Process for the production of nut products with preserved quality

Abstract

This invention provides a processing method for killing pests and maintaining the quality of nuts, belonging to the field of nut processing technology. The invention involves crushing and pre-dispersing the nut raw materials, then applying wet heat pulse treatment to activate the germination window of mold spores. Subsequently, baking is completed under the coordination of microwave-assisted penetrating heating and an online thermal imaging feedback system. Simultaneously, the nitrogen-filled protective atmosphere is controlled by a baking oxidation index adjustment function to inhibit oil oxidation. After baking, the batch is sampled and tested by nitrogen-filled cooling and a multi-scale feature fusion nut insect egg protein thermal denaturation state evaluation system. This invention solves the technical problem of incomplete inactivation of insect eggs due to uneven heat conduction inside the particle accumulation during nut heat treatment.

Description

Technical Field

[0001] This invention belongs to the field of nut processing technology, and more specifically, relates to a processing method for killing pests and maintaining the quality of nuts. Background Technology

[0002] In the nut processing industry, heat treatment is a core technological means to inactivate insect eggs and ensure product safety. Traditional heat treatment processes typically use fixed temperature and duration baking parameters, relying on internal temperature sensors or manual sampling hatching experiments to evaluate the inactivation effect. In the granular aggregate formed after nut crushing or chopping, the particle packing structure leads to complex heat conduction paths, making it difficult for external heat sources to penetrate evenly throughout the aggregate. Industrial production commonly compensates for insufficient heat conduction by extending baking time or increasing baking temperature.

[0003] However, simply extending the baking time or increasing the temperature exacerbates oil oxidation and quality deterioration. On the other hand, it still cannot fundamentally identify and eliminate the heat conduction blind spots inside the pile. External temperature measurement methods can only reflect the surface or local temperature and cannot obtain the actual temperature history of each position in the three-dimensional space of the pile, resulting in the spatial distribution of insect egg inactivation remaining unknown.

[0004] In current nut heat treatment processes, the lack of three-dimensional real-time modeling capabilities for the internal temperature field of the nut pile means that heat conduction blind spots cannot be located or quantified. Process control relies on conservative global parameter settings, making it impossible to implement targeted compensation for specific low-inactivation areas. This makes it difficult to eliminate the risk of incomplete insect egg inactivation throughout the entire batch treatment cycle. In other words, existing technologies suffer from the technical problem of incomplete insect egg inactivation due to uneven heat conduction within the nut pile during nut heat treatment. Summary of the Invention

[0005] In view of this, the present invention provides a processing method for killing pests and maintaining the quality of nuts, which can solve the technical problem in the prior art that uneven heat conduction inside the nut particle accumulation during the nut heat treatment process leads to incomplete inactivation of insect eggs.

[0006] This invention is achieved as follows: This invention provides a processing method for nuts to kill pests and maintain quality, comprising the following steps:

[0007] The raw materials of nuts are crushed or chopped, and the materials are pre-dispersed by a fluidized bed with airflow disturbance combined with an ultrasonic vibration dispersion device, and the pre-dispersed materials are output.

[0008] The pre-dispersed material is fed into the wet heat pulse treatment chamber, where thermal shock is applied in a high humidity environment to activate the germination window of mold spores. Then, the process is switched to the dry heat baking stage, and the material that has completed the wet heat pulse treatment is output.

[0009] The material that has undergone wet heat pulse treatment is sent into the baking chamber, microwave-assisted penetration heating is turned on, and the temperature field data of the material accumulation is collected in real time using an online thermal imaging feedback system. The heating power is adjusted according to the temperature field data, and the temperature field data and the material in baking are output.

[0010] Using temperature field data as input, a fast estimation algorithm for the inactivation rate of insect eggs in nut particle piles is run, and the distribution map of the inactivation rate of insect eggs in the whole field is output in real time. When there are nodes in the distribution map of the inactivation rate of insect eggs in the whole field with an inactivation rate of insect eggs lower than the threshold, the baking time is extended or the airflow disturbance frequency is increased until the inactivation rate of insect eggs in the whole field is not lower than the threshold.

[0011] The current peroxide value is collected by an online peroxide value detection sensor, and the current baking temperature and current water activity are collected simultaneously. The current peroxide value, current baking temperature and current water activity are input into the baking oxidation index adjustment function to calculate the baking oxidation index. The baking temperature or nitrogen protective atmosphere intensity is adjusted according to the range of the baking oxidation index.

[0012] After baking, the material is cooled under nitrogen-filled and sealed conditions. The batch of products is sampled and tested using a multi-scale feature fusion nut egg protein thermal denaturation state assessment system. The batch processing is judged based on the egg activity probability value. Batches with egg activity probability values ​​exceeding the qualified threshold are returned to the baking step for reprocessing.

[0013] Specifically, the pre-dispersion involves using airflow to disturb the fluidized bed and continuously blowing airflow into the material layer from bottom to top, causing the material particles to suspend and tumble to form a fluidized state. This is combined with an ultrasonic vibration dispersion device to promote the dispersal of particle adhesion and agglomeration.

[0014] Specifically, activating the germination window of mold spores involves rapidly injecting water vapor into the humid heat pulse treatment chamber to raise the water activity of the chamber to the target value, causing the mold spores to enter the germination state. In the germination state, the heat resistance of the spores is significantly lower than that in the dormant state, making them easier to inactivate during the subsequent dry heat baking stage.

[0015] Specifically, the fast estimation algorithm for the inactivation rate of insect eggs in the nut particle pile is to model the nut powder pile as a heterogeneous porous medium, based on the unsteady heat conduction partial differential equation, and to discretize the three-dimensional space of the pile using an adaptive finite element mesh. The temperature value of each node is then iteratively solved using a multi-mesh acceleration strategy.

[0016] Specifically, the temperature values ​​of each node are obtained by inputting the iterative temperature results of each node into the Arrhenius-type insect egg protein thermal denaturation kinetic equation, integrating the temperature history of each node from the start of baking to the current moment, obtaining the integral of the degree of insect egg inactivation at each node, and then summarizing and outputting the distribution map of the inactivation rate of insect eggs across the entire field.

[0017] Specifically, the multi-grid acceleration strategy involves constructing a set of multi-layered grids from coarse to fine. Low-frequency error components are quickly eliminated in the coarse grid layer, and high-frequency local errors are corrected in the fine grid layer. The overall iterative convergence speed is accelerated through inter-layer information transmission.

[0018] Among them, the baking oxidation index satisfy ,in This is the current baking temperature. The baseline baking temperature is used. This is the current peroxide value. As the baseline peroxide value, For the current water activity, As a reference water activity, , , The weighting coefficients and .

[0019] Specifically, the baking temperature or nitrogen protective atmosphere intensity is adjusted according to the baking oxidation index range. Specifically, when the baking oxidation index is lower than the first oxidation index threshold, the current state is maintained; when the baking oxidation index is between the first oxidation index threshold and the second oxidation index threshold, the nitrogen protective atmosphere intensity is increased; when the baking oxidation index is between the second oxidation index threshold and the third oxidation index threshold, the baking temperature is reduced and the nitrogen protective atmosphere intensity is increased simultaneously; when the baking oxidation index is not lower than the third oxidation index threshold, the baking temperature is immediately reduced and the forced nitrogen replacement operation is initiated.

[0020] Among them, the multi-scale feature fusion nut egg protein thermal denaturation state assessment system uses the time series data of spectral intensity of nut powder samples collected by near-infrared spectroscopy sensor as input. The main body of the network consists of two parallel feature extraction branches: a coarse-scale branch and a fine-scale branch. The coarse-scale branch captures the overall envelope change features of the spectrum, while the fine-scale branch captures the detailed changes of the characteristic absorption peaks of the egg protein.

[0021] The pulse outputs of the coarse-scale branch and the fine-scale branch are dynamically weighted and fused by the pulse timing fusion gating module. The pulse timing fusion gating module includes three sets of pulse gating units: input gate, forget gate, and output gate. The fusion ratio of the features of the coarse-scale branch and the fine-scale branch is dynamically allocated through the gating weight.

[0022] The fused pulse feature vector is processed by a lateral suppression layer. The redundant responses of adjacent wavelength channels are mutually suppressed by the lateral suppression mechanism. The output layer is connected to a probability calibration module, and the Platt scaling method is used to map the original network output to the egg activity probability value.

[0023] Among them, the training dataset for the multi-scale feature fusion nut egg protein thermal denaturation state assessment system was established. Specifically, nut powder samples were collected under different combinations of baking temperature, baking time, raw material batch, origin, and season. A conditional generative adversarial network combined with the data augmentation strategy of constraining the thermal death dynamics of insect eggs was used to generate synthetic minority class samples and construct a balanced training dataset.

[0024] Among them, the conditional generative adversarial network combined with the data enhancement strategy of constrained egg thermal death kinetics is specifically that the generator takes the egg activity tag and heat treatment parameters as input conditions, and adds a constraint term to the generator loss function, requiring that the trend of the spectral intensity time series data of the synthesized minority samples be consistent with the degree of protein denaturation predicted by the Arrhenius type egg protein thermal denaturation kinetic equation.

[0025] The training of the multi-scale feature fusion nut egg protein thermal denaturation state assessment system specifically involves encoding the spectral intensity time series data in the balanced training dataset into a pulse firing rate sequence according to the time step and inputting it into the network. The cross-entropy loss of the egg activity probability value is used as the training objective, and a stochastic gradient descent optimizer is used for iterative training.

[0026] The parameters are as follows: particle size after pulverization is 0.5–3 mm; target water activity is 0.85–0.92; baking temperature is 105–115℃; baking time is 50–70 min; compensation range for extended baking time is 5–15 min; first oxidation index threshold is 0.85, second oxidation index threshold is 0.95, and third oxidation index threshold is 1.05; target cooling temperature is below 30℃; and training learning rate range is [not specified]. ~ The training rounds are 100 to 300; the coarse-scale branch receptive field covers 50 to 80 adjacent wavelength channels, and the fine-scale branch receptive field covers 5 to 15 adjacent wavelength channels.

[0027] This invention models the nut particle pile as a heterogeneous porous medium, uses the adaptive finite element method to calculate the three-dimensional temperature field of the pile in real time, and couples the temperature history of each node with the thermal denaturation kinetic equation of Arrhenius-type insect egg protein to output a distribution map of the inactivation rate of insect eggs across the entire field. This transforms the unknown state of the heat conduction blind zone inside the pile into a localizable and quantifiable explicit parameter, solving the defect that external temperature measurement methods cannot detect the internal heat conduction blind zone.

[0028] When the present invention detects nodes below the inactivation rate threshold in the overall insect egg inactivation rate distribution map, it automatically triggers directional compensation operations to extend the baking time or increase the airflow disturbance frequency. This transforms the process control from a passive global conservative setting to an active closed-loop feedback control, avoiding blindly increasing the overall parameters. This ensures the integrity of inactivation while inhibiting unnecessary oil oxidation.

[0029] In summary, the present invention solves the technical problem mentioned in the background art of incomplete inactivation of insect eggs due to uneven heat conduction inside the granular aggregate during nut heat treatment. Attached Figure Description

[0030] Figure 1 This is a flowchart of the method of the present invention.

[0031] Figure 2 A schematic diagram showing the distribution of insect egg inactivation rate and the location of nodes with low inactivation rate across the entire nut particle accumulation.

[0032] Figure 3 Near-infrared spectral response curves of a multi-scale feature fusion system for assessing the thermal denaturation state of nut-bearing insect egg proteins. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below.

[0034] like Figure 1 The diagram shows a flowchart of a processing method for killing pests and maintaining the quality of nuts provided by the present invention. This method includes the following steps:

[0035] S01. The nut raw materials are crushed or chopped. After crushing, the particle size is 0.5-3mm. Then, the material is pre-dispersed by airflow turbulence fluidized bed combined with ultrasonic vibration dispersion device to make the material particles fully loose and output the pre-dispersed material.

[0036] S02. The pre-dispersed material is fed into the wet heat pulse treatment chamber and subjected to thermal shock in a high humidity environment where the water activity is increased to 0.85-0.92 to activate the germination window of mold spores. Then, the process is switched to the dry heat baking stage and the material that has completed the wet heat pulse treatment is output.

[0037] S03. The material that has completed the wet heat pulse treatment is sent into the baking chamber. The baking temperature is set to 105-115℃ and the baking time is 50-70min. Microwave-assisted penetration heating is turned on. The online thermal imaging feedback system is used to collect the temperature field data of the material accumulation in real time. The heating power is adjusted according to the temperature field data to ensure uniform temperature throughout the field. The temperature field data and the material in the baking process are output.

[0038] S04. During the baking process, the temperature field data is used as input to run the fast estimation algorithm of the insect egg inactivation rate field of the nut particle accumulation. The distribution map of the insect egg inactivation rate of the whole field is output in real time. When there are nodes in the distribution map of the insect egg inactivation rate of the whole field with an insect egg inactivation rate lower than the insect egg inactivation rate threshold, the baking time is extended by 5 to 15 minutes or the airflow disturbance frequency is increased until the insect egg inactivation rate of the whole field is not lower than the insect egg inactivation rate threshold.

[0039] S05. During the baking process, the current peroxide value is collected by the online peroxide value detection sensor, and the current baking temperature and current water activity are collected simultaneously. The current peroxide value, current baking temperature and current water activity are input into the baking oxidation index adjustment function to calculate the baking oxidation index. The baking temperature or nitrogen protective atmosphere intensity is adjusted according to the range of the baking oxidation index.

[0040] S06. After baking, the material is cooled to below 30°C under nitrogen-filled and sealed conditions. Then, the batch of products is sampled and tested using a multi-scale feature fusion nut egg protein thermal denaturation state evaluation system. The batch processing is determined to be qualified based on the egg activity probability value output by the system. Batches with egg activity probability values ​​exceeding the qualified threshold are returned to step S03 for reprocessing.

[0041] Among them, airflow-turbulent fluidized bed refers to a heat treatment equipment in which a continuous airflow is blown into the material layer from bottom to top, causing the material particles to suspend and tumble to form a fluidized motion state.

[0042] Among them, the ultrasonic vibration dispersion device refers to an auxiliary mechanical device that uses ultrasonic vibration to promote the dispersal of particles that are stuck together and agglomerated, thereby making the material evenly loose.

[0043] Among them, the wet heat pulse treatment chamber refers to a treatment chamber in which water vapor can be rapidly injected into a closed chamber in a short time to raise the water activity of the chamber to the target value, then apply thermal shock, and then switch to dry heat mode.

[0044] Among them, the germination window period of mold spores refers to the short period of time during which mold spores are activated and enter the germination state under high humidity and heat shock conditions. The heat resistance of germinating spores is significantly lower than that of dormant spores, and they are more easily inactivated in the subsequent dry heat baking stage.

[0045] The water activity enhancement range of 0.85–0.92 was derived from the following: standard mold spores were inoculated and subjected to heat shock treatment under water activity conditions of 0.75, 0.80, 0.85, 0.90, 0.92, and 0.95, respectively. The optimal water activity window range was determined by analyzing data from multiple batches of experiments with the goal of maximizing the product of spore germination activation rate and subsequent dry heat inactivation rate.

[0046] The parameter range of baking temperature (105–115℃) and baking time (50–70 min) was determined by: using almond kernel powder as the subject, it was treated for 30 min, 50 min, 70 min, and 90 min at various temperature gradients of 90℃, 100℃, 105℃, 110℃, 115℃, 120℃, and 130℃, respectively. The parameter range was determined by combining the detection of insect egg hatching rate and the detection of free fatty acid oxidation degree, with the dual objectives of reducing the insect egg hatching rate to 0 and minimizing the increase in acid value.

[0047] Among them, the online thermal imaging feedback system refers to an online monitoring system that continuously acquires images of the surface and internal temperature distribution of the material accumulation in the baking cavity through infrared thermal imaging sensors and transmits the temperature field data to the control module in real time.

[0048] The principle and specific implementation of the fast estimation algorithm for the inactivation rate of insect eggs in nut particle piles are as follows: The nut powder pile is modeled as a heterogeneous porous medium. Based on the unsteady-state partial differential equation of heat conduction, an adaptive finite element mesh is used to discretize the three-dimensional space of the pile. The mesh node spacing is dynamically matched with the local particle size, with a sparser mesh in areas with larger particle sizes and a denser mesh in areas with smaller particle sizes. In each time step, the finite element equations iteratively solve the temperature values ​​of each node through a multi-mesh acceleration strategy. The coarse mesh layer quickly captures the global temperature distribution trend, and the fine mesh layer refines the temperature distribution. Positive local temperature error; the temperature iteration results of each node are synchronously input into the Arrhenius-type insect egg protein thermal denaturation kinetic equation, and the temperature history of each node from the start of baking to the current moment is integrated to obtain the integral of the degree of insect egg inactivation at each node. After summarizing, the distribution map of the insect egg inactivation rate of the whole field is output; when there are nodes in the distribution map of the insect egg inactivation rate of the whole field with an insect egg inactivation rate lower than the insect egg inactivation rate threshold, the system feeds back the node coordinate position and temperature deficit to the equipment control module, triggering the compensation operation of extending the baking time or increasing the airflow disturbance frequency, forming a closed-loop control.

[0049] The technical advantages of the algorithm for rapid estimation of insect egg inactivation rate in nut particle accumulation are as follows: In traditional heat treatment, the heat conduction blind zone caused by particle accumulation is difficult to identify directly through external temperature measurement. However, the method based on finite element field calculation couples the temperature history of each spatial location inside the accumulation with the dynamics of insect egg inactivation, transforming the heat conduction blind zone from a hidden risk into a quantifiable, locatable, and feedback-controllable explicit parameter, fundamentally solving the problem of incomplete inactivation caused by uneven spatial distribution of insect eggs. The multi-grid acceleration strategy ensures that the algorithm completes the calculation within the industrial real-time control cycle, making it engineering feasible.

[0050] The source of the insect egg inactivation rate threshold is as follows: using almond powder with different initial insect egg densities as samples, an actual hatching rate verification experiment was conducted under the finite element simulation temperature history. The insect egg inactivation degree integral corresponding to the hatching rate dropping to 0 was used as the benchmark. After fitting multiple batches of experimental data, the range of values ​​for the insect egg inactivation rate threshold was determined.

[0051] Among them, the multi-grid acceleration strategy refers to a numerical calculation acceleration method that constructs a set of multi-layer grids from coarse to fine during the finite element iterative solution process, quickly eliminates low-frequency error components on the coarse grid, corrects high-frequency local errors on the fine grid, and accelerates the overall iterative convergence speed through inter-layer information transfer.

[0052] Among them, the Arrhenius-type insect egg protein thermal denaturation kinetic equation refers to the kinetic equation based on the Arrhenius equation that describes the relationship between the thermal denaturation rate of insect egg protein at different temperatures and temperature. The integral of the degree of inactivation of insect eggs is obtained by integrating the temperature history.

[0053] The principle and specific implementation of the baking oxidation index adjustment function are as follows: Define the baking oxidation index. The formula is expressed as follows: ;in This is the current baking temperature (unit: °C). The reference baking temperature (unit: °C) Current peroxide value (unit: ), Reference peroxide value (unit: ), The current water activity (dimensionless). The reference water activity (dimensionless). , , These are the weighting coefficients, and Each term in the formula represents a ratio of similar quantities, and all terms are dimensionless, including the baking oxidation index. Dimensionless; when When <0.85, maintain the current baking temperature and nitrogen protective atmosphere intensity unchanged; when At that time, the intensity of the nitrogen protective atmosphere will be increased by 10% to 20%; when At that time, the baking temperature should be reduced by 3-5°C while simultaneously increasing the intensity of the nitrogen protective atmosphere by 20%-30%; when Immediately reduce the baking temperature by 5-8°C and initiate forced nitrogen purging to reduce the oxygen partial pressure in the baking chamber to below 30% of the oxygen partial pressure under standard atmospheric pressure.

[0054] The boundary values ​​of 0.85, 0.95, and 1.05 for the baking oxidation index range are derived from simulations of different baking oxidation indices under laboratory conditions. The acid value of the final product was measured and determined, with the baking oxidation index corresponding to the critical point where the acid value exceeded the standard. The values ​​were used as the dividing line, and the boundary values ​​of each interval were determined through multiple batches of experiments.

[0055] Among them, the reference baking temperature Reference peroxide value Reference water activity All parameters are process standard parameters, determined through multiple batches of comparative experiments comparing the company's raw material quality standards and product quality standards.

[0056] Among them, the weighting coefficient , , The data source is as follows: data on the final acid value increment under different combinations of baking temperature, peroxide value, and water activity are collected. With the goal of minimizing the acid value increment, multiple linear regression is performed on the three weight coefficients to determine the value range of each weight coefficient.

[0057] Nitrogen-filled protective atmosphere refers to the protective gas environment created by filling the baking cavity with nitrogen to replace oxygen, thereby reducing the oxygen partial pressure inside the baking cavity and inhibiting the chain reaction of oil self-oxidation.

[0058] The nitrogen-filled protective atmosphere intensity refers to the flow rate of nitrogen gas injected into the baking cavity per unit time. The higher the nitrogen-filled protective atmosphere intensity, the lower the oxygen partial pressure in the baking cavity.

[0059] The specific structure of the multi-scale feature fusion nut egg protein thermal denaturation state assessment system is as follows: The system uses a near-infrared spectral sensor to collect time-series data of spectral intensity in the 900–2500 nm wavelength range of nut powder samples as input. The main network consists of two parallel feature extraction branches: a coarse-scale branch and a fine-scale branch. The coarse-scale branch is composed of wide-field leaky integral excitation neuron layers, each covering 50–80 adjacent wavelength channels, used to capture the overall spectral envelope variation features. The fine-scale branch is composed of narrow-field leaky integral excitation neuron layers, each covering 5–15 adjacent wavelength channels, used to capture the overall spectral envelope variation features. The system captures detailed changes in the characteristic absorption peaks of oocyte protein in the 1700–1800 nm band. The pulse outputs of the coarse-scale and fine-scale branches are dynamically weighted and fused by a pulse temporal fusion gating module, which includes three sets of pulse gating units: an input gate, a forget gate, and an output gate. The fusion ratio of the coarse-scale and fine-scale branch features is dynamically allocated through the gating weights. The fused pulse feature vector is processed by a lateral suppression layer, where redundant responses of adjacent wavelength channels are mutually suppressed through the lateral suppression mechanism, improving the selectivity of the response to the denaturation characteristic absorption peaks of oocyte protein. The output layer is connected to a probability calibration module, which uses the Platt scaling method to map the original network output to oocyte activity probability values.

[0060] The training dataset establishment steps of the multi-scale feature fusion nut egg protein thermal denaturation state assessment system specifically include: collecting almond powder samples under different combinations of baking temperature, baking time, raw material batch, origin, and season; simultaneously collecting spectral intensity time series data and real egg hatching rate labels for each sample; addressing the class imbalance problem of extremely few positive egg samples, a conditional generative adversarial network combined with egg thermal death dynamics constraint data enhancement strategy is adopted. Using the spectral intensity time series data feature distribution of real positive samples and the egg thermal death dynamics law as constraints, synthetic minority class samples that conform to physical laws are generated. The synthetic minority class samples are mixed with real samples to construct a balanced training dataset.

[0061] The training steps of the multi-scale feature fusion nut egg protein thermal denaturation state assessment system specifically include: encoding the spectral intensity time series data in the balanced training dataset into a pulse firing rate sequence by time step and then inputting it into the network; using the cross-entropy loss of the egg activity probability value as the training objective; and performing iterative training using a stochastic gradient descent optimizer with a learning rate range of [range missing]. ~ The training rounds are 100 to 300. After training, the Platt scaling parameters are fitted on the independent validation set to determine the parameters of the probability calibration module.

[0062] The technical advantages of the multi-scale feature fusion system for assessing the thermal denaturation state of nut egg proteins are as follows: the dual-scale parallel branching structure enables the system to simultaneously capture spectral information at two levels—macroscopic spectral envelope changes and microscopic protein characteristic absorption peak details—during the same inference process; the leak-integration-excited neurons encode spectral intensity time-series data with pulse firing rate, exhibiting sparse activation characteristics, which significantly reduces computational energy consumption while maintaining detection accuracy; the lateral inhibition mechanism compresses redundant feature responses, enhancing the selectivity for identifying characteristic bands of thermal denaturation of egg proteins; and the probability calibration module gives the egg activity probability value statistically significant confidence, facilitating threshold setting and risk quantification for process decisions.

[0063] The principle and technical effects of the data augmentation strategy combining conditional generative adversarial networks (GANs) with egg thermal death kinetic constraints are as follows: GANs introduce conditional labels on top of standard GANs. The generator takes egg activity labels and heat treatment parameters as input to generate synthetic minority class samples that match the distribution of spectral intensity time-series data features of real minority class samples. Egg thermal death kinetic constraints are achieved by adding a constraint term to the generator's loss function, requiring that the trend of spectral intensity time-series data features of the synthetic minority class samples match the degree of protein denaturation predicted by the Arrhenius-type egg protein thermal denaturation kinetic equation, thus avoiding the generation of invalid synthetic minority class samples that contradict physical laws. This strategy effectively alleviates the problem of class imbalance in model training caused by the scarcity of positive egg samples, substantially improving the ability of the multi-scale feature fusion nut egg protein thermal denaturation state assessment system to identify batches of eggs at risk of inactivation. Simultaneously, physical constraints ensure the physical rationality of the synthetic minority class samples, avoiding the distribution distortion problem introduced by purely data-driven augmentation methods.

[0064] Among them, the leakage integral excitation neuron refers to the basic computational unit in the spiking neural network that simulates the accumulation and leakage process of membrane potential in biological neurons. When the accumulated membrane potential exceeds the threshold, a pulse is emitted and the system is reset. Information is transmitted in discrete pulse sequences, which has the computational characteristics of sparse activation and low power consumption.

[0065] The pulse firing rate refers to the ratio of the frequency of pulses fired by neurons within a given time window to the length of the time window, and is used to encode spectral intensity time series data into pulse sequences.

[0066] Among them, the pulse timing fusion gating module refers to the fusion module that is modified from the long short-term memory network gating mechanism to be implemented in the pulse domain. It dynamically weights and fuses the pulse outputs of the coarse-scale branch and the fine-scale branch through three sets of pulse gating units: input gate, forget gate and output gate.

[0067] Among them, the lateral inhibition layer refers to a network layer that uses a lateral inhibition mechanism to enable neurons in adjacent wavelength channels to mutually inhibit redundant responses and improve the selectivity of the response to the absorption peak of the denaturation characteristics of insect egg proteins.

[0068] Among them, the Platt scaling method refers to a post-processing method that maps the original output score of the multi-scale feature fusion nut egg protein thermal denaturation state assessment system to the egg activity probability value by fitting a Sigmoid function on an independent validation set.

[0069] The qualified threshold is derived from the following: through verification experiments on the hatching rate of insect eggs in multiple batches of products, the range of qualified threshold values ​​is determined by statistical analysis of the insect egg activity probability value corresponding to a hatching rate of 0% in the product.

[0070] In a specific embodiment of the present invention concerning almond kernels, the parameter ranges involved are as follows: the particle size of almond kernel powder ranges from 0.5 to 3 mm; the water activity enhancement range during the wet heat pulse treatment stage is 0.85 to 0.92; the dry heat baking temperature ranges from 105 to 115°C; the baking time ranges from 50 to 70 min; the frame rate for acquiring temperature field data by the online thermal imaging feedback system is 1 to 5 frames / s; the fast estimation algorithm for the inactivation rate of insect eggs in the nut kernel accumulation compensates for extending the baking time by 5 to 15 min; and the baking oxidation index... The interval boundary values ​​are 0.85, 0.95, and 1.05. The nitrogen-filled protective atmosphere intensity is increased by 10%–30%. After forced nitrogen replacement triggered by the oxygen partial pressure in the baking chamber, the target value is below 30% of the oxygen partial pressure under standard atmospheric pressure. The target cooling temperature after baking is below 30℃. The near-infrared spectral detection band range of the multi-scale feature fusion nut egg protein thermal denaturation state assessment system is 900–2500 nm. The coarse-scale branch receptive field covers 50–80 adjacent wavelength channels, and the fine-scale branch receptive field covers 5–15 adjacent wavelength channels. The training learning rate range is [missing information]. ~ The number of training rounds ranges from 100 to 300.

[0071] The specific implementation of step S01 is as follows: Nut raw materials are fed into a crushing or shredding device for crushing or chopping. The particle size after crushing is controlled within the range of 0.5–3 mm. The purpose of particle size control is to ensure that heat can penetrate the interior of the particles within a reasonable time during subsequent heat treatment, while avoiding severe scattering loss of excessively fine powder under the action of airflow. After crushing, the material is fed into an airflow-turbulent fluidized bed. Airflow is continuously blown into the material layer from bottom to top, causing the particles to suspend and tumble, forming a fluidized state. Simultaneously, an ultrasonic vibration dispersion device applies ultrasonic vibration, causing the agglomeration structure between particles caused by electrostatic or lipid adhesion to dissolve, resulting in fully loose and uniformly distributed material particles. This lays the structural foundation for uniform heating in the subsequent wet heat pulse treatment and baking stages.

[0072] The specific implementation of step S02 is as follows: The pre-dispersed material is fed into the wet heat pulse treatment chamber. After the chamber is sealed, water vapor is rapidly injected to raise the water activity within the chamber to the target range of 0.85–0.92 within a short period of time. The water activity is selected based on the fact that within this range, the product of the germination activation rate of mold spores and the subsequent dry heat inactivation rate is maximized. In the germinating state, the protein structure of spores is more fragile than that in the dormant state, and their heat resistance is significantly reduced. After reaching the target water activity, a heat shock is immediately applied to activate the mold spores into the germination window period. Subsequently, the process is switched to dry heat mode, so that the spores that have entered the germination state are inactivated in a dry heat environment with a lower heat lethal dose, thereby improving the overall sterilization efficiency.

[0073] The specific implementation of step S03 is as follows: The material that has undergone wet heat pulse treatment is sent into the baking chamber, and the baking temperature is set to 105-115℃, with a baking time of 50-70 minutes. The parameter range is determined based on the dual-objective experimental optimization results of reducing the insect egg hatching rate to 0 and minimizing the acid value increment. During the baking process, microwave-assisted penetration heating is activated. Microwaves can penetrate the interior of the particle accumulation, compensating for the insufficient internal heating due to surface heat conduction and improving the heating uniformity inside the accumulation. The online thermal imaging feedback system continuously acquires images of the surface temperature distribution of the material accumulation in the baking chamber at a frame rate of 1-5 frames / s using an infrared thermal imaging sensor. The temperature field data is transmitted to the control module in real time. The control module dynamically adjusts the heating power based on the temperature field data to ensure that the temperature is uniformly distributed within the set range.

[0074] The specific implementation of step S04 is as follows: Using the temperature field data output by the online thermal imaging feedback system as input, a fast estimation algorithm for the inactivation rate of insect eggs in nut particle accumulations is run in real time. This algorithm models the nut powder accumulation as a heterogeneous porous medium. Based on the unsteady-state partial differential equation of heat conduction, an adaptive finite element mesh is used to discretize the three-dimensional space of the accumulation. The mesh node spacing is dynamically matched to the local particle size, with a sparser mesh in areas with larger particle sizes and a denser mesh in areas with smaller particle sizes. Within each time step, the finite element equations iteratively solve for the temperature values ​​of each node using a multi-mesh acceleration strategy. The coarse mesh layer quickly captures the global temperature distribution trend, while the fine mesh layer corrects local temperature errors. The iterative temperature results of each node are synchronously input into the Arrhenius-type insect egg protein thermal denaturation kinetic equation. The temperature history of each node from the start of baking to the current time is integrated to obtain the integral of the insect egg inactivation degree at each node. After summarizing, a distribution map of the overall insect egg inactivation rate is output. When there are nodes in the distribution map where the inactivation rate of insect eggs is lower than the threshold, the system feeds back the node coordinates and temperature deficit to the control module, triggering a compensation operation to extend the baking time by 5 to 15 minutes or increase the frequency of airflow disturbance, until the inactivation rate of insect eggs throughout the field is not lower than the threshold, thus forming a closed-loop control.

[0075] The specific implementation of step S05 is as follows: During the baking process, the online peroxide value detection sensor continuously collects the current peroxide value, and simultaneously records the current baking temperature and current water activity. These three parameters are input into the baking oxidation index adjustment function, according to the formula... Calculate the baking oxidation index The weighting coefficient , , The result was determined by multiple linear regression from multiple batches of controlled experiments, and the sum of the three factors was 1. When When the value is below 0.85, maintain the current baking parameters; when... When the nitrogen protective atmosphere is between 0.85 and 0.95, the intensity of the nitrogen-filled protective atmosphere will be increased by 10% to 20%; when When the pH is between 0.95 and 1.05, reduce the baking temperature by 3-5°C and simultaneously increase the intensity of the nitrogen protective atmosphere by 20%-30%; when When the oxygen partial pressure is not lower than 1.05, immediately reduce the baking temperature by 5-8°C and start the forced nitrogen replacement operation to reduce the oxygen partial pressure in the baking cavity to less than 30% of the oxygen partial pressure under standard atmospheric pressure, so as to cut off the oxygen source required for the self-oxidation chain reaction of oils.

[0076] The specific implementation of step S06 is as follows: After baking, the material is cooled to below 30°C under nitrogen-filled and sealed conditions. The purpose of nitrogen filling and sealing is to prevent secondary oxidation caused by external oxygen contacting the high-temperature material during the cooling process. After cooling, the batch of products is sampled and tested using a multi-scale feature fusion nut egg protein thermal denaturation state assessment system. This system uses a near-infrared spectral sensor to collect the spectral intensity time series data of nut powder samples in the 900-2500nm band. After parallel feature extraction by coarse-scale and fine-scale branches, the data is dynamically fused through a pulse temporal fusion gating module, and then the redundant response is compressed by a lateral suppression layer. Finally, the probability calibration module outputs the egg activity probability value using the Platt scaling method. The batch is judged as qualified based on the comparison result of the egg activity probability value and the qualified threshold. Unqualified batches are returned to the baking step for reprocessing.

[0077] It should be noted that the key technologies of this invention include: a rapid estimation algorithm for the inactivation rate of insect eggs in nut pellet piles, which transforms the blind zone of internal heat conduction in the pile into a locatable and quantifiable explicit parameter through finite element three-dimensional temperature field modeling, enabling process control to shift from passive global setting to active closed-loop feedback, fundamentally eliminating the defect of traditional temperature measurement methods that cannot detect internal blind zones; a baking oxidation index adjustment function that integrates three key parameters—temperature, peroxide value, and water activity—into a single quantifiable index, and forms a graded linkage control mechanism with the intensity of nitrogen-filled protective atmosphere, enabling quality protection to shift from relying on fixed parameter settings to dynamic response based on real-time oxidation risk; and a multi-scale feature fusion nut insect egg protein thermal denaturation state assessment system that simultaneously captures macroscopic spectral envelopes and microscopic protein absorption peak features through a dual-scale parallel spiking neural network architecture, and uses a data augmentation strategy combining conditional generative adversarial networks and kinetic constraints to solve the problem of scarce positive samples, giving batch qualification judgment a physical degree of confidence. The synergistic effect of the above three key technologies enables the entire process to form a mutually supportive closed-loop system in the three dimensions of heat treatment, oxidation control and quality assessment. Precise control of any dimension depends on real-time information feedback provided by other dimensions, and the overall process robustness is significantly better than the level of each technology operating independently.

[0078] It should be noted that this invention also solves the following technical problem: In the heat treatment of nuts, there is an inherent contradiction between the two objectives of oil oxidation and insect egg inactivation in terms of process parameters. Increasing the temperature or extending the time is beneficial for inactivation but exacerbates oxidation, while decreasing the temperature or shortening the time is beneficial for quality preservation but insufficient inactivation. Traditional processes can only take fixed compromise parameters between the two and cannot dynamically adjust them according to the real-time oxidation state. This invention integrates the current peroxide value, baking temperature, and water activity into a unified quantitative index of oxidation risk through a baking oxidation index adjustment function. It also adjusts the baking temperature and nitrogen protective atmosphere intensity in a graded linkage according to the range of the index. This allows the process parameters to dynamically converge in real time towards quality protection, provided that the integrity of inactivation is guaranteed by the closed-loop algorithm of rapid estimation of insect egg inactivation rate field. Thus, the degree of oil oxidation is controlled to the lowest level without sacrificing the inactivation effect, solving the technical problem that it is difficult to dynamically decouple the parameter coupling contradiction between inactivation integrity and quality preservation in traditional processes.

[0079] Specifically, the principle of this invention is:

[0080] The reason why this invention can solve the above-mentioned technical problems is that traditional external temperature measurement methods can only obtain temperature information on the surface of the accumulation or a limited number of points, and cannot reconstruct the complete three-dimensional temperature history inside. However, this invention regards the accumulation as a heterogeneous porous medium, and through adaptive finite element discretization and multi-grid accelerated iteration, completes the temperature history calculation of each spatial node inside the accumulation within the industrial real-time control cycle, so that the internal heat conduction state can be fully characterized.

[0081] Substituting the temperature history of each node into the Arrhenius-type insect egg protein thermal denaturation kinetic equation and integrating over time, the integral of the degree of insect egg inactivation at each node can be obtained. This physically corresponds to the cumulative degree of irreversible denaturation of the insect egg protein under that temperature history, and has a clear biochemical significance. Therefore, the overall inactivation rate distribution map can objectively reflect the actual inactivation status of each spatial location within the batch.

[0082] When there are nodes below the threshold in the distribution map, the system feeds back the node coordinates and temperature deficit to the equipment control module, triggering targeted compensation operations and forming closed-loop control. This eliminates local heat conduction blind spots without blindly increasing global parameters, which is the fundamental reason why this invention can solve the problem of incomplete inactivation of insect eggs in terms of technical logic.

[0083] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0084] The specific implementation of step S01 is as follows: the nut raw material is crushed or chopped to control the particle size to 0.5-3mm. Then, airflow is continuously introduced from bottom to top into the material layer through airflow disturbance fluidized bed to make the particles suspend and tumble to form a fluidized motion state. At the same time, the ultrasonic vibration dispersion device is started to use ultrasonic vibration to promote the adhesion and agglomeration of particles to disintegrate and output uniform and loose pre-dispersed material.

[0085] The specific implementation of step S02 is as follows: the pre-dispersed material is fed into the wet heat pulse treatment chamber, and water vapor is rapidly injected into the sealed chamber to increase the water activity within the chamber. The water activity was increased to 0.85–0.92, and then thermal shock was applied to activate the mold spores into the germination window. In the germinating state, the heat resistance of the spores was significantly lower than that in the dormant state. Subsequently, the process was switched to a dry heat baking stage, and the material was output after completing the wet heat pulse treatment. The method for determining the water activity increase range of 0.85–0.92 was as follows: [The text abruptly ends here, likely due to an incomplete sentence or missing information.] Standard mold spores were inoculated at levels of 0.75, 0.80, 0.85, 0.90, 0.92, and 0.95 and subjected to heat shock to measure the spore germination activation rate. With subsequent dry heat inactivation rate product With the goal of maximizing water activity, the optimal water activity window range was determined through analysis of multiple batches of experimental data. The spore germination activation rate is dimensionless and ranges from 0 to 1. The value represents the subsequent dry heat inactivation rate, which is dimensionless and ranges from 0 to 1.

[0086] The specific implementation of step S03 is as follows: the material is sent into the baking cavity, the baking temperature is set to 105-115℃ and the baking time is 50-70min, microwave-assisted penetrating heating is turned on, and the surface and internal temperature distribution images of the material accumulation are continuously collected by the infrared thermal imaging sensor at a frame rate of 1-5 frames / s. The temperature field data is transmitted to the control module in real time, and the temperature field data and the material in the baking process are output.

[0087] The specific implementation of step S04 is as follows: using temperature field data as input, a fast estimation algorithm for the inactivation rate of insect eggs in nut particle accumulation is run. The nut powder accumulation is modeled as a heterogeneous porous medium, and the model is based on the unsteady-state heat conduction partial differential equation, as expressed in the following formula:

[0088] ;

[0089] In the formula, For the spatial position of the accumulation body The equivalent density at the location, in units of This represents the equivalent specific heat capacity at the corresponding location, in units of... For position At any moment Temperature, in Kelvin The equivalent thermal conductivity at the corresponding location is expressed in units of 1000 m / s. This is the volumetric heat source term for microwave-assisted heating, in units of... gradient operator These are the three-dimensional spatial coordinates within the accumulation, in meters. The baking time is expressed in seconds. An adaptive finite element mesh is used to discretize the three-dimensional space of the accumulation. The mesh node spacing is dynamically matched to the local particle size; the mesh is sparser in areas with larger particle sizes and denser in areas with smaller particle sizes. A multi-mesh acceleration strategy is used to iteratively solve for the temperature values ​​at each node. A coarse mesh layer quickly captures the global temperature distribution trend, while a fine mesh layer corrects local temperature errors. The iterative temperature values ​​at each node are... The update formula is expressed as follows:

[0090] ;

[0091] In the formula, For the first Node at the next iteration The estimated temperature at that location, in Kelvin. For the first Node at the next iteration The estimated temperature at that location, in Kelvin. For the first In the next iteration, the fine mesh layer is applied to the nodes. The correction amount for the temperature, in K. For the first Spatial coordinates of each finite element mesh node, in meters. The iteration number is a dimensionless positive integer. The temperature iteration results at each node are synchronously input into the Arrhenius-type insect egg protein thermal denaturation kinetic equation. The temperature history from the start of baking to the current moment at each node is integrated to obtain the integral of the degree of egg inactivation. The formula is expressed as follows:

[0092] ;

[0093] In the formula, For position At any moment The integral of the degree of inactivation of insect eggs, dimensionless. Pre-exponential factors, in units of Determined through experimental fitting Activation energy, unit: Determined through experimental fitting Let be the gas constant, and take . The variable is the integral variable, and the unit is s. For position At any moment The thermodynamic temperature, in Kelvin, is obtained from the iterative results of the unsteady-state heat conduction equation. The output is a map showing the overall insect egg inactivation rate distribution. When an insect egg inactivation rate falls below a threshold value... At the critical point, extend the baking time by 5–15 minutes or increase the airflow disturbance frequency to form a closed-loop control. Insect egg inactivation rate threshold. The method for determining the hatching rate is as follows: using almond kernel powder with different initial egg densities as samples, an actual hatching rate verification experiment was conducted under a finite element simulation temperature history. The hatching rate was determined by the value corresponding to a hatching rate of 0. The value was determined based on the data from multiple batches of experiments. Dimensionless.

[0094] The specific implementation of step S05 is as follows: the current peroxide value is collected by an online peroxide value detection sensor. Simultaneously collect the current baking temperature With current water activity Calculate the baking oxidation index The formula is expressed as follows:

[0095] ;

[0096] In the formula, Baking oxidation index, dimensionless This is the current baking temperature, in °C. The reference baking temperature is expressed in °C. This is the current peroxide value, in units of The reference peroxide value is expressed in units of... Current water activity, dimensionless Reference water activity, dimensionless , , Let be the weighting coefficient, satisfying All values ​​are dimensionless. The weighting coefficients are determined by collecting data on the final acid value increment under different combinations of baking temperature, peroxide value, and water activity, with the goal of minimizing the acid value increment. , , Perform multiple linear regression to determine the range of values. , , All parameters are standard process parameters, determined through multiple batch control experiments. Adjust the baking temperature or nitrogen protective atmosphere intensity within the specified range: When At that time, maintain the current baking temperature and nitrogen protective atmosphere intensity unchanged; when At that time, the intensity of the nitrogen protective atmosphere will be increased by 10% to 20%; when At that time, the baking temperature should be reduced by 3-5°C while simultaneously increasing the intensity of the nitrogen protective atmosphere by 20%-30%; when Immediately reduce the baking temperature by 5-8°C and initiate forced nitrogen purging to lower the oxygen partial pressure in the baking chamber to below 30% of the oxygen partial pressure under standard atmospheric pressure. The boundary values ​​of 0.85, 0.95, and 1.05 for each interval are determined by simulating different conditions in the laboratory. The acid value of the final product is determined and measured, with the value corresponding to the critical point where the acid value exceeds the standard. The value is used as the dividing line and has been verified through multiple batches of experiments.

[0097] The specific implementation of step S06 is as follows: After baking, the material is cooled to below 30°C under nitrogen-filled and sealed conditions. A multi-scale feature fusion nut egg protein thermal denaturation state assessment system is then used to sample and test the batch of products. This system uses a near-infrared spectral sensor to collect time-series data of spectral intensity in the 900–2500 nm wavelength range from nut powder samples. As input, where Wavelength, in nm The time of spectral acquisition is expressed in seconds (s). The main network consists of two parallel feature extraction branches: a coarse-scale branch and a fine-scale branch. The coarse-scale branch covers 50–80 adjacent wavelength channels per layer of the receptive field, capturing the overall spectral envelope variation features. The fine-scale branch covers 5–15 adjacent wavelength channels per layer of the receptive field, capturing the detailed changes in the characteristic absorption peaks of oocyte proteins in the 1700–1800 nm band. The spectral intensity time series data are encoded according to the pulse firing rate, where the pulse firing rate is... The formula is expressed as follows:

[0098] ;

[0099] In the formula, The pulse firing rate is expressed in units of... The time window is defined as the number of time pulses fired by neurons due to the time window's leakage integral; it is a dimensionless positive integer. The time window length is in seconds. The encoded pulse sequence is input to the coarse-scale branch and the fine-scale branch, and the pulse outputs of the two branches are... and Dynamic weighted fusion is performed via a pulse timing fusion gating module, which includes an input gate. Forgotten Gate With output gate The three pulse gating units are expressed by the following formulas:

[0100] ;

[0101] ;

[0102] ;

[0103] ;

[0104] In the formula, , , These are the gating weight vectors for the input gate, forget gate, and output gate, respectively. They are dimensionless, and each element takes a value between 0 and 1. Element-wise sigmoid activation function , , These are the weight matrices for the input gate, forget gate, and output gate, respectively. They are dimensionless and learned during the training process. , , These are the bias vectors corresponding to the input gate, forget gate, and output gate, respectively. They are dimensionless and learned during the training process. The pulse feature vector output by the coarse-scale branch is dimensionless. The pulse feature vector of the fine-scale branch output is dimensionless. Represents vector concatenation operation Represents element-wise product Element-wise hyperbolic tangent activation function This is the dimensionless pulse feature vector after fusion. After processing with a lateral suppression layer, redundant responses between adjacent wavelength channels are mutually suppressed through the lateral suppression mechanism, thereby improving the selectivity of the response to the characteristic absorption peak of egg protein denaturation. The formula for the lateral suppression layer is as follows:

[0105] ;

[0106] In the formula, For the lateral inhibition layer, the first Each wavelength channel corresponds to the output pulse firing rate of the neuron, dimensionless. To fuse feature vectors The Middle The pulse firing rate of each channel, dimensionless To fuse feature vectors The Middle The pulse firing rate of each adjacent channel, dimensionless The lateral suppression weighting coefficients are dimensionless. The neighborhood range for lateral suppression, expressed in channels. Index of the current wavelength channel To and Adjacent wavelength channel indices. The output layer connects to a probability calibration module, which uses Pratt scaling to adjust the original network output scores. Mapped to the probability value of insect egg viability The formula is expressed as follows:

[0107] ;

[0108] In the formula, This is the probability value of insect egg viability, dimensionless, and ranges from 0 to 1. The original output score of the network is dimensionless. , The Pratt scaling parameter, obtained by fitting on the independent validation set, is dimensionless. The training dataset was created by collecting almond kernel powder samples under different combinations of baking temperature, baking time, raw material batch, origin, and season. Data was collected synchronously for each sample. The data is labeled with the actual egg hatching rate. To address the class imbalance problem caused by the extremely small number of positive egg samples, a data augmentation strategy combining a conditional generative adversarial network with egg heat-induced death dynamics constraints is employed, reducing the generator's total loss. The formula is expressed as follows:

[0109] ;

[0110] Among them, the physical constraint loss term The formula is expressed as follows:

[0111] ;

[0112] In the formula, The generator's total loss is dimensionless. To counteract the loss term, a dimensionless parameter is used to drive the generator to produce synthetic samples that match the distribution of the real minority class samples. The physical constraint loss weight coefficient is dimensionless. The loss term is a dimensionless constraint term representing the kinetics of heat-induced death of insect eggs. To synthesize the number of minority class samples, a dimensionless positive integer. For the first The estimated inactivation degree of each synthetic sample after network feature extraction, dimensionless. The kinetic equation for the thermal denaturation of proteins in Arrhenius-type insect eggs is based on the first... The predicted inactivation degree is obtained by calculating the heat treatment parameters corresponding to each synthetic sample, dimensionless. The baseline inactivation level is a dimensionless reference value, taken as the mean of the integral values ​​of the inactivation level of known completely inactivated samples in the training dataset. A balanced training dataset is constructed by mixing synthetic minority class samples with real samples. During training, the cross-entropy loss of the egg viability probability is used as the training objective, and a stochastic gradient descent optimizer is employed for iterative training with a learning rate ranging from [value missing]. ~ The training iterations range from 100 to 300. After training, the Pratt scaling parameters are finalized on an independent validation set. , The fitting was based on the egg viability probability value output by the system. To determine whether a batch processing is qualified, Batches exceeding the acceptable threshold are returned to step S03 for reprocessing. The acceptable threshold is determined by conducting an egg hatching rate verification experiment on multiple batches of products, with the hatching rate of 0 corresponding to... The value was used as a benchmark, and the range of acceptable threshold values ​​was determined through statistical analysis of multiple batches of experimental data.

[0113] To better understand and implement this invention, the following is a specific application scenario of the invention, Example 2: To verify the effectiveness of the invention, technicians set up a test environment, selecting almond kernels from the same batch and origin as raw materials, with an initial moisture content of 5.8% and an initial peroxide value of The initial insect egg density was set to 50 live insect eggs per 100g of material through artificial inoculation to fully simulate the worst-case scenario of natural pollution. After pulverization, the particle size distribution was concentrated in the range of 1-2mm. After pre-dispersion using an airflow-induced fluidized bed and an ultrasonic vibration dispersion device, the particle agglomeration rate was reduced from approximately 23% to approximately 4%, significantly improving the uniformity of particle dispersion and laying a structural foundation for uniform heating in subsequent stages.

[0114] During the wet heat pulse treatment stage, the water activity within the cavity increased to 0.88 within approximately 90 seconds, followed by a heat shock lasting about 3 minutes. The activation rate of mold spore germination was approximately 82%, after which the system switched to dry heat mode. The decrease in spore heat resistance during germination improved the spore inactivation efficiency during the subsequent baking stage, demonstrating the physical rationality of the wet heat pulse activation strategy.

[0115] The initial baking temperature was set at 110℃, and the target baking time was 60 minutes. Microwave-assisted penetration heating was activated, with the microwave penetration depth covering more than 70% of the pile thickness. An online thermal imaging feedback system continuously acquired temperature field data at a frame rate of 3 frames per second. After running a fast estimation algorithm for the insect egg inactivation rate of the nut pile, at the 38-minute mark of baking, the overall insect egg inactivation rate distribution map showed two nodes where the insect egg inactivation rate was below the threshold. Figure 2 As shown, all nodes are located in the central region at the bottom of the accumulation, where the heat conduction rate is low due to the high particle density. The system automatically triggers a compensation operation to extend the baking time by 10 minutes and simultaneously increases the airflow disturbance frequency. By the 50th minute, the inactivation rate of insect eggs throughout the field has reached or exceeded the inactivation rate threshold, and the compensation operation ends.

[0116] The changes in peroxide value recorded by the online peroxide value detection sensor during the baking process are shown in Table 1:

[0117] Table 1. Changes in key process parameters and baking oxidation index during baking.

[0118]

[0119] As shown in Table 1, the baking oxidation index was [value missing] at 20 minutes of baking. When the concentration rises to 0.88, the system automatically increases the nitrogen protective atmosphere intensity by 15%; by the 40th minute... The temperature was increased to 0.97, and the system lowered the baking temperature by 2°C to 108°C while simultaneously increasing the intensity of the nitrogen-filled protective atmosphere by 25%; this was done at the end of the 70-minute compensation baking period. The peroxide value was 1.01, the baking temperature had been reduced to 103℃, and the nitrogen protective atmosphere intensity was increased by 30%. The forced nitrogen replacement operation corresponding to the third oxidation index threshold was not triggered throughout the process. The final batch product peroxide value was... It is lower than the national standard. The upper limit and quality remain in good condition.

[0120] After baking, the material was cooled to 28°C under nitrogen-filled and sealed conditions. Subsequently, batches of products were sampled and tested, with each sample weighing 50g. The multi-scale feature fusion nut egg protein thermal denaturation state assessment system output egg activity probability values ​​for each sample were all below the acceptable threshold of 0.05. The test results for each sample are shown in Table 2.

[0121] Table 2. Probability values ​​of insect egg viability in batch sampling tests

[0122]

[0123] As shown in Table 2, samples S-04 and S-05, located in the central region at the bottom of the accumulation, had relatively high egg viability probabilities. Figure 2 The location of the low inactivation rate node identified by the algorithm matches the data, verifying the accuracy of the fast estimation algorithm for the inactivation rate field of insect eggs in nut particle accumulations in locating the internal heat conduction blind zone. After compensation, the probability values ​​of insect egg activity in the samples in this area are all below the qualified threshold, and the batch is deemed qualified.

[0124] like Figure 3As shown in this embodiment, the multi-scale feature fusion system for assessing the thermal denaturation state of nut egg proteins, in the response curve of sample S-04 in the near-infrared spectrum 900–2500 nm band, captures the broadband absorption changes caused by protein denaturation at the overall envelope level through the coarse-scale branch, while the fine-scale branch identifies the details of peak shape changes in the characteristic absorption peak region of the 1700–1800 nm band. The fusion of the two significantly improves the discrimination resolution of the thermal denaturation state of egg proteins. The lateral inhibition layer effectively compresses the redundant response outside this band, resulting in a high confidence level for the final output egg activity probability value.

[0125] The advancements of this invention compared to traditional methods are reflected in the following aspects: Traditional heat treatment processes rely on fixed global parameter settings, lacking the ability to perceive the internal thermal conduction state of the accumulation. They can only ensure the inactivation effect through conservative high-temperature, long-term parameters. In contrast, this invention calculates the temperature history of each spatial location within the accumulation through finite element three-dimensional temperature field modeling and couples it with the thermal denaturation kinetic equation of insect eggs. This allows for a comprehensive quantification of the internal thermal conduction state, transforming process control from blind global increases to precise directional compensation. This ensures the integrity of inactivation while avoiding unnecessary parameter increases that could damage quality. Traditional quality monitoring relies on offline sampling and hatching experiments after baking, resulting in delayed feedback and the inability to intervene in real time. This invention uses a baking oxidation index adjustment function to quantify oxidation risk in real time during baking and adjusts process parameters in a graded and linked manner, shifting quality protection from post-inspection to proactive process control. The multi-scale feature fusion nut egg protein thermal denaturation state assessment system incorporates both the macroscopic envelope and microscopic absorption peak characteristics of near-infrared spectra using artificial intelligence methods. It also solves the engineering problem of scarce positive samples through physical constraint data enhancement, making the confidence and robustness of batch qualification determination superior to traditional hatching experiment methods.

[0126] It should be noted that the variables involved in this invention are explained in detail in Tables 3 and 4.

[0127] Table 3. Variable Explanation Table (Part 1)

[0128]

[0129] Table 4. Variable Explanation Table (Part Two)

[0130]

[0131] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A processing method for killing pests and maintaining the quality of nuts, characterized in that, Includes the following steps: The raw materials of nuts are crushed or chopped, and the materials are pre-dispersed by a fluidized bed with airflow disturbance combined with an ultrasonic vibration dispersion device, and the pre-dispersed materials are output. The pre-dispersed material is fed into the wet heat pulse treatment chamber, where thermal shock is applied in a high humidity environment to activate the germination window of mold spores. Then, the process is switched to the dry heat baking stage, and the material that has completed the wet heat pulse treatment is output. The material that has undergone wet heat pulse treatment is sent into the baking chamber, microwave-assisted penetration heating is turned on, and the temperature field data of the material accumulation is collected in real time using an online thermal imaging feedback system. The heating power is adjusted according to the temperature field data, and the temperature field data and the material in baking are output. Using temperature field data as input, a fast estimation algorithm for the inactivation rate of insect eggs in nut particle piles is run, and the distribution map of the inactivation rate of insect eggs in the whole field is output in real time. When there are nodes in the distribution map of the inactivation rate of insect eggs in the whole field with an inactivation rate of insect eggs lower than the threshold, the baking time is extended or the airflow disturbance frequency is increased until the inactivation rate of insect eggs in the whole field is not lower than the threshold. The current peroxide value is collected by an online peroxide value detection sensor, and the current baking temperature and current water activity are collected simultaneously. The current peroxide value, current baking temperature and current water activity are input into the baking oxidation index adjustment function to calculate the baking oxidation index. The baking temperature or nitrogen protective atmosphere intensity is adjusted according to the range of the baking oxidation index. After baking, the material is cooled under nitrogen-filled and sealed conditions. The batch of products is sampled and tested using a multi-scale feature fusion nut egg protein thermal denaturation state assessment system. The batch processing is judged based on the egg activity probability value. Batches with egg activity probability values ​​exceeding the qualified threshold are returned to the baking step for reprocessing.

2. The processing method for killing pests and maintaining the quality of nuts according to claim 1, characterized in that, The pre-dispersion specifically involves using airflow to disturb the fluidized bed and continuously blowing airflow into the material layer from bottom to top, causing the material particles to suspend and tumble to form a fluidized state. This is combined with an ultrasonic vibration dispersion device to promote the dispersal of particle adhesion and agglomeration.

3. The processing method for killing pests and maintaining the quality of nuts according to claim 2, characterized in that, The activation of the mold spore germination window period specifically involves rapidly injecting water vapor into the humid heat pulse treatment chamber to raise the water activity of the chamber to the target value, causing the mold spores to enter the germination state. In the germination state, the heat resistance of the spores is significantly lower than that in the dormant state, making them easier to inactivate in the subsequent dry heat baking stage.

4. The processing method for killing pests and maintaining the quality of nuts according to claim 3, characterized in that, The algorithm for fast estimation of insect egg inactivation rate field of nut particle accumulation is specifically designed to model the nut powder accumulation as a heterogeneous porous medium. Based on the unsteady heat conduction partial differential equation, an adaptive finite element mesh is used to discretize the three-dimensional space of the accumulation, and the temperature value of each node is solved iteratively through a multi-mesh acceleration strategy.

5. The processing method for killing pests and maintaining the quality of nuts according to claim 4, characterized in that, Specifically, the temperature values ​​of each node are obtained by inputting the iterative temperature results of each node into the Arrhenius-type insect egg protein thermal denaturation kinetic equation, integrating the temperature history of each node from the start of baking to the current moment, obtaining the integral of the degree of insect egg inactivation at each node, and then summarizing and outputting a distribution map of the insect egg inactivation rate across the entire field.

6. The processing method for killing pests and maintaining the quality of nuts according to claim 5, characterized in that, The multi-grid acceleration strategy specifically involves constructing a set of multi-layered grids from coarse to fine. Low-frequency error components are quickly eliminated in the coarse grid layer, and high-frequency local errors are corrected in the fine grid layer. The overall iterative convergence speed is accelerated through inter-layer information transmission.

7. The processing method for killing pests and maintaining the quality of nuts according to claim 6, characterized in that, Baking oxidation index satisfy ,in This is the current baking temperature. The baseline baking temperature is used. This is the current peroxide value. As the baseline peroxide value, For the current water activity, As a reference water activity, , , The weighting coefficients and .

8. The processing method for killing pests and maintaining the quality of nuts according to claim 7, characterized in that, The baking temperature or nitrogen protective atmosphere intensity is adjusted according to the baking oxidation index range. Specifically, when the baking oxidation index is lower than the first oxidation index threshold, the current state is maintained; when the baking oxidation index is between the first and second oxidation index thresholds, the nitrogen protective atmosphere intensity is increased; when the baking oxidation index is between the second and third oxidation index thresholds, the baking temperature is decreased and the nitrogen protective atmosphere intensity is increased simultaneously; when the baking oxidation index is not lower than the third oxidation index threshold, the baking temperature is immediately decreased and a forced nitrogen replacement operation is initiated.

9. The processing method for killing pests and maintaining the quality of nuts according to claim 8, characterized in that, The multi-scale feature fusion system for assessing the thermal denaturation state of nut egg protein uses time-series data of spectral intensity from nut powder samples collected by a near-infrared spectral sensor as input. The main network consists of two parallel feature extraction branches: a coarse-scale branch and a fine-scale branch. The coarse-scale branch captures the overall envelope variation of the spectrum, while the fine-scale branch captures the detailed changes in the characteristic absorption peaks of the egg protein.

10. The processing method for killing pests and maintaining the quality of nuts according to claim 9, characterized in that, The pulse outputs of the coarse-scale branch and the fine-scale branch are dynamically weighted and fused by the pulse timing fusion gating module. The pulse timing fusion gating module contains three sets of pulse gating units: input gate, forget gate, and output gate. The fusion ratio of the features of the coarse-scale branch and the fine-scale branch is dynamically allocated through the gating weight.

Images