A method for controlling the baking process of Bazhen powder raw materials based on image processing

Abstract

This invention belongs to the field of intelligent control technology in food processing, and relates to a method for controlling the baking process of Bazhen powder raw materials based on image processing. The method involves synchronously acquiring multi-source image sequences of eight raw materials at adaptive sampling intervals; performing spatial registration and denoising preprocessing on the images; constructing a humidity-heat mutual interference field and calculating the equivalent temperature decay between any two raw materials; defining a normalized maturity index; calculating the actual effective baking temperature of each raw material based on the equivalent temperature decay; calculating the surface moisture diffusion rate pixel by pixel and determining the charring risk under dual conditions; dynamically determining the maximum safe temperature change rate based on the degree of deviation of the thermal inertia parameters of the eight raw materials; comparing the incremental effect of continued baking on the overall quality, and determining the optimal termination time when the incremental effect is negative and persists for multiple sampling times. This invention overcomes the shortcomings of existing technologies that suffer from inaccurate temperature control due to neglecting the humidity-heat mutual interference between raw materials.

Description

Technical Field

[0001] This invention belongs to the field of intelligent control technology for food processing, and in particular relates to a method for controlling the baking process of Bazhen powder raw materials based on image processing. Background Technology

[0002] Bazhen Powder is made from eight medicinal and edible ingredients: yam, lotus seed, poria cocos, fox nut, coix seed, white hyacinth bean, atractylodes macrocephala, and codonopsis pilosula. The process involves baking, grinding, and mixing. Baking is the core step that determines the color, aroma, taste, and retention rate of the effective components in the finished product. Because the Bazhen Powder formula requires the eight ingredients to be mixed in a fixed ratio, in production practice, all eight ingredients must be baked and stopped together in the same oven. However, the initial moisture content, density, specific heat capacity, and characteristic thickness of the eight ingredients vary significantly. Yam has a moisture content of approximately 75%, while codonopsis pilosula has only about 13%. This results in significantly different maturation rates for each ingredient under the same baking conditions. Furthermore, the evaporative cooling effect caused by the large amount of water evaporated from the high-moisture-content ingredients can cause a significant localized cooling effect on adjacent ingredients, further exacerbating the uneven distribution of effective temperatures among the ingredients.

[0003] Existing technologies lack quantitative models for the spatial range and intensity of the mutual interference effect of evaporation and cooling between raw materials. Temperature control is based on a set temperature rather than the actual effective temperature experienced by each raw material. The control strategy aims to make each raw material approach its target independently, without optimizing the simultaneous arrival of the eight raw materials at similar times. This results in over-baking when some raw materials have already met the target but are forced to wait for the non-targeted raw materials. The phase-locked relationship that drives color and texture changes to maintain coordination during normal baking is not utilized, making it impossible to distinguish between synchronous normal changes and out-of-step abnormal changes. Summary of the Invention

[0004] This invention addresses the technical problems associated with the simultaneous roasting of eight ingredients in Bazhen powder, proposing a well-designed, simple, and theoretically sound method for controlling the roasting process of Bazhen powder ingredients based on image processing.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] S1. A visible light camera, a thermal infrared camera, and a multispectral camera, which have been jointly calibrated, are arranged above the baking device to synchronously acquire multi-source image sequences of eight raw materials during the baking process at an adaptive sampling interval.

[0007] S2. Spatial registration is performed on the visible light image, thermal infrared image and multispectral image acquired synchronously at each sampling time. Using the visible light image as the reference coordinate system, the thermal infrared image and multispectral image are mapped to a unified coordinate space, and the registered images are preprocessed for denoising.

[0008] S3. Extract the spatial temperature distribution above each raw material tray from the thermal infrared image at the current sampling time, and extract the spatial moisture concentration distribution above each raw material tray from the ratio of the reflectance of the moisture-sensitive band to the reflectance of the reference band in the multispectral image; jointly encode the spatial temperature distribution and the spatial moisture concentration distribution according to the relative spatial position of each raw material tray to construct a humid heat interference field, and calculate the equivalent temperature attenuation between any two raw material positions in real time.

[0009] S4. Define a normalized maturity index for each raw material. The normalized maturity index uses the water molecule index corresponding to the rate of change of the multispectral moisture estimate as a pre-multiplicative factor, and the color sub-index corresponding to the rate of change of the L value of the brightness channel in the color Lab space, the volume sub-index corresponding to the rate of change of the raw material outline area in the visible light image, and the texture sub-index corresponding to the rate of change of the surface texture roughness are weighted and summed in square brackets. The volume weight, color weight, and texture weight are determined by the dynamic weight coefficients generated by the water molecule index through the quadratic Bernstein basis function.

[0010] S5. At each sampling moment, the equivalent temperature decay amount described in S3 is used as input to calculate the actual effective baking temperature of each raw material, and the adjustment amount of baking temperature and circulating wind speed is dynamically generated so that the ripening progress of the eight raw materials tends to be synchronized and approaches their respective target ripening values ​​at similar times.

[0011] S6. Calculate the surface moisture diffusion rate of the raw material pixel by pixel from the multispectral images of adjacent sampling times. When the surface moisture diffusion rate of any local area exceeds the preset threshold and the local temperature of the area in the thermal infrared image is higher than the average temperature of the raw material by a preset temperature difference, it is determined that there is a risk of coking and local intervention is triggered.

[0012] S7. Pre-determine the thermal inertia parameters of each of the eight raw materials, namely the product of density, specific heat capacity and characteristic thickness. At each sampling time, calculate the degree of deviation of the thermal inertia parameters of the eight raw materials. Based on this, dynamically determine the maximum safe temperature change rate. The larger the deviation, the smaller the allowable temperature change rate, in order to prevent the low thermal inertia raw materials from overheating and the high thermal inertia raw materials from responding with lag. The temperature intervention amount triggered by S5 and S6 is constrained by the maximum safe temperature change rate.

[0013] S8. Define the collective baking quality function as the negative value of the Mahalanobis distance between the current normalized maturity index of each raw material and its target maturity value. At each sampling time, compare the current collective quality that is about to be terminated with the expected collective quality. When the increment of the collective quality due to continued baking is negative and this holds true for multiple consecutive sampling times, it is determined that the optimal termination time for the collective has been reached, and a baking termination command is output.

[0014] Preferably, in step S1, the eight ingredients are placed in their respective independent baking trays, and the trays are arranged in a preset spatial layout within the baking device.

[0015] Preferably, the formula for calculating the equivalent temperature decay in S3 is as follows:

[0016] ,in, Location of the raw materials for evaporation source Location of the disturbed raw materials The resulting equivalent temperature decay, The evaporative cooling intensity coefficient, This is an estimated current moisture content of the raw material used in the evaporation source. For position With position Spatial distance between them The attenuation characteristic length is related to the thermal diffusion properties of air. This refers to the airflow speed within the baking apparatus. It is the angle between the direction of the circulating airflow and the line connecting the positions of the two raw materials.

[0017] Preferably, the formula for calculating the normalized maturity index in S4 is as follows:

[0018] ,

[0019] , ,

[0020] , ,

[0021] , , ,in, For the first At the sampling time of the raw materials The normalized maturity index, For water molecule indicators, For color weights, For color sub-indices, For volume weights, For volume index, For texture weights, For texture sub-indices, The mean value of L at the initial time. For the first raw materials at all times The mean L value of the Lab space brightness channel. The mean of the target L value, This is the initial water content. For the first raw materials at all times The estimated water content, For the target moisture content, The target contour area, This represents the initial contour area. For the first raw materials at all times The visible light image outline area, For the first The surface fractal dimension of the raw material at time t. For the target fractal dimension, Let be the initial fractal dimension.

[0022] Preferably, the formula for calculating the effective baking temperature in step S5 is:

[0023] ,in, For the first The actual effective baking temperature experienced by the raw materials. Set the temperature for the current baking setup. The saturated cooling coefficient, Current temperature The saturated water content of the air below. This represents the average moisture content within the current baking space. Location of the raw materials for evaporation source Location of the disturbed raw materials The resulting equivalent temperature decay.

[0024] Preferably, the formula for calculating the surface moisture diffusion rate of the raw material in S6 is:

[0025] ,in, For the first The raw materials at the pixel position Sampling time Surface moisture diffusion rate, Pixel-level moisture index, For the first The critical water loss coefficient of the raw material. For the first The characteristic length of moisture replenishment of the raw materials. To prevent division by zero constant, This represents the spatial gradient magnitude of the moisture index.

[0026] Preferably, the formula for calculating the maximum safe temperature change rate in S7 is:

[0027] ,in, This represents the maximum safe rate of temperature change allowed at the current sampling time. Based on the allowable temperature change rate, This is the harmonic average of the thermal inertia parameters of the eight raw materials. This is the arithmetic mean of the thermal inertia parameters of the eight raw materials. This refers to the actual temperature inside the baking apparatus. This refers to the maximum permissible operating temperature of the baking equipment. This is the temperature risk acceleration index.

[0028] Preferably, the formula for calculating the current collective quality in S8 is:

[0029] ,

[0030] , ,in, Sampling time The overall baking quality is represented by a negative value; the closer it is to zero, the closer the overall baking state of the eight ingredients is to the ideal target. This represents the normalized maturity vector of the eight raw materials at the current sampling time. Let each of the eight raw materials be a target maturity vector. This is the matrix transpose operator. For the correlation structure matrix, This is the arithmetic mean of the thermal inertia parameters of the eight raw materials. For the first Thermal inertia parameters of the raw materials, Location of the raw materials for evaporation source Location of the disturbed raw materials The resulting equivalent temperature decay, Location of the raw materials for evaporation source Location of the disturbed raw materials The resulting equivalent temperature decay, The current temperature. diagonal elements It is a non-diagonal element.

[0031] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0032] 1. By constructing a humid-thermal mutual interference field and defining an equivalent temperature decay, the local cooling effect of one raw material on other raw materials caused by surface moisture evaporation in multi-raw material co-baking is quantitatively described. This equivalent temperature decay explicitly expresses the functional relationship between evaporative cooling intensity and the moisture content of the evaporation source raw material, the spatial distance between the two raw materials, the circulating wind speed, and the wind direction angle. This upgrades the calculation of the actual effective baking temperature of each raw material from relying on the set temperature to being based on the true effective temperature, overcoming the defect of inaccurate temperature control caused by neglecting the thermal-humid mutual interference between raw materials in existing technologies.

[0033] 2. By defining a normalized maturity index and minimizing the dispersion of maturity deviations among the eight raw materials as the control objective, the asynchronous maturity synchronizing approach is established as the control paradigm. Unlike existing methods that optimize each raw material independently, this method prioritizes applying favorable temperature control to the raw material that deviates furthest from its target value. This synchronizes the maturity progress of the eight raw materials, allowing them to approach their respective differentiated target maturity values ​​at similar times. This resolves the contradiction of over-roasting that occurs when some raw materials have already met the target but are forced to wait for those that have not.

[0034] 3. By calculating the surface moisture diffusion rate pixel by pixel, moisture monitoring is upgraded from static moisture content estimation to dynamic diffusion rate spatial distribution reconstruction. Coking risk is determined by combining the dual conditions of moisture diffusion rate exceeding a threshold and local temperature exceeding the average temperature. This method can locate the local area with the fastest water loss rate on the raw material surface, achieving pixel-level spatial accuracy for early warning. It overcomes the limitation of existing methods that rely solely on overall moisture content thresholds and cannot locate local high-risk coking areas. Attached Figure Description

[0035] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This is a schematic diagram of a method for controlling the baking process of Bazhen powder raw materials based on image processing. Detailed Implementation

[0037] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described below in conjunction with the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0038] Numerous specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways than those described herein, and therefore the invention is not limited to the specific embodiments disclosed in the following specification.

[0039] Examples, such as Figure 1 As shown, a jointly calibrated visible light camera, thermal infrared camera, and multispectral camera are arranged above the baking device to simultaneously acquire multi-source image sequences of eight raw materials during the baking process at an adaptive sampling interval. The visible light, thermal infrared, and multispectral images acquired simultaneously at each sampling moment are spatially registered. Using the visible light image as a reference coordinate system, the thermal infrared and multispectral images are mapped to a unified coordinate space, and the registered images undergo denoising preprocessing. The use of three-source cameras for joint acquisition and adaptive adjustment of the sampling interval is because a single-modal camera cannot comprehensively capture the three-dimensional changes in color, temperature, and moisture content of the raw materials during baking. Visible light reflects surface color and morphology, thermal infrared reflects temperature distribution, and multispectral reflects moisture content; these three complement each other to provide a complete data foundation for subsequent construction of the humidity-heat interference field and calculation of maturity indicators. The adaptive sampling interval solves the problems of redundancy caused by fixed frequency in the preheating stage, omission of key changes in the main baking stage, and untimely monitoring in the drying stage. Spatial registration using visible light images as a reference is necessary because the different physical installation positions of the three cameras result in non-coincident pixel coordinates for the same raw material in the three images. Without registration, temperature and moisture values ​​would correspond to different physical locations, leading to spatial misalignment in the construction of the humidity-heat mutual interference field. The noise reduction preprocessing addresses the noise characteristics of different modal images: thermal infrared exhibits non-uniform noise, multispectral weak signal bands have low signal-to-noise ratios, and visible light bright areas have random noise. Adaptive algorithms are used to suppress these noises to prevent them from being amplified into false moisture diffusion hotspots during the difference calculation between adjacent frames, thus ensuring the accuracy of subsequent calculation steps.

[0040] The spatial temperature distribution above each raw material tray is extracted from the thermal infrared image at the current sampling time. The spatial moisture concentration distribution above each raw material tray is extracted from the ratio of reflectance of the moisture-sensitive band to the reflectance of the reference band in the multispectral image. The spatial temperature distribution and spatial moisture concentration distribution are jointly encoded according to the relative spatial position of each raw material tray to construct a humid-thermal mutual interference field. The equivalent temperature attenuation between any two raw material positions is calculated in real time. The formula for calculating the equivalent temperature attenuation is:

[0041] ,in, Location of the raw materials for evaporation source Location of the disturbed raw materials The resulting equivalent temperature decay, The evaporative cooling intensity coefficient, This is an estimated current moisture content of the raw material used in the evaporation source. For position With position Spatial distance between them The attenuation characteristic length is related to the thermal diffusion properties of air. This refers to the airflow speed within the baking apparatus. The angle between the circulating airflow direction and the line connecting the two raw material positions is represented by the numerator. The current moisture content of the evaporation source raw material is located in the numerator because the evaporation rate is positively correlated with the current moisture content; the higher the moisture content, the more intense the evaporation and the stronger the cooling effect. As baking progresses, the moisture content gradually decreases, and the cooling effect automatically decays without the need for manual stage switching. The spatial distance between the two raw materials is represented by the square in the denominator because in a closed baking space, the water vapor concentration decreases approximately inversely proportionally to the square with distance before becoming uniform. The square decay more accurately describes the physical characteristic of strong cooling at close range and weak cooling at far range than linear decay. The circulating air velocity appears in both the numerator and denominator, capturing the dual effect of air velocity on evaporation and cooling. On the one hand, the higher the air velocity, the thinner the air boundary layer on the raw material surface, the higher the evaporation rate, and the stronger the cooling effect. On the other hand, the higher the air velocity, the faster the water vapor is carried away from the vicinity of the evaporation source, and the influence range of the local cooling effect is compressed. The structure of the denominator, which is the sum of the square of the distance and the square of the wind speed, means that when the two raw materials are very close, the denominator is dominated by the square of the wind speed, and the higher the wind speed, the weaker the local cooling. When the two raw materials are far apart, the denominator is dominated by the square of the distance, and the higher the wind speed, the stronger the cooling over a long distance. This captures both the promoting effect of wind speed on evaporation and the inhibiting effect on water vapor accumulation. The cosine of the angle between the circulating airflow direction and the line connecting the two raw materials reflects the anisotropy of evaporative cooling. Cooling is strongest when the disturbed raw material is downwind of the evaporation source, followed by the crosswind direction, and weakest upwind direction, reflecting the directional control of the circulating airflow direction on the direction of water vapor transport. The evaporative cooling intensity coefficient and the water vapor mixing characteristic parameter are two parameters that have been calibrated through pre-experiments. The former integrates the influence of physical constants such as the latent heat of vaporization of water and the heat capacity of air, while the latter characterizes the speed of water vapor mixing in a closed space, making the formula adaptable to baking equipment of different specifications.

[0042] A normalized maturity index is defined for each raw material. This index is a weighted composite of four sub-indices: the rate of change of the L-value of the lightness channel in the Lab color space, the rate of change of the estimated moisture content in the multispectral image, the rate of change of the raw material's outline area in the visible light image, and the rate of change of surface texture roughness. The formula for calculating the normalized maturity index is as follows:

[0043] ,

[0044] , ,

[0045] , ,

[0046] , , ,in, For the first At the sampling time of the raw materials The normalized maturity index, For water molecule indicators, For color weights, For color sub-indices, For volume weights, For volume index, For texture weights, For texture sub-indices, The mean value of L at the initial time. For the first raw materials at all times The mean L value of the Lab space brightness channel. The mean of the target L value, This is the initial water content. For the first raw materials at all times The estimated water content, For the target moisture content, The target contour area, This represents the initial contour area. For the first raw materials at all times The visible light image outline area, For the first The surface fractal dimension of the raw material at time t. For the target fractal dimension, Using the initial fractal dimension, the color sub-index quantitatively characterizes surface browning caused by Maillard and caramelization reactions using Lab lightness (L) values; the water molecule sub-index quantitatively characterizes the internal dehydration process using the ratio of reflectance in the multispectral moisture-sensitive band to the reference band; the volume sub-index quantitatively characterizes tissue collapse caused by cell dehydration using the area of ​​the raw material outline in the visible light image; and the texture sub-index quantitatively characterizes surface roughening caused by crack propagation and scorch formation using the surface fractal dimension. All four are indispensable. Each sub-index is uniformly expressed as a normalized ratio of the changed amount divided by the target total change because the dimensions and magnitudes of change of the four dimensions differ significantly. Direct weighting would allow the moisture index, with the largest change magnitude, to completely dominate the result. After normalization, all four are expressed as the percentage of their respective dimensions' progress towards the target, making them comparable and compatible. The water molecule index is not included in the weighted summation within square brackets but is used as a pre-multiplicative factor because water evaporation is the physical prerequisite for browning, volume shrinkage, and texture roughening. The Maillard reaction requires appropriate water activity, cell collapse depends on the capillary negative pressure generated by water loss, and surface cracks only occur on the brittle surface after drying. If dehydration is insufficient, the water index value is very low. Therefore, no matter how close the other three sub-indicators are to the target value, the overall maturity is suppressed, thus eliminating the misjudgment of false coloring from a mathematical structure perspective. The three dynamic weights are generated from the water molecule index via a quadratic Bernstein basis function, rather than being preset as fixed constants. This is because the dominance of each dimension differs at different stages of baking, and the transition between stages is solely determined by the rate of water evaporation: the volume weight uses the square of the complementary value of the water index, reaching its maximum in the early stages of baking when moisture is plentiful and cells rapidly lose water and collapse, making volume shrinkage the primary indicator for maturity assessment; the color weight uses twice the product of the water index and its complementary value, peaking at approximately 0.5, precisely corresponding to the physical law that the Maillard reaction has the highest rate at moderate water activity, automatically shifting the maturity assessment in the middle stages of baking to be dominated by color changes; the texture weight uses the square of the water index, strongly suppressed to zero in the early stages of baking to avoid misjudging noisy texture changes on still-wet surfaces as maturity, and rapidly rising to its maximum in the later stages of baking when moisture is nearly depleted, making surface cracks and scorch marks the core dimensions for quality assessment. The three basis functions naturally sum to one, and the weight transfer is entirely driven automatically by the endogenous physical quantity of water evaporation, without the need to introduce any external time parameters or artificial stage discrimination rules. The reason for using a weighted sum rather than a product structure within the square brackets is that color, volume, and texture change in parallel under the unified drive of moisture, rather than having a strict causal order. The weighted sum allows each of the three factors to contribute independently, and the sum of the three weights is always 1. The lag in one dimension can be partially reflected by the progress of other dimensions, unlike the product structure where if any dimension is zero, the whole is zero.The target values ​​for each raw material are set differently based on their functional roles in the finished product of Bazhen Powder. The target values ​​for raw materials with high flavor contribution weight are set slightly higher to allow for slight over-roasting to enhance the caramel aroma. The target brightness for raw materials with high color contribution weight is set higher to maintain a light-colored appearance. The target moisture content for raw materials with high heat sensitivity of medicinal components is set higher to prioritize the retention rate of effective components. This makes the maturity index not only measure the degree of completion of roasting, but also take into account the differentiated quality requirements of each raw material in the final product.

[0047] At each sampling moment, the equivalent temperature decay amount described in S3 is used as input to calculate the actual effective baking temperature of each raw material, dynamically generating adjustments to the baking temperature and circulating air velocity, so that the ripening progress of the eight raw materials tends to be synchronized and approaches their respective target ripeness values ​​at similar times; the formula for calculating the effective baking temperature is:

[0048] ,in, For the first The actual effective baking temperature experienced by the raw materials. Set the temperature for the current baking setup. The saturated cooling coefficient, Current temperature The saturated water content of the air below. This represents the average moisture content within the current baking space. Location of the raw materials for evaporation source Location of the disturbed raw materials The equivalent temperature decay is calculated. The current baking apparatus set temperature serves as the baseline for temperature control and is the starting point for all temperature adjustment commands. Starting from this temperature, the cooling effect caused by adjacent evaporative cooling is subtracted from each raw material, ensuring the calculation result reflects the actual effective thermal environment experienced by each material. Simultaneously, the set temperature, as a stable baseline, filters out measurement noise and local airflow fluctuations introduced by direct temperature measurement from thermal infrared images. The evaporative cooling intensity coefficient integrates physical constants such as the latent heat of vaporization of water, the specific heat capacity at constant pressure of air, and density, aggregating these dispersed physical quantities into a parameter that can be calibrated through pre-experiments, making the formula concise and physically based. The difference between the saturated water content of the air at the current temperature and the average water vapor content in the current baking space measures how much water vapor the air can still hold. This is the physical upper limit of the evaporative cooling effect. In the early stages of baking, the environment is dry, the capacity is large, and the cooling effect is significant. In the later stages of baking, water vapor accumulates, the capacity automatically decreases, and the cooling effect naturally decays, without the need for manual setting of stage switching logic. The equivalent temperature decay represents the independent cooling contribution of a single evaporation source when there is no competition from other evaporation sources. Each evaporation source has a different equivalent temperature decay due to differences in location, water content, distance from the disturbed raw material, and wind direction. This element incorporates the spatial analysis results of the hygrothermal interference field into the temperature correction calculation. The ratio of the equivalent temperature decay of each evaporation source to the saturated cooling limit indicates the proportion of remaining coolable space consumed by that evaporation source, rather than the absolute temperature. When the saturated cooling limit is large, the proportion of the same equivalent temperature decay is small, indicating that there is ample coolable space. When the saturated cooling limit is small, the proportion of the same equivalent temperature decay is large, indicating that the remaining coolable space has been largely consumed by other evaporation sources, and the marginal cooling effect of newly added evaporation sources is compressed. The product structure of subtracting the ratio transforms the combined effect of multiple evaporation sources from independent accumulation to proportional consumption. Each evaporation source consumes not absolute temperature, but a proportion of the remaining coolable space after the previous evaporation source's consumption. This structure perfectly aligns with the physical principle that water vapor accumulates in a finite volume within a closed space, rather than diluting infinitely. The evaporative cooling effect does not increase linearly with the number of evaporation sources but exhibits a decreasing marginal effect. The product of subtracting the ratio represents the total proportion of coolable space consumed by all evaporation sources. Multiplying this by the upper limit of saturated cooling yields the total equivalent cooling amount. When there is only one evaporation source, subtracting the ratio equals the ratio itself, degenerating into a direct subtraction. When the individual cooling amounts of multiple evaporation sources are all much less than the upper limit of saturation, the total cooling is approximately the sum of their individual cooling amounts. When multiple evaporation sources evaporate violently simultaneously, the total cooling is always less than the upper limit of saturated cooling and never exceeds this physical limit. The effective baking temperature of each raw material is obtained by subtracting the total equivalent cooling amount from the set temperature. The greater the difference between the effective temperature and the set temperature, the more severely the raw material is affected by the evaporative cooling of adjacent materials. Its delayed maturity may not be due to its own characteristics but rather passively affected. Based on this, the control system can distinguish between deviations caused by the raw material's own characteristics and deviations caused by the drag from adjacent materials, thereby generating the correct differentiated temperature control strategy.By replacing linear accumulation with a product-proportional structure, the combined cooling effect of multiple evaporation sources is ensured to always be less than the saturation upper limit, which conforms to the physical law of limited water vapor accumulation in a closed space. The difference between the saturated water content and the current water vapor content is used as the dynamic cooling upper limit, so that the inclusion of the cooling effect is automatically adjusted according to the ambient humidity. It is fully included during the initial drying stage of baking and automatically decays during the later water vapor accumulation stage. The effective temperature is calculated for each raw material, so that the temperature control is upgraded from being based on the set temperature to being based on the actual thermal environment experienced by each raw material.

[0049] The surface moisture diffusion rate of the raw material is calculated pixel-by-pixel from the multispectral images of adjacent sampling times. When the surface moisture diffusion rate of any local area exceeds a preset threshold, and the local temperature of that area in the thermal infrared image is higher than the average temperature of the raw material by a preset temperature difference, a risk of coking is determined, and local intervention is triggered. The formula for calculating the surface moisture diffusion rate of the raw material is:

[0050] ,in, For the first The raw materials at the pixel position Sampling time Surface moisture diffusion rate, Pixel-level moisture index, For the first The critical water loss coefficient of the raw material. For the first The characteristic length of moisture replenishment of the raw materials. To prevent division by zero constant, The spatial gradient magnitude of the moisture index. The local temporal change rate of the pixel-by-pixel moisture index measures the rate at which moisture evaporates into the air at that pixel location, and is the original driving force of charring risk. It is used as the basic factor in the formula because there is no possibility of charring without moisture loss. However, if charring is judged solely by this single factor, it implicitly assumes that all moisture changes come from evaporation, ignoring the process of moisture inside the raw material diffusing to the surface through capillary action, and cannot distinguish between a risk-free area with high evaporation but equally efficient diffusion replenishment and a high-risk area with slow evaporation but almost stagnant diffusion replenishment. The critical water loss coefficient characterizes the threshold at which the surface of the raw material changes from safe drying to dangerous drying. It is placed in the denominator of the drying acceleration factor because the charring risk does not increase linearly with the decrease in moisture, but accelerates sharply after the moisture level drops to a certain critical level. When the surface moisture is sufficient, this factor is close to 1 and has almost no effect; when the moisture approaches the critical value, this factor amplifies sharply, transforming the same evaporation rate into a charring risk rate several times higher than before, making the system highly sensitive to small moisture changes at the end of baking, and solving the problem of insufficient sensitivity of the fixed threshold method to the critical zone. The feature length of moisture replenishment measures the effective distance that the capillary network inside the raw material transports moisture to the surface, depending on the porosity, pore size distribution, and capillary connectivity of the raw material. Its product with the spatial gradient amplitude is placed in an exponential decay factor because the gradient represents the diffusion driving force, while the feature length represents the effective diffusion range; their product together determines the intensity of internal replenishment. Different raw materials have different feature lengths due to their microstructure differences. Porous and loose raw materials have a larger replenishment distance and stronger replenishment capacity, while dense and thick raw materials have a smaller replenishment distance and weaker replenishment capacity. This parameter allows the formula to adapt to the internal moisture transport characteristics of the eight raw materials. The spatial gradient amplitude of the moisture index measures the difference in moisture concentration between the pixel and its surrounding neighborhood. The greater the concentration difference, the stronger the capillary driving force and the more timely the internal moisture replenishment to that location. Placing it in the numerator of an exponential function ensures that a larger gradient results in stronger exponential decay and greater suppression of charring risk, while a smaller gradient results in weaker decay and complete conversion of evaporation rate into charring risk. This structure provides a physical basis for the localization of coking risk, explaining why coking often occurs first in the flat central region of the raw material rather than at the edges. While evaporation is rapid at the edges, the strong concentration gradient in adjacent regions drives strong diffusion replenishment, suppressing net risk. In the central region, the gradient approaches zero, diffusion replenishment almost stagnates, and even slow evaporation is fully converted into risk. The current moisture index of this pixel is in the denominator of the exponential function, forming a ratio with the spatial gradient magnitude. This ratio measures the strength of the diffusion driving force relative to the current remaining moisture: the same gradient is relatively weak when moisture is abundant and relatively strong when moisture is scarce. The drier the surface and the more water source in the neighborhood, the more effective the replenishment and the stronger the risk suppression. When the surface is dry and the neighborhood is also dry, the gradient still exists, but there is no water source for replenishment, and suppression naturally fails.This method solves the fundamental flaws of traditional overall moisture content threshold methods, such as coarse spatial positioning and inability to predict local coking risks.

[0051] The thermal inertia parameters of eight raw materials, namely the product of density, specific heat capacity, and characteristic thickness, are pre-determined. At each sampling time, the degree of deviation of the thermal inertia parameters of the eight raw materials is calculated, and the maximum safe temperature change rate is dynamically determined accordingly. The greater the deviation, the smaller the allowable temperature change rate, to prevent overheating of raw materials with low thermal inertia and lag in response of raw materials with high thermal inertia. The temperature intervention amount is constrained by this maximum safe temperature change rate. The formula for calculating the maximum safe temperature change rate is:

[0052] ,in, This represents the maximum safe rate of temperature change allowed at the current sampling time. Based on the allowable temperature change rate, This is the harmonic average of the thermal inertia parameters of the eight raw materials. This is the arithmetic mean of the thermal inertia parameters of the eight raw materials. This refers to the actual temperature inside the baking apparatus. This refers to the maximum permissible operating temperature of the baking equipment. This is the temperature risk acceleration index. The basic allowable temperature change rate is the safe temperature change rate when the thermal inertia of the eight raw materials is exactly the same. It is determined by the safe heating rate when baking the raw material with the lowest thermal inertia alone, and serves as the benchmark value for the entire formula. All subsequent discounts and attenuations are based on this benchmark. The temperature risk attenuation factor is based on the power of the ratio of the current actual temperature to the maximum working temperature. The exponent is greater than 1. When the actual temperature is low, the power of the temperature ratio approaches zero, and the value within the square brackets approaches 1. Temperature has almost no additional constraint on safety. As the actual temperature gradually approaches the maximum working temperature, the power of the temperature ratio approaches 1, the value within the square brackets approaches the uniformity discount factor itself, and the constraint tightens to the most stringent level after being multiplied with the first-level factor. The exponent being greater than 1 makes the risk increase acceleration greater in the high-temperature range than in the low-temperature range, which is consistent with the empirical rule in baking practice that the over-baking risk increase of thin raw materials in the high-temperature range is much greater than that in the low-temperature range. This solves the fundamental flaw of traditional control methods that do not consider the differences in thermal inertia of multiple raw materials, which leads to overheating of thin sheets with low thermal inertia and lag in response of thick blocks with high thermal inertia when adjusting the temperature. It quantifies the degree of thermal inertia deviation into a quantitative constraint on the rate of temperature change.

[0053] The collective baking quality function is defined as the negative Mahalanobis distance between the current normalized maturity index of each raw material and its target maturity value. At each sampling time, the current collective quality (which should be terminated immediately) is compared with the expected collective quality. When the increase in collective quality due to continued baking is negative and this holds true for multiple consecutive sampling times, the optimal termination time for the collective is determined to have been reached, and a baking termination command is output. The formula for calculating the current collective quality in S8 is:

[0054] ,

[0055] , ,in, Sampling time The overall baking quality is represented by a negative value; the closer it is to zero, the closer the overall baking state of the eight ingredients is to the ideal target. This represents the normalized maturity vector of the eight raw materials at the current sampling time. Let each of the eight raw materials be a target maturity vector. This is the matrix transpose operator. For the correlation structure matrix, This is the arithmetic mean of the thermal inertia parameters of the eight raw materials. For the first Thermal inertia parameters of the raw materials, Location of the raw materials for evaporation source Location of the disturbed raw materials The resulting equivalent temperature decay, Location of the raw materials for evaporation source Location of the disturbed raw materials The resulting equivalent temperature decay, The current temperature. diagonal elements It is a non-diagonal element.

[0056] The current collective quality calculation aggregates the maturity deviations of the eight raw materials into a single scalar and assigns a negative value. The closer this scalar is to zero, the closer the overall state is to the ideal target. The reason for taking a negative value is that the original squared Mahalanobis distance is non-negative, and the larger the value, the greater the deviation. After taking a negative value, zero corresponds to the optimal state, and a negative value corresponds to a deduction, making the semantics of the quality function intuitive and consistent. The normalized maturity vector of the eight raw materials treats the complete snapshot of the maturity index of each raw material at the sampling time as a whole vector, rather than treating it as eight independent scalars. This is because the positive or negative sign of each component directly indicates whether the raw material is over-roasted or under-roasted. At the same time, the vector form allows subsequent distance calculations to simultaneously consider the magnitude of the deviation of each component and the correlation structure between the components, which cannot be achieved by scalar accumulation. The target maturity vector includes the target values ​​set by each raw material according to its flavor contribution weight, color contribution weight, and the heat sensitivity of the medicinal components in the finished product of Bazhen powder. If the target values ​​of all raw materials are unified into the same value, the reality of the differentiated quality requirements of each raw material will be ignored, and the true quality constraints of each raw material cannot be accurately reflected in the quality function. The matrix transpose operator converts column vectors into row vectors to ensure dimensional compatibility of multiplication, thus aggregating scattered multidimensional bias information into a unified decision-making basis. By simultaneously encoding the moisture and heat inter-component relationships between raw materials and the thermal inertia and vulnerability of each raw material into the correlation structure matrix, and aggregating multidimensional biases into a single quality score using Mahalanobis distance, the collective quality assessment is upgraded from the mechanical accumulation of eight independent biases to a unified metric that considers the physical correlation structure, enabling termination decisions to perceive the true coupling relationship between maturity biases.

[0057] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments for application in other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An image processing-based control method for the roasting process of eight-treasure powder raw materials, characterized by, Includes the following steps: S1. A visible light camera, a thermal infrared camera, and a multispectral camera, which have been jointly calibrated, are arranged above the baking device to synchronously acquire multi-source image sequences of eight raw materials during the baking process at an adaptive sampling interval. S2. Spatial registration is performed on the visible light image, thermal infrared image and multispectral image acquired synchronously at each sampling time. Using the visible light image as the reference coordinate system, the thermal infrared image and multispectral image are mapped to a unified coordinate space, and the registered images are preprocessed for denoising. S3. Extract the spatial temperature distribution above each raw material tray from the thermal infrared image at the current sampling time, extract the spatial moisture concentration distribution above each raw material tray from the ratio of reflectance of the moisture-sensitive band to the reflectance of the reference band in the multispectral image, jointly encode the spatial temperature distribution and spatial moisture concentration distribution according to the relative spatial position of each raw material tray, construct a humid heat interference field, and calculate the equivalent temperature attenuation between any two raw material positions in real time. S4. Define a normalized maturity index for each raw material. The normalized maturity index uses the water molecule index corresponding to the rate of change of the multispectral moisture estimate as a pre-multiplicative factor, and the color sub-index corresponding to the rate of change of the L value of the brightness channel in the color Lab space, the volume sub-index corresponding to the rate of change of the raw material outline area in the visible light image, and the texture sub-index corresponding to the rate of change of the surface texture roughness are weighted and summed in square brackets. The volume weight, color weight, and texture weight are determined by the dynamic weight coefficients generated by the water molecule index through the quadratic Bernstein basis function. S5. At each sampling moment, the equivalent temperature decay amount described in S3 is used as input to calculate the actual effective baking temperature of each raw material, and the adjustment amount of baking temperature and circulating wind speed is dynamically generated so that the ripening progress of each of the eight raw materials tends to be synchronized and approaches their respective target ripening values ​​at similar times. S6. Calculate the surface moisture diffusion rate of the raw material pixel by pixel from the multispectral images of adjacent sampling times. When the surface moisture diffusion rate of any local area exceeds the preset threshold and the local temperature of the area in the thermal infrared image is higher than the average temperature of the raw material by a preset temperature difference, it is determined that there is a risk of coking and local intervention is triggered. S7. Pre-determine the thermal inertia parameters of each of the eight raw materials, namely the product of density, specific heat capacity and characteristic thickness. At each sampling time, calculate the degree of deviation of the thermal inertia parameters of the eight raw materials. Based on this, dynamically determine the maximum safe temperature change rate. The larger the deviation, the smaller the allowable temperature change rate, in order to prevent the low thermal inertia raw materials from overheating and the high thermal inertia raw materials from responding with lag. The temperature intervention amount triggered by S5 and S6 is constrained by the maximum safe temperature change rate. S8. Define the collective baking quality function as the negative value of the Mahalanobis distance between the current normalized maturity index of each raw material and its target maturity value. At each sampling time, compare the current collective quality that is about to be terminated with the expected collective quality. When the increment of the collective quality due to continued baking is negative and this holds true for multiple consecutive sampling times, it is determined that the optimal termination time for the collective has been reached, and a baking termination command is output.

2. The method for controlling the baking process of Bazhen powder raw materials based on image processing according to claim 1, characterized in that, In S1, the eight ingredients are placed in their respective independent baking trays, and the trays are arranged in a preset spatial layout within the baking device.

3. The method for controlling the baking process of Bazhen powder raw materials based on image processing according to claim 1, characterized in that, The formula for calculating the equivalent temperature decay in S3 is as follows: ,in, Location of the raw materials for evaporation source Location of the disturbed raw materials The resulting equivalent temperature decay, The evaporative cooling intensity coefficient, This is an estimated current moisture content of the raw material used in the evaporation source. For position With position Spatial distance between them The attenuation characteristic length is related to the thermal diffusion properties of air. This refers to the airflow speed within the baking apparatus. It is the angle between the direction of the circulating airflow and the line connecting the positions of the two raw materials.

4. The method for controlling the baking process of Bazhen powder raw materials based on image processing according to claim 1, characterized in that, The formula for calculating the normalized maturity index in S4 is as follows: ， ， ， ， ， , , ,in, For the first At the sampling time of the raw materials The normalized maturity index, For water molecule indicators, For color weights, For color sub-indices, For volume weights, For volume index, For texture weights, For texture sub-indices, The mean value of L at the initial time. For the first raw materials at all times The mean L value of the Lab space brightness channel. The mean of the target L value, This is the initial water content. For the first raw materials at all times The estimated water content, For the target moisture content, The target contour area, This represents the initial contour area. For the first raw materials at all times The visible light image outline area, For the first The surface fractal dimension of the raw material at time t. For the target fractal dimension, Let be the initial fractal dimension.

5. The method for controlling the baking process of Bazhen powder raw materials based on image processing according to claim 1, characterized in that, The formula for calculating the effective baking temperature in S5 is as follows: ,in, For the first The actual effective baking temperature experienced by the raw materials Set the temperature for the current baking setup. The saturated cooling coefficient, Current temperature The saturated water content of the air below. This represents the average moisture content within the current baking space. Location of the raw materials for evaporation source Location of the disturbed raw materials The resulting equivalent temperature decay.

6. The method for controlling the baking process of Bazhen powder raw materials based on image processing according to claim 1, characterized in that, The formula for calculating the surface moisture diffusion rate of the raw material in S6 is as follows: ,in, For the first The raw materials at the pixel position Sampling time Surface moisture diffusion rate, Pixel-level moisture index, For the first The critical water loss coefficient of the raw material. For the first The characteristic length of moisture replenishment of the raw materials. To prevent division by zero constant, This represents the spatial gradient magnitude of the moisture index.

7. The method for controlling the baking process of Bazhen powder raw materials based on image processing according to claim 1, characterized in that, The formula for calculating the maximum safe temperature change rate in S7 is as follows: ,in, This represents the maximum safe rate of temperature change allowed at the current sampling time. Based on the allowable temperature change rate, The harmonic average of the thermal inertia parameters of the eight raw materials is given. This is the arithmetic mean of the thermal inertia parameters of the eight raw materials. This refers to the actual temperature inside the baking apparatus. This refers to the maximum permissible operating temperature of the baking equipment. This is the temperature risk acceleration index.

8. The method for controlling the baking process of Bazhen powder raw materials based on image processing according to claim 1, characterized in that, The formula for calculating the current collective quality in S8 is as follows: ， , ,in, Sampling time The overall baking quality is represented by a negative value; the closer it is to zero, the closer the overall baking state of the eight ingredients is to the ideal target. This represents the normalized maturity vector of the eight raw materials at the current sampling time. Let each of the eight raw materials be a target maturity vector. This is the matrix transpose operator. For the correlation structure matrix, This is the arithmetic mean of the thermal inertia parameters of the eight raw materials. For the first Thermal inertia parameters of the raw materials Location of the raw materials for evaporation source Location of the disturbed raw materials The resulting equivalent temperature decay, Location of the raw materials for evaporation source Location of the disturbed raw materials The resulting equivalent temperature decay, Set the temperature for the current baking setup. diagonal elements It is a non-diagonal element.

Images