, ascorbic acid, and H2O2 and stir until the mixture is uniform.

3. The method according to claim 1, characterized in that, In step S1, the free radical-salt co-factor RSF is the model input: it is used to represent the oxidation strength level corresponding to these treatments, where: NaCl = 0 represents a normal pickling treatment system without the addition of additional free radicals; BHT = -1 represents an antioxidant system, representing a free radical-free oxidation system with the addition of free radical scavengers; AAPH = +1 represents mild oxidation, representing the continuous generation of free radicals, causing moderate oxidation of proteins and lipids; H2O2 = +2 represents strong oxidation, representing the formation of ·OH or other strong oxidants.

4. The method according to claim 1, characterized in that, In step S2, the optical absorption signal corresponds to signals with wavelengths of 412 nm, 370 nm, and 532 nm, and the electromagnetic relaxation signal includes T2 relaxation time and peak area; preferably, the detected indicators are: (1) oxidation indicators: SH-free thiol, Carbonyl-carbonyl, MDA-malondialdehyde; (2) structural characteristics: the sample particle size distribution parameter D is collected using a laser particle size analyzer. 10 D 50 D 90 (3) Water migration characteristics: Low-field NMR T 21 T 22 T 23 and the corresponding peak area A 21 A 22 A 23 Each indicator was measured three times and the average value was taken. The photoelectric detection module includes an optical absorption detection unit, a light source emission unit, a photoelectric conversion unit, an electromagnetic relaxation detection unit, and a signal acquisition unit, which are used to acquire multi-wavelength light absorption signals and low-field NMR T2 signals of egg yolk samples.

5. The method according to claim 1, characterized in that, In step S3, oxidation features are constructed: An exponential function with base e, where The reaction reflects the degree of oxidative damage to egg yolk and the intensity of free radical attack, where SH0 is the initial value of the sulfhydryl content of egg yolk, and SH... t The content of thiol at time t; ΔCarbonyl represents the relative increase in carbonyl content, calculated as follows: Where Carbonyl0 represents the initial carbonyl level, and Carbonyl t Let t represent the carbonyl content at time t; ΔMDA represents the relative increase in malondialdehyde (MDA), a lipid oxidation product, and is calculated as follows: MDA0 represents the initial malondialdehyde level, and MDA t Let be the malondialdehyde content at time t; α1, α2, and α3 are the weighting coefficients of each indicator, used to adjust the contribution of different oxidation characteristics in ODI; p, q, and r are set to real numbers greater than or equal to 1.

6. The method according to claim 1, characterized in that, In step S3, the structural collapse index is: D 10 D represents the particle size corresponding to 10% of the volumetric distribution in the particle size distribution. 50 Median particle size, D 90 : Particle size corresponding to 90% volume distribution; D 50 / D 10 Used to measure the expansion factor of the median particle size relative to the smallest particle size in a sample; D 90 D 10 Indicates the width of the particle size distribution; D 500 D is the initial median particle size. 50t ΔD is the median particle size at time t; 50 The relative rate of particle expansion is represented by β1, β2, and β3, which are weighting coefficients used to control the contribution of different structural change features in SCI. m, n, and k represent structural sensitivity.

7. The method according to claim 1, characterized in that, In step S3, the moisture migration index is: T 21 The relaxation time of water confinement. T 22 The relaxation time of weakly bound water. T 23 The relaxation time of free water T 21,0 , T 22,0 , T 23,0 T is the initial relaxation time of the egg yolk. 21,t ,T 22,t , T 23,t Let t be the relaxation time. A 23 T represents 23 The corresponding peak area ratio, which is the proportion of free water, is represented by γ1, γ2, γ3, and γ4, which are weighting coefficients used to adjust the contribution of different water state changes to WMI.

8. The rapid detection method for egg yolk oil release based on multi-index fusion according to claim 1, characterized in that, In step S4, the three-layer structured prediction model breaks down the egg yolk fat release mechanism into a three-layer prediction structure: The first layer predicts changes in oxidation kinetics based on RSF and time; the second layer predicts gel structure collapse and water migration behavior based on ODI; the third layer uses SCI and WMI as inputs to assess the release of egg yolk fat and outputs low / medium / high risk levels. Preferably, the inputs to the first-layer oxidation kinetics prediction model are: Day and RSF; Output: Predicted ΔSH, ΔCarbonyl, ΔMDA; Constraints: SH decreases over time, and the rate of increase of MDA is limited by the reaction rate; Input to the second-layer structure collapse prediction model: ODI; Outputs: SCI and WMI, where WMI includes particle size diffusion and T2 peak shift; Constraints: irreversible particle size increase, and unidirectional T2 peak shift direction; Inputs to the third-layer oil release prediction model: SCI and WMI; Output: Predicted oil yield and risk level, including low / medium / high risk levels, ultimately forming an explainable oil exudation mechanism.

9. The method according to claim 1, characterized in that, Step S5 includes the following steps: S5.1 Single-Indicator Model Discriminant Model Construction: Using SH, Carbonyl, MDA, and D respectively... 50 T 23 Establish a single-index discrimination model and output the probability of belonging to "low oil extraction H0" or "high oil extraction H1": m ij The i-th single-index model for hypothesis H j The reliability values ​​are: i represents the model number, j represents the category number, H0: low oil yield, H1: high oil yield, m i0 =The probability that model i is determined to have low oil production, m i1 =The probability that model i is determined to be high oil output; S5.2 DS Fusion Rule: Output reliability m of each single-index model ij The conflict coefficient between the two models is used to construct the basic probability distribution BPA. K represents the degree of conflict between different models. The larger K is, the more inconsistent the decisions between the models. i and k are the numbers of the different models. After fusion, the hypothesis H is... j Joint reliability: , Molecules: All participants in the model for H j The confidence scores are multiplied, and the denominator is the normalization factor (1-K), where K is the aforementioned conflict coefficient, used to eliminate the uncertainty caused by the conflict; j∈{0, 1}; j=0 indicates low oil production (H0), and j=1 indicates high oil production (H1). S5.3 Decision Rule - Automatic Classification of Low / Medium / High Risk: M i M represents the predicted category corresponding to the highest confidence level. j For the other category of reliability, j≠i, T1: high-risk threshold, used to determine whether the joint reliability of the model for a certain category is sufficient, and to trigger the high-risk level determination; T2: Difference threshold, used to determine whether the difference between the reliability scores of two categories is large enough to distinguish between medium-risk and low-risk situations; condition 1 is met: M i >T1, then it is judged as high risk; condition 2 is met: M i -M j If the value is >T2, it is classified as medium risk; otherwise, it is classified as low risk. S5.4 Model Adaptability Supporting Partial Feature Input: The three-layer structured prediction model supports partial feature input, i.e., in SH, Carbonyl, MDA, and D... 50 The system can select one or more of the following features from T2 or other features obtained through photoelectric or physicochemical detection as inputs, and output the predicted value of oil release and risk level to achieve self-adaptation to actual detection conditions.

10. The method according to claim 1, characterized in that, In step S6, the Shapley additive explanatory value is quantified based on the marginal contribution of features to the prediction result by calculating each input index, namely SH, MDA, and D. 50 The magnitude of the positive or negative impact of T2 on the predicted value of oil release was determined, and the importance order of each physicochemical characteristic to the model judgment was quantified.

Images