A condensed milk flavor optimization method based on multi-level flavor profile analysis

Abstract

The present application provides a condensed milk flavor optimization method based on multi-level flavor characteristic analysis, comprising: extracting the contours of the fat globules using an edge detection algorithm from the segmented fat globule binary image to obtain a contour curve of the fat globule, calculating the average curvature of the fat globule surface based on the curvature change of the contour curve, and obtaining curvature data characterizing the shape of the fat globule; extracting the membrane area of the fat globule using a morphological processing algorithm from the segmented fat globule binary image, calculating the average membrane thickness of the fat globule based on the number of pixels in the extracted membrane area, and obtaining membrane thickness data characterizing the fat globule membrane characteristics; obtaining fat globule membrane component detection data of different condensed milk samples, including membrane protein content and phospholipid content, calculating the ratio of protein to phospholipid in the fat globule membrane based on the membrane protein content and phospholipid content, and obtaining protein-phospholipid ratio data characterizing the composition of the fat globule membrane.

Description

Technical Field

[0001] The present invention relates to the field of information technology, in particular to a condensed milk flavor optimization method based on multi-level flavor characteristic analysis. Background Art

[0002] The flavor of condensed milk is not simply sweet; it offers a multifaceted, complex experience, closely related to the microstructure of its fat globules. However, the morphological factors that influence condensed milk flavor are numerous and interrelated, making determining the optimal morphology a challenge. First, fat globule size directly influences its surface area. Smaller fat globules have a larger surface area, theoretically allowing for the release of more encapsulated flavor compounds. However, they are also more susceptible to fat oxidation, resulting in unpleasant flavors. Larger fat globules, on the other hand, may result in insufficient flavor release and an overly rich mouthfeel. Second, fat globules are not perfectly spherical, and variations in their surface curvature can affect the structure and stability of the membrane. Regions with varying curvatures may exhibit varying membrane thickness and composition, thus affecting the adsorption and release of flavor compounds. Finally, the thickness and composition of the fat globule membrane are also important factors. Membrane thickness determines the diffusion rate of flavor compounds, while membrane composition, particularly the ratio of protein to phospholipids, influences membrane permeability and its ability to adsorb different flavor compounds. A higher protein ratio may improve membrane stability, but it can also hinder flavor release. How to regulate the membrane thickness and protein-phospholipid ratio to achieve optimal flavor release remains a challenge. Summary of the Invention

[0003] The present invention provides a condensed milk flavor optimization method based on multi-level flavor characteristic analysis, which mainly comprises:

[0004] Obtain fat globule image data of different condensed milk samples, extract the fat globule area using an image segmentation algorithm, obtain a segmented fat globule binary image, calculate the equivalent diameter of the fat globule based on the pixel area of the fat globule in the binary image, and obtain the fat globule size distribution data;

[0005] For the fat globule binary image after segmentation, the edge detection algorithm is used to extract the contour of the fat globule to obtain the contour curve of the fat globule. According to the curvature change of the contour curve, the average curvature of the fat globule surface is calculated to obtain the curvature data representing the shape of the fat globule;

[0006] The morphological processing algorithm is used to extract the membrane area of the fat globules from the segmented binary image. The average membrane thickness of the fat globules is calculated based on the number of pixels in the extracted membrane area to obtain the membrane thickness data that characterizes the fat globule membrane characteristics.

[0007] Obtaining fat globule membrane component test data of different condensed milk samples, including membrane protein content and phospholipid content, and calculating the ratio of protein to phospholipid in the fat globule membrane based on the membrane protein content and phospholipid content, thereby obtaining protein-phospholipid ratio data that characterizes the composition of the fat globule membrane;

[0008] Obtain flavor characteristic evaluation data for different condensed milk samples, including flavor hierarchy scores. Use the flavor hierarchy scores as sample label data, and use fat globule size distribution, surface curvature, membrane thickness, and membrane component ratio as sample feature data to construct a flavor hierarchy prediction model.

[0009] A flavor hierarchy prediction model is trained using a support vector machine algorithm to obtain a trained flavor hierarchy prediction model. The fat globule size distribution, surface curvature, membrane thickness, and membrane component ratio data of the condensed milk sample to be judged are input into the trained prediction model to obtain a flavor hierarchy prediction score for the sample.

[0010] According to the flavor hierarchy prediction scores, the flavor hierarchy of the samples was ranked to obtain the sample ranking results from high to low flavor hierarchy. Based on the ranking results, the combination of fat globule size distribution, surface curvature, membrane thickness, and membrane component ratio corresponding to the sample with the highest flavor hierarchy was determined as the fat globule morphological feature combination with the highest flavor characteristic level.

[0011] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0012] The present invention discloses a method for optimizing the flavor of condensed milk by multi-level flavor feature analysis. The method obtains morphological feature data such as the size distribution, surface curvature and membrane thickness of fat globules in condensed milk samples through image analysis, and constructs a flavor hierarchy prediction model in combination with the fat globule membrane component detection data. The specific steps include: segmenting and extracting features from fat globule images, calculating the equivalent diameter, surface curvature and membrane thickness of fat globules; detecting the protein and phospholipid content ratio of the fat globule membrane; using these feature data and flavor hierarchy scores as training data, and constructing a prediction model using a support vector machine algorithm. The model can predict the flavor hierarchy of the sample to be tested, and sort the samples to determine the combination of fat globule morphological features with the highest flavor hierarchy. The present invention realizes the objective quantitative evaluation of the flavor hierarchy of condensed milk, provides a scientific basis for condensed milk quality improvement and new product development, and helps to improve the flavor quality of condensed milk. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 The present invention provides a flow chart of a method for optimizing condensed milk flavor through multi-level flavor characteristic analysis.

[0014] Figure 2The figure is a schematic diagram of a condensed milk flavor optimization method based on multi-level flavor characteristic analysis of the present invention.

[0015] Figure 3 This is another schematic diagram of the condensed milk flavor optimization method based on multi-level flavor characteristic analysis of the present invention. DETAILED DESCRIPTION

[0016] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] like Figure 1-3 The condensed milk flavor optimization method of this embodiment using multi-level flavor characteristic analysis may specifically include:

[0018] S101. Obtain fat globule image data of different condensed milk samples, extract fat globule regions from the fat globule images using an image segmentation algorithm, obtain segmented fat globule binarized images, calculate equivalent diameters of the fat globules based on the pixel areas of the fat globules in the binarized images, and obtain fat globule size distribution data.

[0019] An imaging processor is used to obtain image data of a condensed milk sample with a calibration plate scale, and a conversion relationship between pixels and actual sizes is established based on the calibration plate scale to obtain a calibration image; Gaussian low-pass filtering is performed on the calibration image to remove background noise, and a multi-level grayscale threshold is calculated using the OTSU algorithm. The image contour boundary is extracted based on the grayscale threshold to obtain a boundary image; seed point growth processing is performed on the boundary image, and if the grayscale value difference in the overlapping area exceeds the standard deviation of the neighborhood average, a watershed algorithm is used to divide the area into independent blocks to obtain a segmented image; fat globule contour area data is extracted from the segmented image, the actual area value is calculated based on the conversion relationship between pixels and actual sizes, and the equivalent diameter is calculated using the circular area formula to obtain the particle size distribution result.

[0020] Specifically, an image processor is used to obtain condensed milk sample image data by comparing it with a standard calibration plate. A histogram equalization operation is performed on the image data. A Gaussian low-pass filter is performed on the central area of the image to remove background noise points. A conversion relationship between pixel points and actual sizes is established according to the scale size of the calibration plate, and enhanced processing data is obtained from the filtered image. The OTSU algorithm is used to iteratively calculate the multi-level grayscale threshold based on the enhanced data. The contour boundary of the independent area after threshold segmentation is extracted. The boundary curvature value is calculated by the directed gradient function. The integrity of the fat globule edge is judged based on the curvature value, and a binary image is generated in the area with complete edges. A seed point growth process is performed on the binary image. The regional grayscale value difference threshold is set to the standard deviation of the neighborhood average value. The overlapping area is divided into independent blocks by the watershed algorithm. The holes in the area are filled by the closing operation. The small area noise area is removed by the opening operation. The pixel set of the independent area is marked. The contour area of each fat globule was extracted from the set of marked regions. The actual area was calculated using the size conversion relationship established by the calibration plate. The equivalent diameter was then calculated using the circular area formula. A frequency distribution histogram was constructed for the obtained data, and the fat globule size distribution was obtained from the histogram data. The calibration plate, a benchmark size conversion tool, has a standard millimeter scale. The calibration plate and condensed milk sample were simultaneously imaged to establish the correspondence between pixels and actual size. Each 1 mm scale mark corresponds to 25 pixels in the image. The calibration plate is matte black, providing a sharp contrast with the condensed milk sample. Histogram equalization adjusts the image grayscale distribution to make it more uniform and enhance image contrast. The original image grayscale values were concentrated between 120 and 180, while the equalized grayscale values were evenly distributed between 0 and 255, enhancing the contrast between the fat globule edges and the background. A Gaussian low-pass filter using a 5×5 mask removed high-frequency noise while maintaining the integrity of the fat globule edges. The OTSU algorithm uses the maximum inter-class variance criterion to calculate the optimal segmentation threshold. It performs multi-level iterative segmentation on the grayscale image. Each iteration calculates the mean grayscale value and inter-class variance of the two subclasses. The segmentation threshold is determined when the inter-class variance reaches its maximum value, resulting in a threshold of 135. Directed gradients are used to calculate boundary curvature, with curvature values greater than 0.85 considered complete edges. Seed point growth selects initial growth points from the center of the image, using a neighborhood grayscale difference of less than 15 as the growth criterion for region segmentation. The watershed algorithm processes overlapping fat globules, segmenting them into independent blocks at the watershed line in the overlapping region. A closing operation fills holes within the region smaller than 50 square pixels, while an opening operation removes noise regions smaller than 100 square pixels. The contour area of the fat globules is calculated by counting the number of pixels in each marked region. Using a calibration relationship, the pixel area is converted to an actual area value, with 25×25 pixels corresponding to 1 square millimeter. The equivalent diameter is calculated using the circular area formula. Statistics show that the fat globule diameters are primarily distributed between 2 and 8 microns, with 4 to 6 microns accounting for 65%, constituting the primary particle size distribution range.In practice, fat globule distribution varies among different types of condensed milk samples. Whole-fat condensed milk has larger fat globules with a wide distribution, while skim milk has smaller, more concentrated fat globules. Quantitative analysis of fat globule size distribution allows for quality evaluation and classification of condensed milk. Image processing parameters must be adjusted based on sample characteristics to ensure accurate segmentation results. A comprehensive image acquisition and processing workflow is crucial for obtaining reliable particle size distribution data.

[0021] S102. For the segmented fat globule binary image, an edge detection algorithm is used to extract the contour of the fat globule to obtain the contour curve of the fat globule. According to the curvature change of the contour curve, the average curvature of the fat globule surface is calculated to obtain curvature data representing the shape of the fat globule.

[0022] The Sobel operator is used to extract edge gradients in the horizontal and vertical directions of the binary image, and an edge intensity distribution map is obtained through iterative calculation based on the horizontal gradient and the vertical gradient; a refinement iterative operation is performed on the edge intensity distribution map, and the unclosed points are connected and compensated by the connectivity propagator to obtain a closed contour line; a point index sequence is constructed according to the closed contour line, and the curvature value of the point index sequence is calculated using a three-point arc fitting function, and a smooth curvature value sequence is obtained through a Gaussian filter; if the curvature of a point in the smooth curvature value sequence is greater than zero, it is marked as a convex point, and if the curvature of the point is less than zero, it is marked as a concave point, and the average curvature result is obtained through a weighted operation of the extreme point curvature.

[0023] Specifically, a 3×3 Sobel edge detection operator is used to extract edge gradients in both the horizontal and vertical directions from the fat globule binary image. An edge intensity distribution map is then synthesized based on the horizontal and vertical gradients. The edge intensity threshold is determined by iteratively calculating the maximum inter-class variance in the map. Points with edge intensities below the threshold are suppressed to generate an edge point map. From the edge point map, the edge line width is compressed to a single pixel using a refinement iterative calculator. A 5×5 window scanner is used to detect the connectivity attribute value of each edge point in the refined edge map. The closure of the line segment in which the point lies is determined based on the connectivity attribute value. A connectivity propagator is used to compensate for unclosed points to generate a closed contour line. A point index sequence is constructed for the closed contour line in a counterclockwise direction. Curvature calculation is performed on the point sequence using a window of five consecutive points. A three-point arc fitting function is used to obtain the curvature values at each point in the sequence. The curvature value sequence is then smoothed using a 7-point Gaussian filter to obtain a smooth curvature value sequence for the contour curve. Based on a smooth curvature sequence, each point's curvature is marked as either convex or concave. The resulting contour is then divided into subsegments. Cubic spline interpolation is used to supplement the continuous curvature distribution within the subsegments. The coordinates of local maxima and minima are extracted from the continuous curvature values, and the average curvature is calculated by weighting the curvature of the extreme points. A Sobel edge detection operator uses a 3×3 template to detect horizontal and vertical gradients. The horizontal template calculates the grayscale difference in the row direction, while the vertical template calculates the grayscale difference in the column direction. The square root of the sum of the squares of the gradients in these two directions is used to determine the edge intensity. The intensity at the edge of the fat globule is significantly higher than that of the background. The edge intensity threshold is set between the background noise and the target edge. The initial threshold is set to the mean edge intensity when iteratively calculating the maximum inter-class variance. The image is divided into two categories: regions above and below the threshold. The average intensity of these two regions is calculated, and the new threshold is taken as the midpoint of these two averages. Iterations are repeated until the threshold change is less than 0.1, with the final threshold being between 150 and 200. During the refinement iteration process, the 8-neighborhood of each edge point is scanned. Edge points meeting these criteria are removed if the number of nonzero pixels in the 8-neighborhood is between 2 and 6. This removal process is repeated on the edge map until the width of all edge lines is reduced to a single pixel. Edge point connectivity is determined by the pixel distribution within a 5×5 window. The presence of two or more branch points in the neighborhood of the window center indicates that the point is an endpoint. Connectivity propagation is performed starting from the endpoint and extending outwards within a search radius of 3 pixels. The nearest endpoint is found to establish a connection and compensate for broken contour segments. Contour closure requires that all endpoints are connected in pairs to form a closed curve. The point index sequence starts at any point on the contour and records the coordinate values in a counterclockwise direction, with the index number starting from 0 and increasing. The curvature calculation window contains 5 consecutive points. The curvature of the center point is calculated by fitting a three-point arc using the three-point least squares method. The curvature value is the inverse of the arc radius.Gaussian filtering uses a 7-point weight template, with a central point weight of 0.4, flanked by 0.25, 0.15, and 0.05, respectively. This filtering eliminates sudden changes in the curvature sequence. Convex points have positive curvature, concave points have negative curvature, and subsegments are defined at the boundary between positive and negative curvature. Curvature within subsegments is continuously distributed using cubic spline interpolation, with the number of interpolation points being three times the number of original points. Local extreme points correspond to the peak positions of the absolute value of the curvature, with maximum curvature exceeding 0.5 and minimum curvature less than -0.5. The average curvature is calculated as a weighted average of the extreme point curvatures, with the weight proportional to the absolute value of the curvature, reflecting the overall shape characteristics of the fat globule.

[0024] S103. For the segmented fat globule binary image, a morphological processing algorithm is used to extract the membrane area of the fat globule, and the average membrane thickness of the fat globule is calculated based on the number of pixels in the extracted membrane area to obtain membrane thickness data representing the characteristics of the fat globule membrane.

[0025] A circular structure corrosion operator is used to erode the edges of the fat globule area in the binarized image, and a membrane area map is obtained through a closing operation; based on the point with the maximum grayscale value in the membrane area map as the seed point, a region growth extractor is used to expand it to all sides, and a membrane area skeleton line is obtained through distance transformation; the membrane area skeleton line is scanned inward and outward to obtain the inner and outer boundary coordinate points of the membrane area contour; the inner and outer boundary coordinate points are measured along the reference direction to obtain a local membrane thickness sequence, and if the sampling interval of the local membrane thickness sequence is a preset value, the average membrane thickness value is obtained based on Gaussian weighted averaging and Kalman filtering.

[0026] Specifically, a 5×5 circular structured erosion operator was used to erode the edges of the fat globule region in the binary image. The extent of erosion was determined by the ratio of the eroded area to the original area. A 7×7 circular structured closing operation was used to fill the edge gaps. The original area was subtracted from the eroded area to obtain a circular difference map. A 3×3 structured opening operation was performed on the difference map to remove independent noise points and obtain the membrane region map. The point with the maximum grayscale value in the membrane region map was selected as the seed point. A region growing extractor was used to expand the region in all directions. The growth threshold was set to the standard deviation of the grayscale of the eight neighborhoods of the seed point. A 5×5 structured closing operation was used to connect the breakpoints in the growing region. The membrane region skeleton was extracted using a distance transform. The complete membrane contour was obtained by scanning from the skeleton line inward and outward. The inner and outer boundary coordinates were extracted from the membrane region contour. The direction of the line connecting the corresponding points was used as the reference for thickness measurement. The distance between the two points was measured at a 90-degree angle along the reference direction to obtain the local membrane thickness. The thickness values were filtered using a bilateral filter with a spatial scale of 5 pixels and a grayscale scale of 0.3. A 7-point sliding median filter was used to remove jump points to obtain a smooth thickness sequence. The smooth thickness sequence was sampled at 10-pixel intervals, and local membrane thickness values were calculated by Gaussian weighted averaging within a 5-pixel radius of the sampling point. The local membrane thickness sequence was filtered using a Kalman filter with a position noise variance of 2 and a measurement noise variance of 0.5. The average membrane thickness value was obtained by averaging the filtered data sequence. The erosion operation reduced pixels from the fat globule edge inward. The 5×5 circular structure corresponds to a spatial scale of approximately 0.5 microns, which matches the typical thickness of the fat globule membrane. The area of the eroded region decreased by 15% to 20% compared to the original region, demonstrating that the edge erosion achieved the desired effect. A closing operation was used to connect disconnected edge points, and a 7×7 structure filled gaps smaller than 0.7 microns to maintain edge continuity. The annular difference map reflects the distribution characteristics of the membrane region. In the difference map, the membrane region has grayscale values between 180 and 220, significantly higher than the background region. An opening operation removed noise points with an area smaller than 9 pixels to highlight the true membrane structure. The seed point was selected in the region with a grayscale value of 215 or above to ensure that the growth started from the true membrane region. During the region growing process, the grayscale standard deviation of the 8-neighborhood is about 12, which can effectively control the region expansion range as the growing threshold.

[0027] A 5×5 closing operation connects the 2- to 3-pixel gaps. A distance transform calculates the shortest distance from each pixel to the background, and the point with the maximum distance forms the skeleton line. The width of the skeleton line in the membrane region is maintained at 1 pixel, reflecting the center of the membrane. The distance between the inner and outer boundary points reflects the local membrane thickness. The measurement reference direction is perpendicular to the membrane extension direction. Bilateral filtering smoothes membrane thickness variations while preserving edge transitions. A spatial scale of 5 pixels corresponds to an actual distance of 0.5 microns, and a grayscale of 0.3 retains significant thickness variations. A median filter window of 7 points eliminates isolated thickness outliers. A sampling interval of 10 pixels ensures independence between thickness measurement points while maintaining sufficient spatial resolution. Gaussian weighted averaging calculates the local mean within a 5-pixel range, with the weight decaying exponentially with increasing distance. The position noise variance of the Kalman filter reflects the natural fluctuations in membrane thickness, while the measurement noise variance corresponds to instrumental measurement error. In practice, membrane thickness distribution characteristics vary among different types of fat globules. Fresh samples have good membrane thickness uniformity, while samples with long storage periods exhibit uneven thickness. Quantitative analysis of membrane thickness data can be used to assess the structural integrity of fat globules. Image processing parameters need to be calibrated according to the microscope resolution and magnification to ensure the accuracy of the measurement results. A complete membrane area extraction and thickness measurement process is important for evaluating fat globule stability.

[0028] S104. Obtain fat globule membrane component detection data of different condensed milk samples, including membrane protein content and phospholipid content, and calculate the ratio of protein to phospholipid in the fat globule membrane based on the membrane protein content and phospholipid content to obtain protein-phospholipid ratio data characterizing the composition of the fat globule membrane.

[0029] A methanol-water gradient eluent is used to separate components through a solid phase extraction column, and the component separation obtains a protein absorbance signal and a phospholipid fluorescence intensity signal; baseline smoothing is performed on the protein absorbance signal and the phospholipid fluorescence intensity signal, and corrected signal data is obtained by internal standard correction; the corrected signal data is filtered using a wavelet filter and a low-pass filter, and component content values are obtained by normal distribution function fitting; the molar concentration of the component content values is calculated using a protein and phospholipid molecular weight conversion coefficient, and the protein-phospholipid molar ratio is obtained by variance-weighted averaging.

[0030] Specifically, the condensed milk sample was separated using an 18-carbon octyl silica gel solid-phase extraction column. A linear gradient elution with methanol-water from 20% to 90% was used. The protein peak absorbance was measured at 280 nm, and the phospholipid fluorescence peak intensity was measured using an excitation wavelength of 485 nm and an emission wavelength of 520 nm. The peak areas were calculated using the trapezoidal integration method, and the integrated baseline was smoothed using a five-point method to eliminate matrix interference. A standard curve with concentrations of 0.1 to 1.0 mg / mL was prepared using bovine serum albumin standards. A standard curve with concentrations of 0.05 to 0.5 mg / mL was prepared using fluorescently labeled phosphatidylcholine as a phospholipid standard. The correlation coefficient of the standard curve was determined, with a threshold of 0.995. The sample matrix effect was corrected using the internal standard method, and the concentrations of each component in the sample were calculated from the calibration curve. A 4th-order wavelet filter was used to eliminate background noise from the detection signal. High-frequency interference was then filtered through a low-pass filter with a cutoff frequency of 20 Hz. The filtered data were fitted with a normal distribution function, and the chi-square test was used to determine the consistency of the data distribution. The mean protein and phospholipid component contents were obtained from the data distribution. The protein-to-phospholipid mass ratio was converted based on the component mean content. The molar concentration was calculated using a conversion factor of 65,000 for protein and 7500 for phospholipid. The calculated results were weighted by variance and the final protein-to-phospholipid molar ratio was obtained from the weighted average. Solid-phase extraction (SPE) used an 18-carbon octyl silica gel column to separate proteins and phospholipids. The hydrophobic surface of the silica gel interacts with the hydrophobic regions of the proteins, and the hydrophobic tails of the phospholipid molecules also bind to the stationary phase. The hydrophilicity of the components was altered by adjusting the methanol concentration in the eluent to achieve selective elution. Gradient elution began at 20% methanol and increased by 5% per minute, reaching 90% methanol in 90 minutes. Proteins eluted between 45% and 65% methanol, and phospholipids eluted between 70% and 85% methanol. The elution intervals of the two components differed by at least 5% to avoid peak overlap. A wavelength of 280 nm corresponds to the characteristic absorption of tryptophan residues in proteins. During fluorescence detection, excitation light at 485 nm caused electronic transitions in the phospholipid labeling groups, and emission light was collected at 520 nm. For standard curve preparation, five concentration points were selected: 0.1, 0.3, 0.5, 0.7, and 1.0 mg / mL for bovine serum albumin, and 0.05, 0.1, 0.2, 0.3, and 0.5 mg / mL for phosphatidylcholine. Three measurements were performed at each concentration point, and peak area versus concentration curves were plotted. A correlation coefficient of 0.995 indicated a good linear relationship between concentration and response. An internal standard was added at 20% of the sample concentration to correct for response deviations caused by matrix effects. A 4th-order wavelet filter corresponds to four decomposition levels. The signal is decomposed into different frequency bands through high-pass and low-pass filters. High-frequency noise is removed and a smooth signal is reconstructed. A 20 Hz low-pass filter removes detector electronic noise while preserving information about component concentration changes. A normal distribution fit is used to obtain the kurtosis and skewness coefficients. A chi-square test with a p-value greater than 0.05 confirms that the data conform to a normal distribution.The ratio of a protein molecular weight of 65,000 to a phospholipid molecular weight of 7500 is approximately 87. This coefficient converts the mass ratio into a molar ratio. In variance-weighted averaging, the weight is inversely proportional to the variance of the measurement, with greater weight given to data with smaller measurement errors. In actual testing, the protein-phospholipid molar ratio of fresh samples ranges from 3.5 to 4.5. This ratio decreases during storage due to protein degradation. The integrity of the fat globule membrane is closely related to the protein-phospholipid ratio; a stable ratio indicates a stable membrane structure. Accurate quantification of membrane components is achieved by combining solid-phase extraction with fluorescence detection.

[0031] S105. Obtain flavor characteristic evaluation data of different condensed milk samples, including flavor hierarchy scores, use the flavor hierarchy scores as sample label data, and use fat globule size distribution, surface curvature, membrane thickness, and membrane component ratio as sample characteristic data to construct a flavor hierarchy prediction model.

[0032] The characteristic data of the condensed milk sample are normalized by a normalizer to obtain normalized fat globule distribution values, surface curvature values, membrane thickness values and membrane component ratio values; the Pearson correlation coefficient is calculated based on the normalized characteristic values, and the feature contribution is used to delete the feature combinations with correlation coefficients exceeding the threshold; a flavor hierarchy label vector is constructed for the dataset after the features are deleted, and a stratified sampler is used to obtain training data and verification data; the feature importance value is calculated for the training data, and the features with a significance level lower than a preset threshold are selected by the forward stepwise regression method; a prediction model is established for the selected features, and the root mean square error and the determination coefficient are used to judge the accuracy of the prediction model on the verification data, and the prediction reliability level is determined according to the residual normality test results.

[0033] Specifically, the condensed milk sample feature data was normalized to the interval 0 to 1. The fat globule size distribution, surface curvature, membrane thickness, and membrane component ratio values were normalized using a normalizer. The Pearson correlation coefficient was used to calculate the correlation between features. Features with low contributions from feature combinations with correlation coefficients greater than 0.85 were removed. Missing value and outlier detection were performed on the dataset, and data quality scores were assigned from the test results. A label vector was constructed based on the flavor hierarchy score. A stratified sampler was used to partition the training and validation data into a 7:3 ratio. Random forest parameters were optimized using 10-fold cross-validation, with tree depths ranging from 5 to 15, the number of trees ranging from 50 to 200, and feature sampling ratios ranging from 0.6 to 0.9. A grid search was performed on the parameter combinations, and the validation set accuracy for each parameter combination was calculated. For the training data, feature importance was calculated using the Gini coefficient, and feature contributions were normalized. Features were selected using forward stepwise regression, with a significance level of less than 0.05 as the inclusion criterion. Features with a variance inflation factor greater than 5 were excluded. A prediction model was constructed based on these selected features, and model convergence was assessed using learning curves. Prediction operations were performed on the validation data, with root mean square error (RMS) and coefficient of determination used to assess prediction accuracy. Confidence intervals for predictions were calculated using residual normality tests. Samples with prediction residuals exceeding two standard deviations were flagged. The reliability of predictions was determined based on the data quality score, and resampled data was collected for low-reliability predictions. Normalization was performed to convert features of different dimensions to the same scale. Original values for the fat globule size distribution ranged from 2 to 8 μm, curvature values from 0.5 to 1.5, membrane thickness from 20 to 50 nm, and protein-phospholipid ratios from 3 to 5. All features were mapped to the range 0 to 1 after normalization to facilitate subsequent calculations of inter-feature correlations. The correlation coefficient reflects the degree of linear correlation between features, ranging from -1 to 1, with 0 indicating no correlation and larger absolute values indicating stronger correlation. The correlation coefficient between fat globule size and membrane thickness was 0.87, exceeding the threshold of 0.85. Calculation indicated that the membrane thickness feature contributed little to prediction and was therefore removed. Data quality was scored on a scale of 0 to 100, taking into account the proportion of missing values and the number of outliers. Scores below 60 required re-collection. Stratified sampling maintained a consistent ratio of samples in the training and validation sets, with 70% of the original dataset used as the training set and 30% as the validation set. 10-fold cross-validation assessed model stability by equally splitting the training data into 10 parts, with one part used for validation and the remaining for training. Tree depth controls random forest complexity; deeper trees are prone to overfitting, while shallower trees are underfitting. Optimal parameters were selected based on validation set accuracy. The Gini coefficient measures the reduction in feature impurity, reflecting the contribution of a feature to classification. After normalization, it ranges from 0 to 1. Forward stepwise regression starts with an empty feature set and adds the most significant features one at a time. The significance level of 0.05 is the standard threshold for statistical tests.A variance inflation factor greater than 5 indicates strong collinearity between features, affecting model stability. Learning curves show the trend of training and validation errors as sample size increases; convergence indicates a sufficient sample size. The root mean square error (RMS) reflects the average deviation between the predicted and true values, while the coefficient of determination (CDR) reflects the degree of explained variation in the model, ranging from 0 to 1. Residual normality tests determine whether the distribution of prediction errors is symmetrical and concentrated by calculating skewness and kurtosis. The standard deviation is used to set the prediction interval; residuals exceeding 2 standard deviations indicate that the prediction deviates significantly from the true value. The reliability of prediction results is categorized as high, medium, and low, based on a comprehensive assessment of the data quality score and prediction residuals. A low reliability rating indicates that the prediction is unreliable.

[0034] S106. Use a support vector machine algorithm to train the flavor hierarchy prediction model to obtain a trained flavor hierarchy prediction model, input the fat globule size distribution, surface curvature, membrane thickness, and membrane component ratio data of the condensed milk sample to be judged into the trained prediction model to obtain a flavor hierarchy prediction score for the sample.

[0035] Standard deviation normalization is performed on the characteristic data of the condensed milk sample to obtain normalized characteristic data, and a support vector machine is constructed using a radial basis kernel function to obtain a training model from the normalized characteristic data; a feature contribution value is calculated using the training model, a feature weight screening threshold is obtained based on the feature contribution value, and features below the feature weight screening threshold are eliminated and then trained to obtain a new support vector machine; a Mahalanobis distance value of the sample to be judged is calculated based on the new support vector machine, and if the Mahalanobis distance value exceeds the mean Mahalanobis distance threshold of the training sample, the sample to be judged is marked as an abnormal sample; a kernel density estimator is used to calculate a predicted score probability for the abnormal sample, and if the predicted score probability is less than a preset probability threshold, the abnormal sample is determined to be non-compliant.

[0036] Specifically, standard deviation normalization was performed on the characteristic data of the condensed milk samples, normalizing the values to an interval with a mean of 0 and a standard deviation of 1. The normality of the characteristic distribution was verified through skewness and kurtosis tests. A support vector machine was constructed using a radial basis kernel function. A grid search was performed with kernel function parameters ranging from 0.01 to 100 and penalty coefficients ranging from 0.1 to 1000. The search step size was set to exponential growth. The parameter combination was determined by the accuracy of the validation set. The validation accuracy and convergence rate were calculated for the training data. Support vector density statistics were performed on the trained support vector machine. A sequence-based minimum optimization searcher was used to extract the support vector set. The contribution of each feature to the prediction result was obtained using a mutual information calculator. The feature weight screening threshold was set to a maximum contribution value of 20%. Features below the threshold were removed, and a new support vector machine was trained using the remaining features. The degree of convergence of the new support vector machine was determined from the accuracy curve. Standard deviation normalization was performed on the condensed milk sample data to be judged. A single-sample Mahalanobis distance calculator was used to obtain sample feature distances. The discrimination threshold was set at two standard deviations of the mean Mahalanobis distance of the training samples. Samples with distance values exceeding the threshold were marked as abnormal, and prediction scores for normal samples were obtained using a support vector machine prediction algorithm. A confidence level was calculated based on the prediction error distribution of the training data, with a 95% confidence level used to determine the upper and lower bounds of the prediction score range. The prediction errors were tested for normality using the Shapiro normality test. Kernel density estimation was used to calculate the probability that the predicted score fell within the historical data distribution. Prediction scores with a probability of less than 5% were flagged, and the final prediction results were output. Standard deviation normalization converted the different feature data to a comparable scale. The original features had a mean of 5 microns and a standard deviation of 1.5 microns for fat globule size, a mean of 0.8 and a standard deviation of 0.2 for surface curvature, a mean of 35 nanometers and a standard deviation of 8 nanometers for membrane thickness, and a mean of 4.2 and a standard deviation of 0.6 for protein-to-phospholipid ratio. After normalization, all features had a mean of 0 and a standard deviation of 1. Skewness coefficients between -0.5 and 0.5, and kurtosis coefficients between 2.5 and 3.5, indicate that the data are close to a normal distribution. The radial basis kernel function constructs a nonlinear classification boundary in the feature space. The kernel parameter controls the complexity of the mapping space. A too small value leads to underfitting, while a too large value leads to overfitting. A grid search was performed starting at 0.01 and increasing 10-fold to 100, with a total of six kernel parameter values selected. The penalty coefficient balances empirical risk and structural risk, starting at 0.1 and increasing 10-fold to 1000, with a total of five penalty coefficient values selected. The validation set accuracy was optimized with a kernel parameter of 1.0 and a penalty coefficient of 10. Support vectors are sample points located near the classification boundary; the number of support vectors reflects the complexity of the classification boundary. Approximately 15% of the points in the training sample serve as support vectors. Mutual information calculations revealed contributions of 0.45 to the fat globule size distribution, 0.25 to the surface curvature, 0.18 to the membrane thickness, and 0.12 to the protein-phospholipid ratio. By setting the weight threshold to 0.09 and removing the protein-phospholipid ratio feature, the number of support vectors in the new model was reduced to 12%.The Mahalanobis distance measures the standardized distance between a sample point and the data center, taking into account inter-feature correlation. The mean Mahalanobis distance for the training sample was 2.8, the standard deviation was 0.9, and the anomaly threshold was set at 4.6. For the predicted scores, the 95% confidence interval corresponds to 1.96 times the standard deviation. The Shapiro test p-value for the prediction error greater than 0.05 indicates a normal distribution. Kernel density estimation constructs a distribution curve for historical data. A predicted score with a probability of less than 5% generally indicates unreliable results. In practical applications, the distribution characteristics of flavor level scores vary among condensed milk samples of different qualities. Fresh samples exhibit more stable scores, while samples with longer storage experience greater fluctuations. A support vector machine (SVM) is used to establish a mapping between features and scores, enabling predictive evaluation of flavor level. Model parameters must be optimized based on sample characteristics and scoring criteria to ensure reliable predictions. A comprehensive feature extraction and prediction process plays a crucial role in evaluating condensed milk quality.

[0037] S107. Sort the flavor levels of the samples according to the flavor level prediction scores to obtain a ranking result of the samples from high to low flavor levels. Based on the ranking result, determine the combination of fat globule size distribution, surface curvature, membrane thickness, and membrane component ratio corresponding to the sample with the highest flavor level as the fat globule morphological characteristic combination with the highest flavor characteristic level.

[0038] The condensed milk samples are sorted in descending order according to their predicted scores for flavor hierarchy, and the Mann-Whitney test is used to obtain the sample difference significance results, which are used to mark the sample ranking levels; the Ward minimum variance clustering method is used to calculate the Euclidean distance for the top samples in the ranking levels, a distance matrix is constructed using the Euclidean distance, a cluster pedigree diagram is obtained from the distance matrix, and a representative sample set is obtained from the cluster pedigree diagram through silhouette coefficient analysis; principal component analysis is performed on the representative sample set, and principal components are obtained through eigenvalue decomposition of the covariance matrix. The principal components are subjected to Schmidt orthogonalization to obtain orthogonal basis vectors, and the orthogonal basis vectors are used to construct a feature space; the Euclidean distance between sample points is calculated in the feature space, and a nearest neighbor sample set is constructed for the sample point whose Euclidean distance is less than a threshold value. The most densely populated area sample points are obtained from the most densely populated area sample points through density clustering, and the most densely populated area sample points are used to determine the optimal morphological feature combination value.

[0039] Specifically, the predicted flavor levels of condensed milk samples were sorted in descending order. The Mann-Whitney test was used to determine the significance of differences between adjacent sample scores based on the 95% confidence interval of the predicted scores, with a test level of 0.05. Samples with significant differences were marked with a ranking rank. The ranking stability index was calculated using bootstrap repeated sampling, and reliable ranking results were obtained from samples with a stability index greater than 0.9. Ward minimum variance clustering was used for the top-ranked samples. The Euclidean distances between samples were calculated to construct a distance matrix. A cluster pedigree diagram was drawn based on the distance matrix. The optimal number of clusters was determined using the silhouette coefficient. A stability test was performed on the clustering results, with a silhouette coefficient threshold of 0.7. A representative sample set was extracted from the cluster center. Principal component analysis was performed on the representative sample set. Principal components were extracted through eigenvalue decomposition of the covariance matrix. A cumulative contribution threshold of 85% was set. Principal components that met the threshold were selected to construct a feature space. Schmidt orthogonalization was used to obtain orthogonal basis vectors, and feature combinations were obtained by projecting the original features onto the orthogonal basis. Based on the feature combinations, a four-dimensional feature vector space was constructed. Within this feature space, Euclidean distances between sample points were calculated. A distance threshold of 20% of the maximum distance was set. Sample points with distances less than the threshold were selected to form the nearest neighbor set. Density clustering was then used to identify the densest areas of sample points, and the optimal morphological feature combination value was obtained from these densest areas. Ranking of predicted scores involves assessing the comparability of scores. The 95% confidence interval (CI) represents the range of fluctuation in the predicted values. For example, a predicted score of 8.5 has a CI of 8.2 to 8.8, which overlaps with the CI of 8.7 to 9.3 for a neighboring sample score of 9.0. This requires a Mann-Whitney test to determine the significance of the differences. A test level of 0.05 corresponds to a critical value of 1.96, with differences above this threshold considered significant. A bootstrap method was used to assess the stability of the ranking by repeated sampling. 1000 random samples were drawn from the original data, and the sample ranking position was calculated each time. The frequency of each sample appearing at a specific ranking position was then counted. A stability index of 0.9 indicates that 90% of the sample results yielded the same ranking position, reflecting the reliability of the ranking results. Samples ranked high tend to have higher stability indices. Ward minimum variance clustering groups samples based on inter-sample distance. The Euclidean distance calculation considers all feature dimensions, and the eigenvalues are normalized to ensure equal weighting across dimensions. The cluster pedigree diagram displays the hierarchical relationships between samples, with closer samples clustering earlier. The silhouette coefficient measures the tightness of clustering; a value of 0.7 indicates that samples within a cluster are more similar than samples between clusters. Principal component analysis reduces the dimensionality of the feature space. The covariance matrix reflects the correlation between features, and the eigenvalues indicate the extent to which the principal components explain the variation in the original features. A cumulative contribution of 85% means that the retained principal components contain 85% of the information in the original data. Schmidt orthogonalization eliminates correlation between principal components and constructs a standard orthogonal basis. The four-dimensional eigenvector contains information on fat globule size distribution, surface curvature, membrane thickness, and membrane component ratios. The Euclidean distance threshold is set at 20% of the maximum distance to ensure that sufficiently similar sample points are selected.Density clustering searches for densely populated areas of sample points in Euclidean space. Feature combinations in densely populated areas are highly representative. In practice, samples with a high degree of flavor depth typically possess stable feature combinations, such as concentrated fat globule size distribution, uniform surface curvature, moderate membrane thickness, and a balanced protein-phospholipid ratio. Multidimensional feature analysis identifies the optimal feature combination and establishes a correspondence between flavor depth and morphological characteristics. Statistical tests and stability assessments ensure the reliability of the results, while feature space dimensionality reduction and orthogonalization enhance the representativeness of the feature combinations.

[0040] The above disclosure is only a preferred embodiment of the present invention, and it is certainly not intended to limit the scope of the present invention. A person skilled in the art can understand that all or part of the processes of the above embodiment and equivalent changes made in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. A method for optimizing condensed milk flavor by multi-level flavor characteristic analysis, characterized in that: The method comprises: Obtain fat globule image data of different condensed milk samples, extract the fat globule area using an image segmentation algorithm, obtain a segmented fat globule binary image, calculate the equivalent diameter of the fat globule based on the pixel area of the fat globule in the binary image, and obtain the fat globule size distribution data; For the fat globule binary image after segmentation, the edge detection algorithm is used to extract the contour of the fat globule to obtain the contour curve of the fat globule. According to the curvature change of the contour curve, the average curvature of the fat globule surface is calculated to obtain the curvature data representing the shape of the fat globule; The morphological processing algorithm is used to extract the membrane area of the fat globules from the segmented binary image. The average membrane thickness of the fat globules is calculated based on the number of pixels in the extracted membrane area to obtain the membrane thickness data that characterizes the fat globule membrane characteristics. Obtaining fat globule membrane component test data of different condensed milk samples, including membrane protein content and phospholipid content, and calculating the ratio of protein to phospholipid in the fat globule membrane based on the membrane protein content and phospholipid content, thereby obtaining protein-phospholipid ratio data that characterizes the composition of the fat globule membrane; Obtain flavor characteristic evaluation data for different condensed milk samples, including flavor hierarchy scores. Use the flavor hierarchy scores as sample label data, and use fat globule size distribution, surface curvature, membrane thickness, and membrane component ratio as sample feature data to construct a flavor hierarchy prediction model. A flavor hierarchy prediction model is trained using a support vector machine algorithm to obtain a trained flavor hierarchy prediction model. The fat globule size distribution, surface curvature, membrane thickness, and membrane component ratio data of the condensed milk sample to be judged are input into the trained prediction model to obtain a flavor hierarchy prediction score for the sample. According to the flavor hierarchy prediction scores, the flavor hierarchy of the samples was ranked to obtain the sample ranking results from high to low flavor hierarchy. Based on the ranking results, the combination of fat globule size distribution, surface curvature, membrane thickness, and membrane component ratio corresponding to the sample with the highest flavor hierarchy was determined as the fat globule morphological feature combination with the highest flavor characteristic level.

2. The method according to claim 1, characterized in that The method comprises obtaining fat globule image data of different condensed milk samples, extracting fat globule regions from the fat globule images using an image segmentation algorithm, obtaining segmented fat globule binarization images, and calculating equivalent diameters of the fat globules based on the pixel areas of the fat globules in the binarization images to obtain size distribution data of the fat globules, including: An image processor is used to obtain image data of a condensed milk sample with a calibration plate scale, and a conversion relationship between pixel points and actual size is established according to the calibration plate scale to obtain a calibration image; Performing Gaussian low-pass filtering on the calibration image to remove background noise, using the OTSU algorithm to calculate multi-level grayscale thresholds, and extracting image contour boundaries according to the grayscale thresholds to obtain a boundary image; A seed point growth process is performed based on the boundary image. If the grayscale value difference in the overlapping area exceeds the standard deviation of the neighborhood average, a watershed algorithm is used to divide the area into independent blocks to obtain a segmented image. The fat globule contour area data is extracted from the segmented image, the actual area value is calculated according to the conversion relationship between the pixel point and the actual size, and the particle size distribution result is obtained by calculating the equivalent diameter using the circular area formula.

3. The method according to claim 1, characterized in that The method comprises the following steps: extracting the contour of the fat globule from the segmented binary image of the fat globule using an edge detection algorithm to obtain a contour curve of the fat globule; calculating the average curvature of the fat globule surface according to the curvature change of the contour curve, and obtaining curvature data characterizing the shape of the fat globule, including: Using the Sobel operator to extract edge gradients in the horizontal and vertical directions of the binary image, and obtaining an edge intensity distribution map through iterative calculation based on the horizontal gradient and the vertical gradient; Performing a refinement iterative operation on the edge intensity distribution map, and connecting and compensating unclosed points through a connectivity propagator to obtain a closed contour line; Constructing a point index sequence according to the closed contour line, calculating a curvature value of the point index sequence using a three-point arc fitting function, and obtaining a smooth curvature value sequence through a Gaussian filter; If the curvature of a point in the smooth curvature value sequence is greater than zero, it is marked as a convex point. If the curvature of a point is less than zero, it is marked as a concave point. The average curvature result is obtained by weighted operation of the extreme point curvatures.

4. The method according to claim 1, wherein The segmented fat globule binary image is extracted by using a morphological processing algorithm to extract the membrane area of the fat globule, and the average membrane thickness of the fat globule is calculated according to the number of pixels in the extracted membrane area to obtain membrane thickness data characterizing the fat globule membrane characteristics, including: A circular structure erosion operator is used to perform edge erosion on the fat globule region of the binary image, and a membrane region map is obtained through a closing operation; According to the maximum gray value point in the membrane region map as a seed point, a region growing extractor is used to expand to the surrounding areas, and a skeleton line of the membrane region is obtained by distance transformation; Scanning from the skeleton line of the membrane area to the inner and outer sides to obtain the inner and outer boundary coordinate points of the membrane area contour; A local film thickness sequence is obtained by measuring the inner and outer boundary coordinate points along the reference direction. If the sampling interval of the local film thickness sequence is a preset value, an average film thickness value is obtained according to Gaussian weighted averaging and Kalman filtering.

5. The method according to claim 1, wherein The method comprises obtaining fat globule membrane component detection data of different condensed milk samples, including membrane protein content and phospholipid content, and calculating the ratio of protein to phospholipid in the fat globule membrane according to the membrane protein content and phospholipid content, to obtain protein-phospholipid ratio data characterizing the composition of the fat globule membrane, including: A methanol-water gradient eluent is used to separate the components through a solid phase extraction column, and the component separation obtains a protein absorbance signal and a phospholipid fluorescence intensity signal; performing baseline smoothing processing on the protein absorbance signal and the phospholipid fluorescence intensity signal, and obtaining corrected signal data by internal standard correction; The corrected signal data is filtered using a wavelet filter and a low-pass filter, and the component content value is obtained by fitting a normal distribution function; The molar concentrations were calculated using protein and phospholipid molecular weight conversion coefficients for the component contents, and the protein-phospholipid molar ratios were obtained by variance-weighted averaging.

6. The method according to claim 1, characterized in that The method comprises obtaining flavor characteristic evaluation data of different condensed milk samples, including flavor hierarchy scores, using the flavor hierarchy scores as label data of the samples, using fat globule size distribution, surface curvature, membrane thickness, and membrane component ratio as characteristic data of the samples, and constructing a flavor hierarchy prediction model, comprising: The characteristic data of condensed milk samples were normalized by a normalizer to obtain normalized fat globule distribution values, surface curvature values, membrane thickness values and membrane component ratio values; The Pearson correlation coefficient is calculated based on the normalized eigenvalues, and the feature contribution is used to delete the feature combinations whose correlation coefficient exceeds the threshold; A flavor hierarchy label vector is constructed for the dataset after feature deletion, and a stratified sampler is used to obtain training data and validation data. Calculating feature importance values for the training data, and selecting features whose significance levels are lower than a preset threshold using a forward stepwise regression method; A prediction model is established based on the selected features, and the root mean square error and determination coefficient are used to judge the accuracy of the prediction model on the validation data. The prediction reliability level is determined based on the residual normality test results.

7. The method according to claim 1, characterized in that The flavor hierarchy prediction model is trained using a support vector machine algorithm to obtain a trained flavor hierarchy prediction model, and the fat globule size distribution, surface curvature, membrane thickness, and membrane component ratio data of the condensed milk sample to be judged are input into the trained prediction model to obtain a flavor hierarchy prediction score for the sample, including: performing standard deviation normalization on the characteristic data of the condensed milk sample to obtain normalized characteristic data, and constructing a support vector machine using a radial basis kernel function to obtain a training model from the normalized characteristic data; Calculate the feature contribution value through the training model, obtain the feature weight screening threshold according to the feature contribution value, remove the features below the feature weight screening threshold, and then train a new support vector machine; Calculating the Mahalanobis distance value of the sample to be judged according to the new support vector machine, and marking the sample to be judged as an abnormal sample if the Mahalanobis distance value exceeds the mean threshold of the Mahalanobis distance of the training samples; A kernel density estimator is used to calculate the predicted score probability for the abnormal sample. If the predicted score probability is less than a preset probability threshold, the abnormal sample is determined to be unqualified.

8. The method according to claim 1, characterized in that The flavor layering of the samples is sorted according to the flavor layering prediction scores to obtain a sample sorting result from high to low flavor layering, and according to the sorting result, a combination of fat globule size distribution, surface curvature, membrane thickness, and membrane component ratio corresponding to the sample with the highest flavor layering is determined as the fat globule morphological feature combination with the highest flavor characteristic level, including: Sorting the condensed milk samples in descending order according to their predicted flavor levels, and obtaining a significant difference between the samples using a Mann-Whitney test. The significant difference between the samples is used to mark the ranking of the samples. For the top samples in the ranking, the Ward minimum variance clustering method is used to calculate the Euclidean distance, a distance matrix is constructed using the Euclidean distance, a cluster pedigree diagram is obtained from the distance matrix, and a representative sample set is obtained by analyzing the cluster pedigree diagram through the silhouette coefficient; Performing principal component analysis on the representative sample set, obtaining principal components by eigenvalue decomposition of the covariance matrix, obtaining orthogonal basis vectors by Schmidt orthogonalization of the principal components, and using the orthogonal basis vectors to construct a feature space; The Euclidean distance between sample points is calculated in the feature space, and a nearest neighbor sample set is constructed for the sample points whose Euclidean distance is less than a threshold. The most densely populated area sample points are obtained from the nearest neighbor sample set through density clustering, and the most densely populated area sample points are used to determine the optimal morphological feature combination value.

Images