Method for predicting sorting of fermented sponge spicules through machine learning

Abstract

The invention provides a method for predicting sorting of fermented sponge spicules through machine learning, which comprises the following steps of: analyzing the interaction strength of spicule structures and hypha attachment according to a deformation aggravation degree to obtain a potential rising level of a sorting misjudgment rate; if the potential rising level of the sorting misjudgment rate exceeds a preset rising level standard, the airflow intensity and the screen mesh aperture are obtained from historical data, the airflow intensity and the screen mesh aperture are adjusted according to the adhesion stacking risk, and a dynamic adjustment instruction set is obtained; the airflow intensity and the screen mesh aperture of the sorting equipment are adjusted by executing the adjusting instruction set, the state of the single-layer distribution bone needles adjusted by the sorting equipment is obtained, and the single-layer distribution uniformity of the bone needles is recognized; the final sorting parameters are applied to a sponge spicule sorting system to sort sponge spicules, single-layer distributed sponge spicules are obtained, and the stability of the sponge spicule sorting process is evaluated by analyzing the uniformity and the sorting misjudgment rate of the single-layer distributed spicules.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information technology, and in particular to a machine learning prediction method for sorting fermented sponge spicules. BACKGROUND

[0002] The sorting technology of sponge spicules is a core link in the field of marine biological material processing, and is widely used in the production of medical, cosmetic and high-value-added biological products. The sorting precision directly determines the product quality and market competitiveness. As a kind of natural fiber with small size and variable shape, the sorting process of sponge spicules needs to ensure that each spicule can be accurately separated and classified. However, due to the complex physical properties of spicules, the optimization of sorting technology not only concerns the production efficiency, but also has a decisive influence on the stability of product quality, and is a key technical field to promote the upgrading of related industries. The adhesion and flexibility of the spicule surface will change significantly during the fermentation process, and the existing equipment is difficult to adapt to these changes, resulting in a decrease in the uniformity of the single-layer distribution of spicules. The adhesion determines whether the spicules will be adhered to each other during the sorting process, and the flexibility affects the degree of deformation of the spicules under the action of air flow or screen. The increase of adhesion is due to the chemical action of temperature fluctuation in the fermentation process, which makes the spicule surface generate sticky substances, increasing the possibility of spicule stacking. The stacked spicules will deform during sorting due to the increase of flexibility, further interfering with the separation effect of air flow or screen. For example, in actual sorting, spicules may form clumps due to excessive adhesion, block the screen aperture, or bend due to excessive flexibility, resulting in failure to pass through the preset aperture range, and thus causing sorting errors. The increase of adhesion will result in the failure of single-layer distribution of spicules during sorting, and the stacking phenomenon makes the air flow or screen unable to effectively act on each spicule. The change of flexibility further magnifies this problem, and the deformed spicules may produce unpredictable motion trajectories in the air flow, or be misclassified on the screen due to deformation. For example, in a certain sorting scenario, the spicules form multi-layer stacking due to excessive adhesion, resulting in the misclassification of some spicules as defective products during air flow sorting, seriously affecting the precision and consistency of sorting. Therefore, how to real-time respond to the interactive influence of the adhesion and flexibility of the spicule surface in the dynamic changing fermentation environment, and ensure that the spicules can be accurately separated with single-layer distribution, has become a key problem in the sorting technology of sponge spicules. SUMMARY

[0003] The present application provides a machine learning prediction method for sorting fermented sponge spicules, mainly comprising: The fermentation temperature and sponge spicule surface images in the fermentation environment are continuously collected, the hypha density and morphology in the sponge spicule surface images are identified, and the adhesion force change trend and the flexibility change trend of the sponge spicule under different fermentation temperatures are generated; according to the adhesion force change trend and the flexibility change trend, the hypha adhesion degree change amplitude is identified, the adhesion stacking risk is determined according to the change amplitude, and the deformation aggravation degree is obtained through surface tension analysis; according to the deformation aggravation degree, the interaction strength of the spicule structure and the hypha adhesion is analyzed, and the potential increase level of the sorting error rate is obtained; the dynamic adjustment instruction set is generated according to the potential increase level of the sorting error rate; the air flow intensity and the screen mesh size of the sorting equipment are adjusted by executing the adjustment instruction set, the adjusted single-layer distributed spicule state is obtained, and the distribution uniformity of the single-layer distributed spicule is identified; according to the comparison between the distribution uniformity and the target distribution uniformity, the distribution uniformity deviation level is obtained, and the sorting parameters are determined according to the deviation level, the adhesion stacking risk and the deformation aggravation degree; the sponge spicule is sorted through the sorting parameters, and the single-layer distributed sponge spicule is obtained, and the uniformity and the sorting error rate of the single-layer distributed sponge spicule are analyzed to obtain the sorting process stability index.

[0004] Further, the continuous collection of the fermentation temperature and the sponge spicule surface image in the fermentation environment, the identification of the hypha density and morphology in the sponge spicule surface image, and the generation of the adhesion force change trend and the flexibility change trend of the sponge spicule under different fermentation temperatures, include: The temperature data at different positions in the fermentation tank is collected, the sponge spicule surface image is obtained, the fermentation area where the spicule is located is determined according to the temperature gradient distribution, the mean filtering and histogram equalization processing of the surface image is performed, and the spicule surface texture image is obtained; the gray processing of the spicule surface texture image is performed, the hypha contour is identified by using the edge detection algorithm, the hypha density value is determined by calculating the number of hypha interlacing points in unit area, and the hypha structure characteristics are extracted according to the curvature radius and branch angle of the hypha morphology; the adhesion coefficient is obtained by calculating the ratio of the hypha coverage area to the spicule surface area according to the hypha density value, and the flexibility modulus is calculated according to the deviation angle of the spicule bending degree in the spicule surface texture image and the standard straight line; the adhesion coefficient and the flexibility modulus are arranged in time sequence, the fixed time window is segmented, and the curve fitting method is used to determine the adhesion force change trend and the flexibility change trend under different temperature gradients.

[0005] Further, the adhesion force change trend and the flexibility change trend are used to identify the hypha adhesion degree change amplitude, the adhesion stacking risk is determined according to the change amplitude, and the deformation aggravation degree is obtained through surface tension analysis, including: Respectively extract the numerical difference of adjacent time in the adhesion force change trend and the flexibility change trend, calculate the standard deviation of the data sequence as the fluctuation amplitude, construct the time series feature vector, process the time series feature vector through the classification algorithm, output the mycelium adhesion degree level value, calculate the difference value of adjacent time point level value, and obtain the mycelium adhesion degree change amplitude; According to the correlation mapping of the mycelium adhesion degree change amplitude and the adjacent bone needle spacing, the adsorption force between the adjacent bone needles is calculated, and the adhesion stacking risk level is determined by comparing the product of the adsorption force and the bone needle mass and the gravitational acceleration; Obtain the mycelium network density data, calculate the surface tension coefficient according to the liquid drop contact angle measurement method, and obtain the mycelium constraint strength value through the product of the surface tension coefficient and the mycelium network density; According to the distribution of the mycelium constraint strength value in the bone needle length direction, the difference value of the constraint strength of the adjacent positions is calculated, and the tension gradient is obtained. Multiply the tension gradient by the original form curvature of the bone needle to obtain the deformation stress of each part, and obtain the deformation aggravation degree after weighted summation.

[0006] Further, according to the deformation aggravation degree, the interaction strength of the bone needle structure and the mycelium adhesion is analyzed to obtain a potential increase level of sorting misjudgment rate, including: Through the deformation aggravation degree, the elastic modulus and the cross-sectional moment of inertia of the bone needle material are combined to calculate the maximum deflection value of the bone needle; The coverage ratio of the mycelium in each region on the surface of the bone needle is counted to obtain the mycelium adhesion density distribution, and the average value of the mycelium adhesion density distribution is multiplied by the deformation aggravation degree to obtain the interaction strength.

[0007] Further, according to the potential increase level of sorting misjudgment rate, a dynamic adjustment instruction set is generated, including: According to the ratio of the adhesion stacking risk level to the preset level, an adjustment coefficient is determined, the air flow intensity and the screen mesh size value are adjusted through the adjustment coefficient, and a dynamic adjustment instruction set containing a control code, a device address identifier and a time stamp is generated.

[0008] Further, the air flow intensity and the screen mesh of the sorting device are adjusted by executing the adjustment instruction set to obtain the state of the single-layer distributed bone needles, and the distribution uniformity of the single-layer distributed bone needles is identified, including: The control code in the adjustment instruction set is sent to adjust the fan speed and the screen opening degree, the sorting area image is shot to obtain the state of the single-layer distributed bone needles, the image of the single-layer distributed bone needles is divided into grids, the number of bone needles in each grid is counted, the ratio of the number standard deviation to the average value is calculated, and the distribution uniformity is obtained.

[0009] Further, according to the comparison of the distribution uniformity and the target distribution uniformity, the distribution uniformity deviation level is obtained, and the sorting parameters are determined according to the deviation level, the adhesion stacking risk and the deformation aggravation degree, including: The difference between the distribution uniformity and the target distribution uniformity is calculated to obtain a deviation level, and the airflow intensity, the screen mesh size and the sorting speed are determined in combination with the risk value of adhesion stacking and the value of deformation aggravation degree, to form a sorting parameter set.

[0010] Further, the single-layer distributed sponge bone needles are obtained by sorting the sponge bone needles through the sorting parameters, and the uniformity and the sorting misjudgment rate of the single-layer distributed sponge bone needles are analyzed to obtain a sorting process stability index, including: The sorting parameter set is transmitted to a sorting equipment controller to run according to the set airflow intensity, screen mesh size and sorting speed, and the single-layer distributed sponge bone needles are obtained by collecting images of a sorting area; the images of the single-layer distributed sponge bone needles are grid-divided, the ratio of the standard deviation of the number of bone needles in each grid to the average value is calculated as the uniformity index, and the proportion of misclassified bone needles is calculated as the sorting misjudgment rate; and the sorting process stability index is obtained according to the weighted average of the uniformity index and the sorting misjudgment rate.

[0011] The technical scheme provided by the embodiment of the present application can include the following beneficial effects: The present application discloses a machine learning prediction method for sorting fermented sponge bone needles, aiming at the problem of increased bone needle sorting misjudgment rate caused by mycelium adhesion during fermentation. The fermentation temperature and bone needle surface images are continuously collected, the mycelium density and morphology are analyzed, the adhesion force and flexibility change trend are extracted, the mycelium adhesion degree and adhesion stacking risk are evaluated in combination with a machine learning algorithm, and then the deformation aggravation degree is determined through surface tension analysis, the interaction strength between bone needle structure and mycelium attachment is quantified, and the sorting misjudgment rate is predicted. When the misjudgment rate exceeds the standard, the present application obtains the airflow intensity and screen mesh size from historical data, dynamically adjusts the sorting equipment parameters, optimizes the uniformity of single-layer bone needle distribution, and iteratively calculates the final sorting parameters, including airflow intensity, screen mesh size and sorting speed, which are applied to a sorting system to realize high-stability single-layer bone needle distribution. The present application effectively reduces the sorting misjudgment rate by integrating environmental monitoring, image analysis and machine learning optimization, improves the stability and uniformity of the sponge bone needle sorting process, and significantly improves the production efficiency and product quality. BRIEF DESCRIPTION OF DRAWINGS

[0012] Fig. 1 It is a flowchart of the machine learning prediction method for sorting fermented sponge bone needles.

[0013] Fig. 2 It is a schematic diagram of the machine learning prediction method for sorting fermented sponge bone needles. DETAILED DESCRIPTION

[0014] In order for those skilled in the art to better understand the technical solutions in the specification, the technical solutions in the specification will be clearly and completely described below in conjunction with the drawings in the embodiments of the specification. Obviously, the described embodiments are only a part of the embodiments of the specification, not all the embodiments. Based on the embodiments in the specification, all other embodiments obtained by those of ordinary skill in the art without creative labor should belong to the protection scope of the specification.

[0015] As Figs. 1-2 The machine learning prediction method for sorting fermented sponge spicules in the embodiment can specifically include the following steps. In step S101, the fermentation temperature in the fermentation environment and the spicule surface image are continuously collected, the hypha density and morphology in the spicule surface image are identified, the adhesion and flexibility are extracted according to the hypha density and morphology in the image, and the adhesion change trend and flexibility change trend of the spicule under different fermentation temperatures are obtained.

[0016] In the fermentation environment, temperature sensors are uniformly arranged along the wall of the fermentation tank, temperature data at different positions in the fermentation tank are collected at a preset time interval, at the same time, a sponge spicule surface image is obtained through an industrial camera, a fermentation area where the spicule is located is determined according to the temperature gradient distribution, image noise is removed through mean filtering and histogram equalization processing, and a spicule surface texture image is obtained. The spicule surface texture image is subjected to grayscale processing, a Canny edge detection algorithm is used to identify the hypha profile, the number of hypha interlacing points in a unit area is calculated to determine the hypha density value, the hypha structure features are extracted according to the curvature radius and branch angle of the hypha morphology, the hypha growth feature data are obtained through correlation analysis of the hypha density value and the hypha structure features and the fermentation time length. According to the hypha density value in the hypha growth feature data, the adhesion coefficient is obtained by calculating the ratio of the hypha coverage area to the spicule surface area, the flexibility modulus is calculated by using the deviation angle of the bending degree of the spicule in the image and the standard straight line, if the environmental humidity exceeds the preset humidity threshold, the adhesion coefficient is corrected according to the linear relationship between the humidity and the adhesion coefficient, and the actual adhesion force value is obtained. The actual adhesion force value and the flexibility modulus are arranged in time sequence, the data are segmented by setting a fixed time window, the numerical change law under different temperature gradients is analyzed in each window, the least square method is used to curve fit the corresponding relationship between the temperature and the adhesion force value and the temperature and the flexibility modulus, and the adhesion change trend and the flexibility change trend of the sponge spicule in the fermentation process are determined.

[0017] Specifically, in an embodiment, the temperature monitoring of the fermentation environment is realized by the temperature sensors arranged at equal intervals on the inner wall of the fermentation tank.

[0018] Specifically, one temperature sensor is installed every 60 degrees along the circumferential direction of the fermenter, and three layers of sensors are arranged according to one-third of the height of the tank body in the vertical direction, forming a three-dimensional monitoring network. The temperature sensor uses a PT100 platinum resistance thermometer, and the measurement accuracy reaches ±0.1℃. The industrial camera selects a CMOS camera with a resolution of 2048×1536 pixels, equipped with a macro lens, and is installed in the direct position of the observation window of the fermenter, ensuring that the mycelium surface can be clearly captured.

[0019] It should be noted that the image preprocessing stage adopts a method combining mean filtering and histogram equalization. Mean filtering replaces the center pixel value by averaging the pixel values in the 3×3 neighborhood around each pixel point, effectively eliminating random noise in the image. Histogram equalization adjusts the gray scale distribution of the image, redistributes the pixel values originally concentrated in a certain gray scale interval to the entire gray scale range, enhances the contrast of the image, and makes the mycelium texture more clear and identifiable.

[0020] Preferably, the Canny edge detection algorithm performs in five steps when identifying the mycelium contour. First, apply a Gaussian filter to the gray image for smoothing processing, set the filter size to 5×5, and the standard deviation σ to 1.4 to eliminate high-frequency noise in the image. Second, calculate the image gradient, use the Sobel operator to calculate the gradient components in the horizontal and vertical directions respectively, and get the gradient amplitude and direction of each pixel point. In the third step, non-maximum suppression is performed, compare the gradient amplitudes of the current pixel and adjacent pixels along the gradient direction, and retain the local maximum value point to thin the edge width. In the fourth step, double threshold detection is applied, set the high threshold and the low threshold, the high threshold is usually 2 to 3 times the low threshold, the pixel points with gradient amplitude higher than the high threshold are determined as strong edges, and the weak edges between the two thresholds are retained. By edge connection, the weak edges connected with the strong edges are retained, and the isolated weak edges are removed to obtain the complete mycelium contour. In the fermentation process of the sponge bone needle, the mycelium will form a network structure on the surface of the bone needle, and the mycelium density is quantified by calculating the number of intersection points of the mycelium in unit area. The intersection point is defined as the position where three or more mycelia intersect.

[0021] Illustratively, the extraction of mycelium structure features focuses on two parameters: curvature radius and branch angle. The curvature radius is obtained by calculating the second derivative after polynomial fitting of the mycelium contour, reflecting the bending degree of the mycelium. The branch angle is obtained by identifying the bifurcation point of the mycelium and calculating the included angle between the two sub-mycelia and the main mycelium at the bifurcation. When the fermentation temperature rises, the mycelium grows faster, the branch angle tends to be acute, the curvature radius decreases, and the mycelium presents a more intensive growth trend.

[0022] In an embodiment, the calculation of the adhesion coefficient is based on the ratio of the hyphae coverage area to the surface area of the spicule. The hyphae coverage area is identified by image segmentation techniques, and the number of pixels is calculated, and then converted to the actual area according to the camera calibration parameters. The surface area of the spicule is calculated by measuring the length and diameter of the spicule in advance according to the lateral area formula of a cylinder. The adhesion coefficient K = hyphae coverage area / spicule surface area × correction factor, and the correction factor is dynamically adjusted according to the thickness of the hyphae, ranging from 0.8 to 1.2.

[0023] It can be understood that the determination of the flexibility modulus is achieved by analyzing the deformation degree of the spicule in the image. The central axis of the spicule is selected as the reference, and the maximum deviation distance between the actual spicule profile and the ideal straight line is measured, and the ratio of the deviation distance to the length of the spicule is defined as the deformation coefficient. The flexibility modulus E is inversely proportional to the deformation coefficient, and the greater the deformation coefficient, the easier the spicule bends, and the smaller the flexibility modulus.

[0024] Specifically, the humidity correction uses a linear compensation model. When the environmental humidity exceeds a preset threshold, the adhesion coefficient needs to be corrected. The correction formula is Kcorrected = Koriginal × (1 + α × (H - H0)), where Koriginal is the uncorrected adhesion coefficient, H is the actual humidity value, H0 is the humidity threshold, and α is the humidity influence coefficient, which is determined by experimental calibration, and is usually valued between 0.01 and 0.03. This correction method takes into account the effect of increased adhesion caused by increased moisture on the surface of the hyphae in a high humidity environment.

[0025] For example, the time window analysis uses a sliding processing method with a fixed window width. The window width is set to 30 minutes of data collection period, and the window slides forward with a step of 5 minutes. In each window, the average temperature, adhesion force value and flexibility modulus in that time period are extracted to form a set of data points. These data points are curve fitted by the least squares method, and the fitting function is a quadratic polynomial y = ax2 + bx + c, where x represents the temperature, and y represents the adhesion force or flexibility modulus, respectively. The coefficients a, b, c obtained by fitting reflect the quantitative relationship between temperature and these two parameters, and the change in the slope of the curve reveals the change trend of the adhesion force and flexibility.

[0026] Step S102, according to the change trend of the adhesion force and the change trend of the flexibility, the change amplitude of the hyphae adhesion degree is identified by a machine learning algorithm, and the adhesion and stacking risk is determined according to the change amplitude of the hyphae adhesion degree, and the deformation aggravation degree is obtained by surface tension analysis from the adhesion and stacking risk.

[0027] According to the adhesion force change trend and the flexibility change trend, the numerical difference of adjacent time points is extracted as the change rate, the standard deviation of the data sequence is calculated as the fluctuation amplitude, the main frequency is identified as the periodic characteristic through Fourier transform, the time series feature vector is constructed, the random forest algorithm is used for classification of the feature vector, and the level value of the mycelium adhesion degree is output. The difference between the level values of adjacent time points is calculated to obtain the change amplitude of the mycelium adhesion degree. The change amplitude of the mycelium adhesion degree is associated with the adjacent spicule spacing obtained by image measurement, and when the change amplitude exceeds a preset amplitude threshold, the adhesion force between the spicules is calculated according to the product of the adhesion strength and the spicule contact area. The adhesion force is compared with the product of the spicule mass and the acceleration of gravity, and if the adhesion force is greater than the gravity value, it is determined that there is a risk of adhesion stacking, and the adhesion stacking risk level is determined according to the ratio of the adhesion force to the gravity value. Based on the adhesion stacking risk level, the number of stacked spicules is counted by image recognition, the mycelium network density data is obtained, the contact angle θ between the spicule surface and the mycelium interface is obtained by the droplet contact angle measurement method, and the surface tension coefficient is calculated according to the surface tension γ = σcosθ, wherein σ is the liquid surface tension constant. The product of the surface tension coefficient and the mycelium network density obtains the mycelium constraint strength value. According to the distribution of the mycelium constraint strength value in the length direction of the spicule, the difference of the constraint strength of adjacent positions is calculated to obtain the tension gradient, and the tension gradient is multiplied by the curvature of the original shape of the spicule to obtain the deformation stress of each part. The deformation stresses of all parts are weighted and summed, and the weight value is proportional to the distance from the part to the center of gravity of the spicule. The accumulated deformation aggravation degree is obtained.

[0028] Specifically, in an embodiment, the construction of the time series feature vector is based on multi-dimensional analysis of continuous monitoring data of adhesion force and flexibility. The change rate is obtained by calculating the numerical difference of adjacent sampling time points, specifically, the current time value is subtracted from the previous time value, and then divided by the time interval. The fluctuation amplitude is represented by the standard deviation of the data in the sliding window, and the window width is set to 10 sampling points, sliding one sampling point at a time, which reflects the dispersion degree of the data in real time.

[0029] Specifically, Fourier transform is used to identify periodic features. The original time series is first detrended to eliminate the influence of long-term trends. The time-domain signal is converted to the frequency domain by the fast Fourier transform algorithm, and the frequency component with the largest amplitude in the frequency spectrum is identified as the dominant frequency. When the temperature fluctuates periodically during fermentation, the adhesion and flexibility of the hyphae also exhibit corresponding periodic changes, and the dominant frequency reflects the strength of this periodicity. The rate of change, fluctuation amplitude, and dominant frequency are combined into a feature vector, which is input into a random forest classifier. The random forest contains 100 decision trees, each trained based on different feature subsets and sample subsets, and the final mycelial adhesion level is determined by a voting mechanism. The mycelial adhesion level is divided into five levels, from complete separation to severe adhesion, and each level corresponds to a specific numerical interval.

[0030] It should be noted that the measurement of the distance between the bone needles is achieved through image processing. In the field of view of the microscope, the edge profiles of adjacent bone needles are identified, the pixel distance between the closest points of the two profiles is calculated, and the actual physical distance is converted according to the microscope calibration parameters.

[0031] Preferably, the calculation of the adsorption force takes into account the bridging effect of the mycelial network. When there is a hyphal connection between two bone needles, the tensile force generated by the hyphae constitutes the main adsorption force. The adhesion strength is determined by a nanoindenter, which contacts the hyphal surface at a constant rate and then pulls away, recording the force-displacement curve during the pulling process. The peak value of the curve is the adhesion strength. The contact area of the bone needle is calculated based on the overlapping part of the hyphal coverage area on the bone needle surface and the adjacent bone needle.

[0032] Illustratively, the implementation process of the droplet contact angle measurement method includes precise droplet release and image acquisition steps. A microsyringe is used to release a 2-microliter deionized water droplet on the surface of the bone needle, and a high-speed camera is used to take a side view of the droplet profile at a rate of 1000 frames per second, capturing the morphological changes during the contact of the droplet. The droplet profile is identified by image processing software, and the baseline and tangent of the droplet are fitted, and the angle between the two is the contact angle θ. The surface tension coefficient γ is calculated according to Young's equation, γ = σcosθ, where σ is the surface tension constant of water at room temperature, about 72.8 mN / m. When the hyphal density increases, the hydrophilicity of the bone needle surface increases, the contact angle decreases, and the surface tension coefficient increases accordingly. The hyphal network density is quantified by the number of interlacing points per unit area, and the higher the density, the stronger the constraint on the bone needle.

[0033] In one possible implementation, the distribution of the mycelial constraint strength value along the length direction of the bone needle exhibits uneven characteristics. The root of the bone needle has a high constraint strength due to its proximity to the culture medium, while the tip of the bone needle has a low constraint strength due to the sparse hyphae.

[0034] For example, the calculation of the tension gradient adopts the difference method, dividing the bone needle into several micro-element segments along the length direction, each segment with a length of 0.5 millimeter. The difference of the constraint strength of adjacent micro-element segments is calculated, and then divided by the distance between the segments to obtain the local tension gradient. The curvature of the original shape of the bone needle is determined by the three-point method, selecting three consecutive points on the bone needle to form an arc, and the reciprocal of the radius of the arc is the curvature value.

[0035] It can be understood that in the weighted summation process of the deformation stress, the setting of the weight coefficient is based on the principle of material mechanics. The farther the position from the center of gravity of the bone needle, the greater the bending moment generated under the action of external force, and the more significant the contribution to the overall deformation. The weight value is linearly proportional to the distance, and the proportional coefficient is determined by experiment calibration. The deformation stress of each part is multiplied by the corresponding weight and then accumulated to obtain the value of the deformation aggravation degree, which directly reflects the deformation tendency of the bone needle under the condition of adhesion and stacking.

[0036] In step S103, the interaction strength between the bone needle structure and the mycelium attachment is analyzed according to the deformation aggravation degree, and the potential increase level of the sorting misjudgment rate is obtained.

[0037] The elastic modulus of the bone needle material and the moment of inertia of the bone needle cross section are obtained, the deformation aggravation degree is multiplied by the cube of the length of the bone needle, and then divided by the product of the elastic modulus and the moment of inertia to obtain the maximum deflection value of the bone needle. The ratio of the maximum deflection value to the length of the bone needle is taken as the structural deformation rate. Based on the structural deformation rate, the coverage of the mycelium on the surface of the bone needle in each region is identified by image segmentation, the proportion of the number of mycelium pixels in each region to the total number of pixels in the region is counted, and the mycelium attachment density distribution data is obtained. The average value of the attachment density distribution data is multiplied by the deformation aggravation degree to obtain the interaction strength between the bone needle structure and the mycelium attachment. According to the interaction strength, the sorting cases with the interaction strength in the same interval are screened from the historical sorting records, the number of bone needles incorrectly classified among them is counted, and the proportion of the number of misjudgments to the total number of sorting in the interval is taken as the benchmark misjudgment rate. The difference between the current interaction strength and the average value of the historical interaction strength is divided by the standard deviation of the historical interaction strength to obtain the deviation coefficient. The deviation coefficient is multiplied by the benchmark misjudgment rate to determine the potential increase level of the sorting misjudgment rate.

[0038] Specifically, in one embodiment, the elastic modulus of the bone needle material is determined by a three-point bending test. The bone needle is placed horizontally on two support points with a distance of two-thirds of the length of the bone needle, and a vertical downward load is applied at the midpoint of the bone needle. The load and displacement curve is recorded. The slope of the linear segment of the curve and the combination of the geometric parameters of the bone needle can be used to calculate the elastic modulus. The moment of inertia of the cross section of the bone needle is determined according to the cross-sectional shape of the bone needle. For a bone needle with a nearly circular cross section, the moment of inertia is equal to π times the fourth power of the diameter divided by 64.

[0039] It should be noted that the structural deformation rate reflects the overall deformation degree of the spicule under the constraint of the mycelium. The deformation aggravation degree, as the equivalent value of external load, is proportional to the cube of the length of the spicule, which is consistent with the deflection calculation principle of the cantilever beam in material mechanics. When the structural deformation rate exceeds 5%, the spicule is prone to morphological changes during sorting, affecting the sorting accuracy.

[0040] Specifically, the identification of the mycelium attachment density distribution adopts a threshold-based image segmentation method. First, the spicule surface image is converted into a gray image, and the mycelium area presents a darker gray value due to the difference in optical properties. Set the gray threshold, and the pixels below the threshold are determined as the mycelium coverage area, and the pixels above the threshold are the spicule body. The spicule surface is divided into several grid areas, and the proportion of mycelium pixels in each grid is counted to form a density distribution matrix. The average value of the density distribution data reflects the overall coverage degree of the mycelium, and the product of the deformation aggravation degree embodies the synergistic effect of mechanical constraint and biological attachment.

[0041] Preferably, the historical sorting records are divided into multiple intervals according to the interaction intensity value, and the width of each interval is 0.1 standardized unit. In each interval, the number of spicules that are incorrectly classified into other grades is counted, and these spicules are mainly misjudged due to changes in size or morphological characteristics caused by deformation. The calculation of the deviation coefficient adopts standardization processing to make the data of different batches comparable.

[0042] For example, when the deviation coefficient is 2, it means that the current interaction intensity is two standard deviations higher than the historical average value, which belongs to the case of abnormally high deviation, and the sorting misjudgment rate may reach more than twice the benchmark value. Therefore, it is necessary to adjust the sorting parameters in time to reduce the risk of misjudgment.

[0043] Step S104, if the potential increase level of the sorting misjudgment rate exceeds the preset increase level standard, the air flow intensity and the screen aperture are obtained from the historical data, the air flow intensity and the screen aperture are adjusted according to the risk of adhesion and stacking, and a set of dynamic adjustment instructions is obtained.

[0044] If the potential increase level of the sorting error rate exceeds the preset increase level standard, historical sorting records matching the current adhesion force value and flexibility value are retrieved from the historical database, the air flow intensity value and the screen mesh aperture value are extracted, the average value of the air flow intensity in the same adhesion force interval is calculated as the air flow intensity reference value, and the average value of the screen mesh aperture in the same flexibility interval is calculated as the screen mesh aperture reference value. The adjustment coefficient is determined according to the ratio of the adhesion and stacking risk level to the preset risk level, the adjustment coefficient is greater than 1 when the risk level is higher than the preset level, and vice versa. The air flow intensity reference value is multiplied by the adjustment coefficient to obtain the adjusted air flow intensity value, and the screen mesh aperture reference value is multiplied by the reciprocal of the adjustment coefficient to obtain the adjusted screen mesh aperture value. If the adjusted value exceeds the allowable range of the equipment, the boundary value is taken as the actual execution value. The actual execution value is converted into a hexadecimal control code, device address identification and command type identification are added in front of the control code, the execution priority is set according to the adjustment amplitude, the larger the adjustment amplitude, the higher the priority, time stamp information and check code are added, and a dynamic adjustment instruction set containing the air flow intensity adjustment instruction and the screen mesh aperture adjustment instruction is formed.

[0045] Specifically, in an embodiment, the historical database adopts a relational database structure, and is indexed according to the numerical ranges of adhesion force and flexibility. Each historical record contains environmental parameters, bone needle characteristic parameters, device running parameters and sorting results. When retrieval is needed, the numerical interval to which the current adhesion force and flexibility belong is first determined, the interval width is set to 0.5 times the standard deviation, and then all the historical records of successful sorting in the interval are queried.

[0046] It should be noted that the determination of the air flow intensity reference value and the screen mesh aperture reference value adopts a statistical average method. For all historical records in the same adhesion force interval, the air flow intensity value is extracted, and the arithmetic mean value is calculated as the air flow intensity reference value of the interval after removing the outliers. The calculation method of the screen mesh aperture reference value is similar, but the influence of flexibility on aperture selection needs to be considered.

[0047] Specifically, the determination of the adjustment coefficient is based on the quantitative evaluation of the risk level. The adhesion and stacking risk level is divided into five levels, from low to high, corresponding to adjustment coefficients of 0.8, 0.9, 1.0, 1.1 and 1.2 respectively. When the actual risk level is between two standard levels, the corresponding adjustment coefficient is calculated by using the linear interpolation method. The air flow intensity is proportional to the adjustment coefficient, and the higher the risk, the stronger the air flow required to separate the adhesion of the bone needles. The screen mesh aperture is inversely proportional to the adjustment coefficient, and the sorting accuracy is improved by reducing the aperture.

[0048] Preferably, the boundary limits of the actual execution values are determined according to the physical capabilities of the device. The upper limit of the air flow intensity is limited by the power of the fan, generally not exceeding 90% of the rated power; the lower limit needs to ensure that the needles can be blown up, generally not less than 20% of the rated power. The adjustment range of the screen mesh size is determined by the mechanical structure, the minimum mesh size cannot be smaller than the minimum diameter of a single needle, and the maximum mesh size cannot exceed one-third of the target needle length.

[0049] Illustratively, the encoding of the control instructions adopts the format of a standard industrial communication protocol. The hexadecimal control code occupies 4 bytes, the first two bytes represent the parameter type and value, and the last two bytes are used for verification. The device address identification uses an 8-bit binary code, supporting addressing of up to 256 devices. The execution priority is divided into three levels: emergency adjustment, regular adjustment, and preventive adjustment, corresponding to different response time requirements.

[0050] For example, when the adjustment amplitude exceeds 30% of the reference value, it is set to the emergency adjustment level, and the device responds within 100 milliseconds after receiving the instruction; the adjustment amplitude is between 10% and 30% for the regular level, with a response time of 500 milliseconds; less than 10% is preventive adjustment, which can be completed within 1 second.

[0051] Step S105, adjust the air flow intensity and screen mesh size of the sorting device by executing the adjustment instruction set, obtain the single-layer distributed needle state of the adjusted sorting device, and identify the distribution uniformity of the single layer of needles.

[0052] The control codes in the adjustment instruction set are sent to the sorting device through the serial communication protocol, the fan speed is adjusted according to the air flow intensity adjustment instruction, and the step motor is driven to change the screen opening according to the screen mesh size adjustment instruction. After the parameters are adjusted and stable, an industrial camera is used to shoot the sorting area image to obtain the single-layer distributed needle state of the adjusted sorting device. The single-layer distributed needle state image is uniformly divided according to a pre-set grid size, the number of needles in each grid is counted, the ratio of the standard deviation to the average value of all grid needle numbers is calculated, and the distribution uniformity of the single layer of needles is obtained.

[0053] Specifically, in an embodiment, the serial communication adopts RS485 protocol, the baud rate is set to 9600 bps, the data bits are 8 bits, and the stop bits are 1 bit. The control code is assembled according to the pre-defined format of the device, including the command header, parameter value and check bit. The fan speed is adjusted by the frequency converter, and the speed range is between 500 and 3000 revolutions per minute. The screen opening is driven by a stepping motor, and the precise adjustment of the screen distance is realized through a gear transmission mechanism. The size of the grid division is determined according to the average length of the bone needles. The image is divided into several square grids, and the length of each grid is set to 1.5 times the average length of the bone needles, so as to avoid repeated counting caused by a single bone needle spanning multiple grids. The calculation of the distribution uniformity adopts the coefficient of variation method. After counting the number of bone needles in each grid, the average value and the standard deviation of the number of bone needles in all grids are calculated, and the coefficient of variation is obtained by dividing the standard deviation by the average value. The smaller the coefficient of variation, the more uniform the distribution of bone needles. When the coefficient of variation is less than 0.3, it is considered that a good single-layer distribution state is achieved.

[0054] Step S106, comparing the distribution uniformity of the single layer of bone needles adjusted by the sorting device with the target distribution uniformity to obtain a distribution uniformity deviation level, if the distribution uniformity deviation level exceeds the set range, calculating the adhesion force change trend and the flexibility change trend according to the risk of adhesion and stacking and the degree of deformation aggravation, and determining the final sorting parameters according to the adhesion force and flexibility change trends.

[0055] The distribution uniformity of the single layer of bone needles adjusted by the sorting device is calculated by difference with the pre-set target distribution uniformity to obtain a distribution uniformity deviation value, and the distribution uniformity deviation value is divided by the target distribution uniformity to obtain a deviation level, if the deviation level exceeds the pre-set deviation threshold, the current adhesion and stacking risk value and the deformation aggravation value are extracted as a feature vector. The feature vector and the adhesion force and flexibility time series data extracted from the database are input into the support vector regression algorithm, and a regression model is obtained by training. The regression model outputs the adhesion force change slope and the flexibility change slope, and the adhesion force and flexibility values in the future time period are predicted according to the adhesion force change slope and the flexibility change slope respectively, and the adhesion force change trend and the flexibility change trend are obtained. Based on the adhesion force change trend and the flexibility change trend, the airflow intensity reference value is determined according to the adhesion force value interval, the screen aperture reference value is determined according to the flexibility value interval, the sorting speed adjustment coefficient is calculated by the change rate of the change trend, the reference value is multiplied by the adjustment coefficient to obtain the adjusted airflow intensity and screen aperture, and the sorting speed is calculated according to the number of bone needles per unit area. The adjusted airflow intensity, screen aperture and sorting speed are combined to form a final sorting parameter set containing three parameters, and the final sorting parameter set is output to the sorting device actuator.

[0056] Specifically, in one embodiment, the setting of the target distribution uniformity is based on the industry standard and product quality requirement of the sponge needle sorting. According to the specification requirements of different grades of bone needle products, the coefficient of variation of the target distribution uniformity is usually set between 0.15 and 0.25. The calculation of the deviation level uses the relative deviation method, that is, the difference between the actual distribution uniformity and the target value is divided by the target value, and the percentage form of the deviation level can intuitively reflect the deviation degree of the sorting quality.

[0057] It should be noted that the adhesion stacking risk value and the deformation aggravation degree value as two dimensions of the feature vector reflect the current physical state characteristics of the bone needles. These two characteristic values have time correlation and will dynamically change with changes in environmental factors such as temperature and humidity during fermentation. The construction of the feature vector requires standardization processing of the original data to eliminate the dimension influence, so that different physical quantities can be compared and calculated on the same scale.

[0058] Preferably, the implementation process of the support vector regression algorithm includes four stages of training data preparation, kernel function selection, parameter optimization and model verification. The training data comes from the adhesion and flexibility time series data collected in the historical fermentation batches, each batch contains complete data sequence from the beginning to the end of fermentation, and the sampling interval is 10 minutes. The kernel function is selected as the radial basis function, and the parameter γ is determined by cross-validation, usually with a value range of 0.001 to 0.1. The penalty parameter C controls the balance between the complexity and fitting error of the model, and too large C value will cause overfitting, and too small will cause underfitting. The grid search method is used to find the parameter combination that minimizes the validation set error within the preset parameter range. The trained regression model can predict the change slope of adhesion and flexibility according to the current feature vector, and the positive and negative of the slope represents the increasing or decreasing trend, and the absolute value of the slope reflects the change speed. Based on the change slope, the adhesion and flexibility values in the next 30 minutes are predicted by linear extrapolation method, forming the change trend curve.

[0059] Exemplarily, the determination of the airflow intensity reference value uses a segmented mapping method. The adhesion value range is divided into several intervals, and each interval corresponds to an airflow intensity reference value. When the adhesion is in the low interval, the adhesion between the bone needles is weak, and a lower airflow intensity can achieve separation; as the adhesion increases, the airflow intensity needs to be increased accordingly. The determination principle of the screen aperture reference value is similar, but it has an inverse relationship with flexibility. The higher the flexibility, the easier the bone needle deforms through the screen aperture, and the aperture needs to be reduced to improve the sorting precision.

[0060] Specifically, the calculation of the sorting speed adjustment coefficient considers the dynamic characteristics of the change trend. When the adhesion and flexibility change rapidly, it indicates that the fermentation environment is unstable, and the sorting speed needs to be reduced to improve the sorting accuracy; on the contrary, when the change trend is gentle, the sorting speed can be appropriately increased to increase the production capacity. The adjustment coefficient is obtained by normalizing the change rate, and the value range is between 0.5 and 1.5.

[0061] In one possible implementation, the number of bone needles per unit area is calculated through image processing. Image acquisition is performed on the sorting area, the bone needle contour is identified through morphological processing, and the number of bone needles per unit area is counted. The higher the bone needle density, the more bone needles pass through the sorting area per unit time, and the sorting speed needs to be correspondingly reduced to avoid the accumulation of bone needles affecting the sorting effect.

[0062] For example, when the adhesion change trend shows that the future will rise rapidly, and the flexibility remains stable, the air flow strength needs to be increased in advance, while the screen aperture remains unchanged, and the sorting speed is appropriately reduced. This predictive adjustment can complete parameter adjustment before the physical properties of the bone needle change significantly, avoiding sudden decline in sorting quality.

[0063] It can be understood that there is a mutual restraint relationship between the three parameters of the final sorting parameter set. Too large air flow strength may cause bone needle splashing, affecting single-layer distribution; too small screen aperture will increase the risk of blockage; and too fast sorting speed will reduce the sorting accuracy. By adjusting the three parameters in coordination, the production capacity is maximized under the premise of ensuring the sorting quality. Further, after the parameter set is received by the actuator of the sorting equipment, each parameter is adjusted in turn according to the preset execution order. The screen aperture is adjusted first, the air flow strength is adjusted after the mechanical structure is stable, and finally the sorting speed is set. This order arrangement takes into account the difference in response time of the equipment. Mechanical adjustment requires a long time, while electrical parameter adjustment is relatively fast. A reasonable execution order can shorten the overall adjustment time.

[0064] Step S107, sorting the sponge bone needles by applying the final sorting parameters to the sponge bone needle sorting system to obtain single-layer distributed sponge bone needles, and evaluating the stability of the sponge bone needle sorting process by analyzing the uniformity and sorting misjudgment rate of the single-layer distributed bone needles.

[0065] The final sorting parameter set is transmitted to the controller of the sponge bone needle sorting device through serial communication. The device runs the sorting process according to the set air flow intensity, screen mesh size and sorting speed, and real-time collects the bone needle throughput data and sorting area images in the sorting process to obtain single-layer distributed sponge bone needles. The image of the single-layer distributed sponge bone needles is grid divided, the number of bone needles in each grid is counted, and the ratio of the standard deviation to the average of the number of bone needles in all grids is calculated as the uniformity index. At the same time, bone needle samples are randomly taken from the sorting output end, the proportion of the number of misclassified bone needles to the total sampling number is counted, and the sorting misjudgment rate is obtained. According to the uniformity index and the sorting misjudgment rate, the weighted average of the two is calculated as the comprehensive quality index according to the preset weight coefficient. If the fluctuation range of the comprehensive quality index in the continuous five sampling periods is less than the preset stability threshold, it is determined that the sponge bone needle sorting process meets the stability requirement.

[0066] Specifically, serial communication adopts Modbus protocol for parameter transmission, and the air flow intensity, screen mesh size and sorting speed are packaged into a data frame and sent to the sorting device controller through RS485 bus. After the controller parses the data frame, it sends control instructions to the fan frequency converter, screen adjustment motor and conveyor belt driver respectively to realize synchronous updating of parameters. During the execution of the sorting process, the bone needle throughput data is collected by the photoelectric sensor installed at the outlet of the sorting channel. A pulse signal is generated every time a bone needle passes through, and the controller accumulates the pulse number to obtain the throughput. The sorting area image is taken by the industrial camera fixed above the screen every 5 seconds. The image resolution is set to 2048x1536 pixels to ensure that a single bone needle can be clearly identified. The implementation of grid division adopts a fixed-size square grid, and the grid side length is determined according to 1.5 times the average length of the bone needle. After the collected image is binarized, the bone needle contour is recognized, and the number of bone needle center points in each grid is counted. The uniformity index is obtained by calculating the coefficient of variation of the number of bone needles in all grids. The coefficient of variation is equal to the standard deviation divided by the average value, and the smaller the value, the more uniform the distribution. The misjudgment rate is calculated by stratified random sampling. Three sampling points are set at the sorting output end, corresponding to different levels of bone needle outlets. Every 10 minutes, 50 bone needles are taken as samples from each sampling point. Whether each bone needle is correctly classified is determined by manual re-inspection or image recognition, the number of misclassified bone needles is counted, and the misjudgment rate at that time point is obtained by dividing the total sampling number.

[0067] For example, the weight coefficient of the comprehensive quality index is dynamically adjusted according to the product quality requirements. When producing high-grade bone needle products, the uniformity weight is set to 0.7 and the misjudgment rate weight is set to 0.3, which focuses more on the uniformity of the distribution; when producing ordinary grade products, the weights of the two are each 0.5, taking into account the uniformity and sorting accuracy.

[0068] For example, the stability determination is evaluated by using a moving average method. In five consecutive sampling periods, the comprehensive quality index of each period is calculated, and then the range of the five index values, i.e. the maximum value minus the minimum value, is calculated. If the range is less than a preset stability threshold value 0.05, it is considered that the sorting process reaches a stable state, and the current parameters can be maintained to continue production.

[0069] The above description is merely preferred embodiments of the present application, and is not intended to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for predicting the sorting of spicules in fermented sponges using machine learning, characterized in that, include: The fermentation temperature and sponge spicule surface images in the fermentation environment are continuously collected. The hyphal density and morphology in the sponge spicule surface images are identified, and the trends of adhesion and flexibility of sponge spicules under different fermentation temperatures are generated. Based on the trends of adhesion and flexibility, the range of change in hyphal adhesion is identified, the risk of adhesion and stacking is determined based on the range of change, and the degree of deformation aggravation is obtained through surface tension analysis. Based on the degree of deformation aggravation, the interaction strength between the spicule structure and hyphal attachment is analyzed to obtain the potential increase level of the sorting misjudgment rate. A dynamic adjustment instruction set is generated based on this potential increase level. The airflow intensity and screen aperture of the sorting equipment are adjusted by executing the adjustment instruction set to obtain the adjusted monolayer spicule distribution state and identify the distribution uniformity of the monolayer spicules. The distribution uniformity is compared with the target distribution uniformity to obtain the distribution uniformity deviation level. Sorting parameters are determined based on the deviation level, the risk of adhesion and stacking, and the degree of deformation aggravation. Sponge spicules are sorted using these sorting parameters to obtain monolayer sponge spicules. The uniformity of the monolayer sponge spicules and the sorting misjudgment rate are analyzed to obtain the stability index of the sorting process.

2. The method for sorting fermented sponge spicules using machine learning prediction according to claim 1, characterized in that, The process involves continuously acquiring fermentation temperature and sponge spicule surface images within the fermentation environment, identifying hyphal density and morphology in the sponge spicule surface images, and generating trends in the adhesion and flexibility of the sponge spicules under different fermentation temperatures, including: Temperature data from different locations within the fermenter were collected to obtain surface images of the sponge spicules. The fermentation zone of the spicules was determined based on the temperature gradient distribution. Mean filtering and histogram equalization were applied to the surface images to obtain a spicule surface texture image. The spicule surface texture image was then converted to grayscale. An edge detection algorithm was used to identify hyphal contours. The hyphal density was determined by calculating the number of hyphal interlacing points per unit area. Hyphae structural features were extracted based on the radius of curvature and branching angle of the hyphal morphology. The ratio of the hyphal coverage area to the spicule surface area was calculated based on the hyphal density value to obtain the adhesion coefficient. The flexibility modulus was calculated based on the deviation angle of the spicule curvature from a standard straight line in the spicule surface texture image. The adhesion coefficient and flexibility modulus were arranged in a time series, segmented by a fixed time window, and a curve fitting method was used to determine the trends of adhesion force and flexibility under different temperature gradients.

3. The method for sorting fermented sponge spicules using machine learning prediction according to claim 1, characterized in that, The process of identifying the variation range of hyphal adhesion based on the trends of adhesion force and flexibility, determining the risk of adhesion and stacking based on the variation range, and obtaining the degree of deformation aggravation through surface tension analysis includes: The numerical differences between adjacent moments in the trends of adhesion force and flexibility are extracted, and the standard deviation of the data sequence is calculated as the fluctuation amplitude. A time series feature vector is constructed, and the time series feature vector is processed by a classification algorithm to output the hyphal adhesion level value. The difference between the level values ​​at adjacent time points is calculated to obtain the amplitude of hyphal adhesion change. Based on the correlation mapping between the amplitude of hyphal adhesion change and the distance between adjacent spicules, the adsorption force between spicules is calculated. By comparing the adsorption force with the product of the spicule mass and gravitational acceleration, the adhesion and stacking risk level is determined. The hyphal network density data is obtained, and the surface tension coefficient is calculated using the droplet contact angle measurement method. The hyphal constraint strength value is obtained by multiplying the surface tension coefficient with the hyphal network density. Based on the distribution of the hyphal constraint strength value along the length direction of the spicule, the difference in constraint strength at adjacent positions is calculated to obtain the tension gradient. The tension gradient is multiplied by the original curvature of the spicule to obtain the deformation stress at each part. The weighted sum is then used to obtain the degree of deformation aggravation.

4. The method for sorting fermented sponge spicules using machine learning prediction according to claim 1, characterized in that, The analysis of the interaction strength between the bone spur structure and hyphal attachment based on the degree of deformation aggravation yields the potential increase level of sorting misclassification rate, including: The maximum deflection value of the bone needle is calculated by combining the elastic modulus and cross-sectional moment of inertia of the bone needle material with the degree of deformation aggravation. The coverage ratio of hyphae in each region of the bone needle surface is statistically analyzed to obtain the hyphae attachment density distribution. The interaction strength is obtained by multiplying the average value of the hyphae attachment density distribution with the degree of deformation aggravation.

5. The method for sorting fermented sponge spicules using machine learning prediction according to claim 1, characterized in that, The step of generating a dynamic adjustment instruction set based on the potential increase level of the sorting misjudgment rate includes: An adjustment coefficient is determined based on the ratio of the adhesion and stacking risk level to a preset level. The airflow intensity and screen aperture value are adjusted using the adjustment coefficient to generate a dynamic adjustment instruction set containing control code, device address identifier, and timestamp.

6. The method for sorting fermented sponge spicules using machine learning prediction according to claim 1, characterized in that, The step of adjusting the airflow intensity and screen aperture of the sorting device by executing the adjustment instruction set, obtaining the adjusted single-layer distributed bone needle state, and identifying the distribution uniformity of the single-layer distributed bone needle includes: Send control codes from the adjustment instruction set to adjust the fan speed and screen opening, capture images of the sorting area, and obtain the state of the single-layer distributed bone needles; divide the image of the single-layer distributed bone needle state into grids, count the number of bone needles in each grid, calculate the ratio of the standard deviation of the number to the mean, and obtain the distribution uniformity.

7. The method for sorting fermented sponge spicules using machine learning prediction according to claim 1, characterized in that, The process involves comparing the distribution uniformity with the target distribution uniformity to obtain the distribution uniformity deviation level. Based on this deviation level, the risk of adhesion and stacking, and the degree of deformation aggravation, sorting parameters are determined, including: The difference between the distribution uniformity and the target distribution uniformity is calculated to obtain the deviation level. Combined with the adhesion and stacking risk value and the deformation aggravation value, the airflow intensity, screen aperture and sorting speed are determined and combined to form a sorting parameter set.

8. The method for sorting fermented sponge spicules using machine learning prediction according to claim 1, characterized in that, The process involves sorting sponge spicules using the sorting parameters to obtain a single-layer distribution of sponge spicules. The uniformity and sorting error rate of the single-layer distribution of sponge spicules are analyzed to obtain stability indicators for the sorting process, including: The sorting parameter set is transmitted to the sorting equipment controller, and the equipment operates according to the set airflow intensity, screen aperture, and sorting speed. Images of the sorting area are acquired to obtain the single-layer distributed sponge spicules. The images of the single-layer distributed sponge spicules are divided into grids, and the ratio of the standard deviation to the average value of the number of spicules in each grid is calculated as a uniformity index. The proportion of misclassified spicules is calculated as the sorting misclassification rate. The stability index of the sorting process is obtained based on the weighted average of the uniformity index and the sorting misclassification rate.

Images