A method for detecting the content of pigskin collagen

Abstract

The application relates to the technical field of optical detection, and discloses a pigskin collagen content detection method. Before contacting a detection medium, the initial optical profile of a pigskin sample is established. According to the profile, a special detection path for the sample is generated, and the moving track and the stay node of the probe sampling are specified. The medium is injected at the starting point of the path, and a synchronization mechanism is started to synchronize the probe movement with the medium diffusion front. When the probe reaches the terminal node, the spectral sequence of each node on the path is recorded completely, and the sequence contains the dynamic characteristic absorption information of collagen. The spectral sequence is subjected to cross-node correlation fusion to generate the fusion spectral feature map of the sample. The map is compared with a pre-stored feature-content mapping model, and the collagen content value is output. The application realizes adaptive sampling and dynamic process tracking, and improves the accuracy and reliability of detection.

Description

Technical Field

[0001] This invention relates to the field of optical detection technology, specifically to a method for detecting collagen content in pig skin. Background Technology

[0002] Current techniques for detecting collagen content in pigskin commonly employ near-infrared spectroscopy or other optical methods. These methods typically involve pre-setting fixed measurement points or regular grid paths on the sample surface for spectral acquisition. Regardless of variations in sample surface color uniformity, hair distribution, or thickness, the detection probe operates according to a pre-defined, uniform pattern. This static sampling approach treats each sample as an object with uniform optical properties, ignoring the inherent spatial heterogeneity of pigskin as a biological tissue.

[0003] Fixed sampling points may be located in areas of scarring, pigmentation, or excessive stratum corneum, where spectral signals are strongly interfered with, failing to effectively penetrate and reflect the true information of subcutaneous collagen. Furthermore, in detections involving the detection medium, the diffusion of the medium within pig skin tissue is a dynamic process that changes over time. Existing techniques typically measure at a fixed time point after medium injection or in a stable state after complete medium infiltration. This results in spectral information acquired either from the initial stage before the reaction has fully progressed or from a homogenized state after the reaction has reached equilibrium, failing to capture the most characteristic dynamic spectral changes during the interaction between collagen and the medium. Therefore, a detection method is needed that can adapt to individual sample differences and capture key dynamic reaction information. Summary of the Invention

[0004] The purpose of this invention is to provide a method for detecting collagen content in pig skin, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for detecting collagen content in pig skin, the method comprising:

[0006] Before the pigskin sample comes into contact with the detection medium, an initial optical profile of the pigskin sample is established. The initial optical profile is composed of the surface reflectance distribution and the internal transmittance distribution of the sample.

[0007] Based on the initial optical profile, a dedicated detection path is generated for the pigskin sample. The dedicated detection path defines the movement trajectory and dwell points of the detection probe on the sample surface for spectral sampling.

[0008] The detection medium is injected at the starting node of the dedicated detection path, and the path following mechanism is activated simultaneously to keep the movement of the detection probe synchronized with the diffusion front of the detection medium.

[0009] When the detection probe reaches the end node of the dedicated detection path, the spectral sequence collected at all the stopping nodes on the dedicated detection path is completely recorded. The spectral sequence contains the characteristic absorption information of collagen at different nodes.

[0010] Cross-node correlation fusion is performed on the spectral sequence to generate a fused spectral feature map of the pigskin sample;

[0011] The fused spectral feature map is compared with the pre-stored feature-content mapping model to output the collagen content value of the pig skin sample.

[0012] Preferably, after establishing the initial optical profile of the pigskin sample, the method further includes the following steps:

[0013] By analyzing the spatial gradient changes of reflectance and transmittance in the initial optical profile, high heterogeneity regions and low heterogeneity regions on the surface of the pigskin sample were identified.

[0014] In the planning of the dedicated detection path, the movement trajectory is made to preferentially traverse highly heterogeneous regions and reduce the density of dwelling nodes in low heterogeneous regions.

[0015] Based on the identified regional characteristics, an independent sampling integration time is assigned to each stop node on the dedicated detection path, with nodes in highly heterogeneous regions being assigned a longer sampling integration time.

[0016] Preferably, the step of injecting the detection medium at the starting node of the dedicated detection path and simultaneously activating the path following mechanism specifically includes the following steps:

[0017] The actual diffusion rate and direction of the detection medium on the surface of pigskin samples were monitored.

[0018] Based on the actual diffusion rate and diffusion direction, the moving speed of the detection probe and the local orientation of the dedicated detection path are dynamically fine-tuned to ensure that the detection probe is always located in the area where the detection medium has been fully impregnated for spectral sampling.

[0019] The dynamic fine-tuning is based on a pre-established medium diffusion model, which relates the influence of pigskin texture density and environmental temperature and humidity on the diffusion process.

[0020] Preferably, after fully recording the spectral sequences collected at all stopping nodes along the dedicated detection path, the method further includes the following steps:

[0021] Check the continuity of the spectral sequence in the time dimension. If a spectral line interruption is found due to probe movement or signal fluctuation, the spectral line repair process is initiated.

[0022] The spectral line repair process calls the optical information of the corresponding spatial position in the initial optical profile, and performs interpolation reconstruction on the spectral data at the interruption point based on the trend of the spectral data of adjacent nodes.

[0023] Preferably, the cross-node correlation fusion of the spectral sequence specifically includes the following steps:

[0024] A spatial correlation weight is assigned to the spectral data collected at each stop node on the dedicated detection path. The spatial correlation weight is calculated based on the reciprocal of the Euclidean distance between the stop node and all other nodes on the path.

[0025] Using the spatial correlation weights, the spectral data of all nodes are weighted and averaged to generate a preliminary fused spectrum;

[0026] The common background spectrum caused by non-collagen components of the pigskin sample itself is subtracted from the preliminary fusion spectrum. The common background spectrum is established by analyzing the spectral data of collagen-deficient regions in the historical sample library.

[0027] The result after deducting common background spectra is used as the fused spectral feature map.

[0028] Preferably, the following steps are also included:

[0029] After generating the fused spectral feature map, the spectral signal-to-noise ratio and feature peak integrity index of the fused spectral feature map are calculated;

[0030] If the signal-to-noise ratio of the spectrum is lower than a preset threshold or the feature peak integrity index shows that key features are missing, a re-examination instruction is triggered.

[0031] The re-inspection command controls the detection equipment to resample a specific sub-region of the pigskin sample according to a supplementary detection path, wherein the supplementary detection path is determined by the region with the most blurred information in the initial optical profile.

[0032] The calculation of the spectral signal-to-noise ratio and characteristic peak integrity index of the fused spectral feature map specifically includes:

[0033] Multiple calculation points are randomly selected within the baseline stable region of the fused spectral feature map, and the average signal intensity of the multiple calculation points is calculated as the reference noise level;

[0034] Identify the characteristic absorption peaks belonging to collagen in the fused spectral feature map, and determine the peak intensity and full width at half maximum (FWHM) of each characteristic absorption peak;

[0035] The ratio of the peak intensity of the characteristic absorption peak to the reference noise level is used as the signal-to-noise ratio of the characteristic absorption peak.

[0036] The average signal-to-noise ratio of all the characteristic absorption peaks is taken as the signal-to-noise ratio of the spectrum;

[0037] Count the number of characteristic absorption peaks actually identified in the fused spectral feature map;

[0038] Obtain the theoretical total number of characteristic absorption peaks that a predefined collagen must contain;

[0039] The ratio of the number of actual identified characteristic absorption peaks to the theoretical total number is used as the characteristic peak integrity index.

[0040] Preferably, after the re-inspection command controls the detection device to resample a specific sub-region of the pigskin sample according to a supplementary detection path, the method further includes the following steps:

[0041] The resampled spectral data is then replaced with the data of the corresponding spatial region in the fused spectral feature map.

[0042] The new spectrum after replacement is subjected to cross-node correlation fusion and common background spectrum subtraction steps again to generate an updated fused spectral feature map;

[0043] The fused spectral feature maps before and after the update are simultaneously input into the feature-content mapping model, and the average of the two outputs is taken as the final collagen content value.

[0044] Preferably, the method for establishing the pre-stored feature-content mapping model includes the following steps:

[0045] A large number of pig skin standard samples with known precise collagen content were collected. An initial optical profile was established for each standard sample, and spectral sequences were acquired according to its dedicated detection path.

[0046] The cross-node correlation fusion processing is performed on the spectral sequences of all standard samples to construct a standard fused spectral feature map library;

[0047] A nonlinear iterative optimization algorithm is used to find the optimal mapping function between the standard fused spectral feature map library and known collagen content values. This optimal mapping function is then solidified into the feature-content mapping model, specifically including:

[0048] The structure of the optimal mapping function is defined as a network model with multi-layer nonlinear transformations;

[0049] All spectral data in the standard fusion spectral feature map library are used as the input vector of the network model, and the corresponding known collagen content value is used as the target output.

[0050] Assign initial weight parameters and bias parameters to each layer of the network model;

[0051] The forward computation process is performed, passing the input vector layer by layer through the network model to obtain the predicted content value under the current parameters;

[0052] Calculate the total error between the predicted content value and the known collagen content value;

[0053] Based on the total error, the weight parameters and bias parameters of each layer of the network model are adjusted through a backpropagation process.

[0054] Repeat the forward calculation process and the backward propagation process until the change in the total error is less than the preset convergence threshold or the maximum number of iterations is reached;

[0055] The final stabilized network model and all its parameters are then solidified into the feature-content mapping model.

[0056] Preferably, after outputting the collagen content value of the pigskin sample, the method further includes the following steps:

[0057] A detection record is generated based on the dedicated detection path used in this test, the actual diffusion rate, and the final output content value.

[0058] The detection records are added to a dynamic experience base, which is used to optimize the generation rules for dedicated detection paths and the parameters of the medium diffusion model for subsequent samples.

[0059] Preferably, the method further includes the following steps:

[0060] When a new pigskin sample detection task is received, its initial optical profile is matched with the initial optical profiles of all records in the dynamic experience base. If a matching existing record with a similarity exceeding the threshold is found, the dedicated detection path scheme and estimated diffusion rate stored in the existing record are directly used as the initialization parameters for the new sample detection, and the path planning calculation process is skipped.

[0061] Compared with the prior art, the beneficial effects of the present invention are:

[0062] By acquiring the initial optical profile of the pigskin sample and generating a dedicated detection path accordingly, the detection system can identify and avoid areas on the sample surface with optical defects or severe interference. The path planning algorithm guides the probe to a preset stop node with more uniform surface optical properties and better internal transmittance for measurement. This process transforms spectral acquisition from passively receiving information from all areas of the sample to actively selecting areas with better optical windows, thereby ensuring that the acquired raw spectral data has a higher signal-to-noise ratio and stronger representativeness, reducing measurement errors and result fluctuations caused by sample surface inhomogeneity.

[0063] By synchronizing the movement of the detection probe with the diffusion front of the detection medium, spectral acquisition is precisely positioned at the tissue front region where the medium is infiltrating but has not yet reached equilibrium. In this region, the molecular structure of collagen is interacting with the medium, and its optical properties are in a rapidly changing transient state. The synchronization mechanism ensures that the spectra acquired at each dwell point capture the most sensitive reaction information in this dynamic process. Compared to spectra measured at a static endpoint, these dynamic spectral sequences acquired along the diffusion sequence contain richer reaction kinetics features, allowing subsequent feature maps generated based on cross-node correlation fusion to reveal collagen content and state information more profoundly, improving the model's analytical capability and prediction accuracy. Attached Figure Description

[0064] Figure 1 This is a schematic diagram illustrating the working principle of the method for detecting collagen content in pig skin according to the present invention.

[0065] Figure 2 This is a flowchart for optimizing the detection path based on the initial optical profile;

[0066] Figure 3 A flowchart for cross-node correlation fusion;

[0067] Figure 4 A comparison of the spectral characteristics of the original and updated porcine skin collagen fusion.

[0068] Figure 5 Comparison chart showing the effects of optimizing the testing process. Detailed Implementation

[0069] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] Please see Figure 1This invention provides a method for detecting collagen content in pig skin. The method includes: in the initial detection stage, acquiring the reflectance distribution of the surface and the internal transmittance distribution of the pig skin sample using an optical scanning system, and combining them to form an initial optical profile of the sample. Based on the spatial characteristics of the initial optical profile, a dedicated detection path covering key areas of the sample is generated. This path consists of a series of spectral sampling and dwell nodes and their connecting trajectories. The path planning must ensure that the probe can capture representative information of the spatial distribution of collagen. At the starting node of the path, a detection medium that specifically binds to collagen is quantitatively injected, and a path following mechanism is activated to synchronize the movement speed of the detection probe with the diffusion front of the medium in the pig skin tissue. When the probe reaches the ending node of the path, spectral sequences at all dwell nodes are acquired. The sequence data contains intensity information of characteristic absorption peaks of collagen. The spectral sequences are integrated into a fused spectral feature map using a cross-node correlation fusion algorithm. Finally, this map is input into a pre-trained feature-content mapping model to output the collagen content value.

[0071] In one embodiment of the present invention, see [reference] Figure 2 After establishing the initial optical profile of the pigskin sample, the spatial gradient changes of reflectance and transmittance within this profile were analyzed to identify high-heterogeneity and low-heterogeneity regions. When planning a dedicated detection path, the movement trajectory was prioritized to traverse high-heterogeneity regions, while reducing the density of stopping nodes in low-heterogeneity regions. Based on the degree of regional heterogeneity, an independent sampling integration time was allocated to each stopping node, with longer sampling integration times allocated to nodes in high-heterogeneity regions to improve the signal-to-noise ratio. Simultaneously with the injection of the detection medium at the path's starting node, a path-following mechanism was activated. This mechanism dynamically fine-tunes the probe's movement speed and local path orientation by real-time monitoring of the actual diffusion rate and direction of the medium on the pigskin surface, ensuring that the probe is always located within the fully wetted medium area for sampling. The medium diffusion model upon which this dynamic fine-tuning relies is linked to the pigskin texture density and environmental temperature and humidity parameters.

[0072] In practical implementation, after establishing the initial optical profile of the pigskin sample, the optical profile analysis module analyzes the spatial gradient changes of reflectance and transmittance. For example, it quantifies local heterogeneity by calculating the standard deviation of reflectance in the neighborhood of each pixel. Regions with a standard deviation of reflectance higher than a set threshold of 0.15 are identified as high heterogeneity regions, while regions with a standard deviation of reflectance lower than 0.05 are identified as low heterogeneity regions. In a specific detection of a back skin sample, the analysis showed that the standard deviation of reflectance in the areas surrounding pores and wrinkles was between 0.18 and 0.25, and was marked as a high heterogeneity region, while the standard deviation of reflectance in the smooth area in the middle of the epidermis was approximately 0.03, and was marked as a low heterogeneity region. When planning a dedicated detection path, the path generation algorithm prioritizes the movement trajectory through high heterogeneity regions based on the region identification results. For example, it plans a broken line trajectory that sequentially connects three high heterogeneity regions, and reduces the density of stopping nodes in the low heterogeneity regions it passes through, increasing the node spacing from the standard 2 mm to 5 mm. Based on the identified regional characteristics, the control software assigns an independent sampling integration time to each stop node on the dedicated detection path. In a specific operation, nodes located in high heterogeneity regions are assigned 100 milliseconds of sampling integration time, while nodes located in low heterogeneity regions are assigned 40 milliseconds of sampling integration time.

[0073] In some embodiments, a detection medium is injected at the starting node of the dedicated detection path, and a path-following mechanism is simultaneously activated. This mechanism monitors the actual diffusion rate and direction of the detection medium on the surface of the pigskin sample through a real-time image analysis unit. For example, under ambient temperature of 25 degrees Celsius and relative humidity of 60%, the diffusion rate of Sirius Red solution along the pigskin texture direction is monitored to be 0.8 mm / s, and the diffusion rate perpendicular to the texture direction is 0.5 mm / s. Based on the actual diffusion rate and direction, the path-following controller dynamically fine-tunes the moving speed of the detection probe and the local orientation of the dedicated detection path. For example, when the probe moves along the texture direction, the speed is fine-tuned to 0.75 mm / s; when it needs to traverse the texture laterally, the speed is fine-tuned to 0.45 mm / s, and the path orientation is slightly adjusted to follow the wetting front of the medium, ensuring that the detection probe is always located in the area where the detection medium has been fully wetting for spectral sampling.

[0074] It is understandable that the dynamic fine-tuning is based on a pre-established medium diffusion model. This model correlates the influence of pigskin texture density and environmental temperature and humidity on the diffusion process. The specific implementation of the medium diffusion model is based on the analysis of historical detection data. This model uses a multivariate nonlinear regression algorithm to determine key parameters, thereby integrating factors such as the texture density of the pigskin sample, environmental temperature, and humidity into a predictive framework. During model construction, a large amount of actual observation data on medium diffusion under different texture densities, temperature, and humidity conditions was collected. Model coefficients were obtained through fitting and optimization, enabling the model to accurately reflect the comprehensive influence of these variables on the diffusion rate. In actual detection applications, the control system reads the values ​​from the environmental temperature and humidity sensors in real time, and simultaneously quantifies and extracts texture density information per unit area from the initial optical profile of the pigskin sample. Inputting this real-time data into the medium diffusion model allows for the calculation of an estimated diffusion coefficient. The model obtains the expression for the diffusion coefficient D through fitting historical data:

[0075] ;

[0076] in: Represents the diffusion coefficient of the medium. This represents the texture density per unit area on the surface of a pigskin sample. Indicates ambient temperature. Indicates the relative humidity of the environment. , , These are the model coefficients determined through multivariate nonlinear regression. In practical implementation, the control system collects real-time temperature and humidity sensor data of the current environment, extracts texture density information from the initial optical profile, and substitutes it into the medium diffusion model to calculate the estimated diffusion coefficient, which serves as the basic parameter for dynamically fine-tuning the probe speed.

[0077] In one embodiment of the present invention, see [reference] Figure 3After acquiring the spectral sequences of all nodes along the dedicated detection path, the continuity of the spectral sequences in the time dimension is checked. If spectral line interruptions are found due to probe movement jitter or signal fluctuations, a spectral line repair process is initiated. This process calls upon the optical information of the corresponding spatial position in the initial optical profile and performs interpolation reconstruction at the interruption point based on the trend of the spectral data of adjacent nodes. During cross-node correlation fusion, a spatial correlation weight is assigned to the spectral data of each node. The weight value is calculated based on the reciprocal of the Euclidean distance between the node and all other nodes on the path. The weights are used to perform a weighted average of the spectral data of all nodes to generate a preliminary fused spectrum. The common background spectrum caused by non-collagen components of pigskin itself is then subtracted. This background spectrum is established by analyzing the spectral data of collagen-deficient regions in the historical sample library, ultimately yielding a fused spectral feature map. In practice, after fully recording the spectral sequences collected at all nodes along the dedicated detection path, the spectral sequence integrity verification module checks the continuity of the spectral sequence in the time dimension. The verification method is to compare whether the intensity difference between adjacent spectra at a specific reference wavelength exceeds the set interruption judgment threshold. For example, when detecting collagen labeled with Sirius red, a wavelength of 540 nanometers is selected as the reference wavelength. If the absorbance difference between two adjacent node spectra at this wavelength is greater than 0.15, it is determined that there is a spectral line interruption. If a spectral line interruption is detected due to probe movement vibration or instantaneous fluctuations in electronic signals, the system immediately initiates a spectral line repair process. This process retrieves the optical information corresponding to the spatial position in the initial optical profile, specifically reading the reflectance and transmittance values ​​corresponding to the coordinates of the interrupted node. Based on the trend of the spectral data from at least two adjacent nodes before and after the interrupted node, interpolation reconstruction is performed. In a specific operation, when the spectrum of node 7 on the dedicated detection path is interrupted, the repair algorithm extracts the spectral data of nodes 5, 6, 8, and 9 within the wavelength range of 1200 nm to 2200 nm, calculates their average slope of change, and combines this with the higher transmittance value recorded in the initial optical profile of node 7 to generate an interpolated spectrum that conforms to the local trend to fill the interruption.

[0078] In some embodiments, cross-node correlation fusion of spectral sequences specifically includes the following steps: assigning a spatial correlation weight to the spectral data collected at each stop node on the dedicated detection path. The spatial correlation weight is calculated based on the normalized distance relationship between the stop node and all other nodes on the path, and the calculation method conforms to the formula:

[0079] ;

[0080] in: This represents the spatial association weight of the i-th dwell node. This indicates the total number of nodes stopped on the dedicated detection path. This represents an exponential function used to calculate powers with the natural constant e as the base. Let represent the Euclidean distance between the i-th node and the j-th node. This is a preset distance scale parameter used to control the rate at which the weights decay with distance. In a detection process containing 15 dwell nodes, it is set... The spatial association weight calculated for a node located at the center of the node cluster is 5.0 mm. The value can reach 0.82, while isolated edge nodes have a lower value. The value is only 0.31. Using spatial correlation weights, a weighted average is applied to the spectral data of all nodes to generate a preliminary fused spectrum. The weighted average is achieved by multiplying the entire spectral vector of each node by its normalized weight. Then sum them up.

[0081] It is understandable that subtracting the common background spectrum caused by non-collagen components of the pigskin sample itself from the initial fusion spectrum is a crucial step in subsequent analysis. The common background spectrum is established by analyzing spectral data from collagen-deficient regions in a historical sample database. In practice, this database contains spectral data from over 200 pigskin samples whose collagen content was confirmed to be less than 0.5% by biochemical analysis. Principal component analysis was performed on these spectral data to extract the first three principal components, which were then linearly combined to construct a common background spectrum template. The result after subtracting the common background spectrum is used as the fusion spectral feature map. The subtraction operation involves subtracting the absorbance value of the common background spectrum template at each wavelength from the absorbance value of the initial fusion spectrum at that wavelength. Optionally, multiple sub-templates can be established for different parts of the pigskin, and the matching sub-template is automatically selected based on the initial optical profile characteristics and sample location information of the current sample during subtraction.

[0082] In one embodiment of the present invention, after generating the fused spectral feature map, its signal-to-noise ratio (SNR) and characteristic peak integrity index are calculated. If the SNR is lower than a preset threshold or the characteristic peak integrity indicates the absence of key features, a re-detection command is triggered. The calculation process of the SNR includes: randomly selecting multiple calculation points in the baseline stable region of the fused spectral feature map, using their average intensity as the baseline noise level; identifying characteristic absorption peaks belonging to collagen, calculating the peak intensity and full width at half maximum (FWHM) of each peak; using the ratio of peak intensity to noise level as the SNR of that peak, and taking the average of the SNRs of all characteristic peaks as the spectral SNR. The characteristic peak integrity index is obtained by statistically analyzing the ratio of the number of actually identified characteristic absorption peaks to the theoretical number of characteristic peaks that collagen must contain. The re-detection command controls the detection device to resample a specific sub-region of the pigskin sample according to a supplementary detection path, which is determined by the region with the most blurred information in the initial optical profile.

[0083] In practice, after generating the fused spectral feature map, the spectral evaluation module calculates the spectral signal-to-noise ratio (SNR) and characteristic peak integrity index of the fused spectral feature map. The calculation process is based on data from the fused spectral feature map within a specific wavelength range, such as the visible and near-infrared bands. The SNR calculation first involves randomly selecting multiple calculation points within the baseline stable region of the fused spectral feature map. The baseline stable region refers to an area where absorbance values ​​change smoothly and there are no characteristic absorption peaks, for example, within the wavelength range of 1700 nm to 1800 nm. At least 20 calculation points are uniformly selected, and the average signal intensity of these multiple calculation points is calculated as the baseline noise level. Identify characteristic absorption peaks belonging to collagen in the fused spectral feature map. The identification of characteristic absorption peaks is based on a pre-stored standard collagen spectral library. For example, for Sirius red stained samples, characteristic absorption peaks should exist near 540 nm, 1200 nm, and 2050 nm. Determine the peak intensity and full width at half maximum (FWHM) of each characteristic absorption peak. The peak intensity is the difference in absorbance between the highest point of the absorption peak and the adjacent baseline, and the FWHM is the wavelength width corresponding to half the peak height.

[0084] In some embodiments, the ratio of the peak intensity of a characteristic absorption peak to a reference noise level is used as the signal-to-noise ratio (SNR) of that characteristic absorption peak, and the average of the SNRs of all characteristic absorption peaks is taken as the spectral SNR. This calculation conforms to the following relationship:

[0085] ;

[0086] in: This indicates the signal-to-noise ratio of the spectrum. This represents the total number of characteristic absorption peaks identified from the fused spectral feature map. This represents the peak intensity of the p-th characteristic absorption peak. This represents the calculated baseline noise level. The number of characteristic absorption peaks actually identified in the fused spectral feature map is statistically analyzed to obtain the theoretical total number of characteristic absorption peaks that collagen must contain, as defined by the predefined criteria. This theoretical total number is pre-set based on the specific binding spectral characteristics of the detection medium and collagen; for example, for the aforementioned Sirius red staining method, the theoretical total number is set to three key peaks.

[0087] If the spectral signal-to-noise ratio is lower than a preset threshold or the feature peak integrity index indicates missing key features, a re-detection command is triggered. The preset threshold is determined based on statistical analysis of historical detection data; for example, the spectral signal-to-noise ratio threshold is set to 15, and the feature peak integrity index threshold is set to 0.67. The re-detection command controls the detection equipment to resample a specific sub-region of the pigskin sample according to a supplementary detection path. The planning of the supplementary detection path is determined by the region with the most blurred information in the initial optical contour. The ambiguity is quantified by calculating the image entropy value or the optical gradient mean square error of the local region. In a specific operation, the transmittance gradient mean square error of each 10-pixel × 10-pixel sub-region in the initial optical contour is calculated, and the center point of the region with the lowest mean square error (i.e., the most blurred region) is selected as the sampling point of the supplementary detection path.

[0088] In one embodiment of the present invention, after triggering a re-detection command and completing supplementary sampling, the resampled spectral data is replaced with the data of the corresponding spatial region in the fused spectral feature map; cross-node correlation fusion and common background spectral subtraction are performed again on the replaced new map to generate an updated fused spectral feature map. The pre-stored feature-content mapping model is established through the following steps: a large number of pig skin standard samples with known precise collagen content are collected, an initial optical profile is established for each sample, and spectral sequences are collected according to a dedicated detection path; cross-node correlation fusion is performed on the spectral sequences of all samples to construct a standard fused spectral feature map library; a nonlinear iterative optimization algorithm is used to find the optimal mapping function from the map library to the known content value, which is a network model with multi-layer nonlinear transformation; with the standard map library as the input vector and the known content value as the target output, the network weights and bias parameters are adjusted through forward calculation and error backpropagation until the error converges, and the final network model is solidified as a feature-content mapping model.

[0089] In practice, after the re-inspection command controls the detection equipment to resample a specific sub-region of the pigskin sample according to a supplementary detection path, the system executes a data replacement and result update process. The resampled spectral data is replaced with the data from the corresponding spatial region in the fused spectral feature map. The replacement operation is based on spatial coordinate mapping. For example, if the supplementary detection path covers the region represented by nodes 5 to 7 in the original dedicated detection path, the three sets of resampled spectral data replace the spectral vectors from nodes 5, 6, and 7 in the original fused spectral feature map generation process. The new dataset after replacement undergoes cross-node correlation fusion and common background spectral subtraction steps again to generate an updated fused spectral feature map. This process repeats the weighted average and background subtraction algorithms, but the node spectral data sets involved in the calculation have been locally updated. The fused spectral feature maps before and after the update are simultaneously input into the feature-content mapping model. The feature-content mapping model calculates the two maps separately and outputs two predicted collagen content values. The arithmetic mean of the two outputs is taken as the final collagen content value.

[0090] In some embodiments, the method for establishing a pre-existing feature-content mapping model includes two stages: standard sample library construction and model training. A large number of pig skin standard samples with known precise collagen content are collected. The known content is obtained through hydroxyproline chemical determination. An initial optical profile is established for each standard sample, and spectral sequences are acquired according to a dedicated detection path generated based on this profile. Cross-node correlation fusion processing is performed on the spectral sequences of all standard samples to construct a standard fused spectral feature map library. Each data point in the library is associated with a real collagen content value. See Table 1 for some example data from the standard spectral map library. A nonlinear iterative optimization algorithm is used to find the optimal mapping function between the standard fused spectral feature map library and the known collagen content value, and this optimal mapping function is solidified into a feature-content mapping model.

[0091] Table 1: Partial Tables of the Standard Fusion Spectral Feature Library

[0092]

[0093] It can be understood that the structure of the optimal mapping function is defined as a network model with multiple nonlinear transformations. This network model contains one input layer, at least two hidden layers, and one output layer. The number of nodes in the input layer is equal to the data dimension of the fused spectral feature map, and the number of nodes in the output layer is 1, corresponding to the predicted collagen content value. All spectral data from the standard fused spectral feature map library are used as the input vector of the network model, and the corresponding known collagen content value is used as the target output. Initial weight parameters and bias parameters are assigned to each layer of the network model. The initial weight parameters are usually generated by a random number generator within a specific interval. The forward computation process is performed, passing the input vector layer by layer through the network model. At each layer, a linear weighted sum is performed and then transformed by an activation function to obtain the predicted content value under the current parameters. The total error between the predicted content value and the known collagen content value is calculated. The total error function uses the mean squared error, and its expression is:

[0094] ;

[0095] in: Indicates the total error. This represents the total number of standard samples in the training set. This represents the known collagen content value of the k-th sample. This represents the predicted content value output by the network model for the k-th sample.

[0096] Optionally, based on the total error, the weight and bias parameters of each layer of the network model are adjusted through a backpropagation process. The backpropagation process calculates the partial derivative of the error with respect to each parameter using the gradient descent algorithm and updates the parameters along the negative gradient direction. The forward calculation and backpropagation processes are repeated until the change in the total error is less than a preset convergence threshold or the maximum number of iterations is reached; for example, the convergence threshold is set to 0.001, and the maximum number of iterations is 10,000. The finally stabilized network model and all its weight and bias parameters are then solidified into a feature-content mapping model. This model can be stored as a data file in the detection system's memory for subsequent comparisons.

[0097] See Figure 4In the data optimization process following re-detection of porcine skin collagen content, the comparison results of the original fused spectral feature maps and the updated fused spectral feature maps are presented. Specifically, the figure shows the original fused spectral feature map corresponding to the purple curve and the updated fused spectral feature map corresponding to the green curve, with the detection wavelength (nm) as the horizontal axis and the spectral feature intensity (AU) as the vertical axis. Different color blocks are used to mark the corresponding regions of feature peaks A, B, and C. From the differences in the spectral maps, the original map shows a lower signal intensity and stronger fluctuation noise in the feature peak B region (around 300-350nm), while the updated map, through supplementary sampling data replacement and secondary fusion processing, shows a significant increase in signal intensity and a reduction in fluctuation amplitude in this region. At the same time, the integrity of feature peaks in the feature peak A (250-275nm) and feature peak C (350-375nm) regions of the updated map is also enhanced, as evidenced by improved peak intensity and baseline discrimination. This difference reflects the optimization effect of the re-examination process on the signal-to-noise ratio and feature peak integrity of the fused spectral feature map, providing more reliable input data for the accurate prediction of the feature-content mapping model.

[0098] In one embodiment of the present invention, after outputting the collagen content value of the pigskin sample, a detection record is generated based on the dedicated detection path used in this detection, the actual diffusion rate, and the output content value. This record is added to a dynamic experience base to optimize the path generation rules and media diffusion model parameters for subsequent samples. When a new sample detection task is received, its initial optical profile is matched with the initial optical profiles of all records in the dynamic experience base. If a matching existing record with a similarity exceeding a threshold is found, the dedicated detection path scheme and estimated diffusion rate stored in that record are directly used as the initialization parameters for the new sample detection, skipping the path planning calculation process.

[0099] In practice, after outputting the collagen content value of the pigskin sample, the system's record generation module generates a structured detection record based on the dedicated detection path used in this test, the actual diffusion rate, and the final output content value. The detection record is stored as data entries, including but not limited to the sample identifier, the feature hash value of the initial optical profile, the node coordinate sequence of the dedicated detection path, the actual diffusion rate of the medium, environmental temperature and humidity data, and the final content value. The detection record is added to a dynamic experience base, a relational database used to optimize the dedicated detection path generation rules and medium diffusion model parameters for subsequent samples. For example, when the dynamic experience base accumulates more than 1000 records from pigskin samples, the path generation algorithm adjusts the recognition threshold for highly heterogeneous regions of new samples based on the node distribution patterns of historically successful paths, and the medium diffusion model refits its parameters using the newly added actual diffusion rate and environmental data.

[0100] In some embodiments, when a new pigskin sample detection task is received, the system performs a similarity match between its initial optical profile and the initial optical profiles of all records in the dynamic experience base. The matching process is based on the feature vectors of the optical profiles. For example, the reflectivity and transmittance matrices of the initial optical profile are compressed into 512-dimensional feature vectors, and the cosine similarity between these feature vectors and the corresponding feature vectors of each record in the dynamic experience base is calculated. If an existing record with a similarity exceeding a threshold is matched, the dedicated detection path scheme and estimated diffusion rate stored in the existing record are directly used as the initialization parameters for the new sample detection, and the path planning calculation process is skipped. The similarity threshold can be set to 0.95. In a specific operation, if the cosine similarity calculation result between the feature vector of the new sample and the record numbered "REC-2024-0782" in the dynamic experience base is 0.968, the system automatically loads the dedicated detection path node coordinate sequence and the estimated diffusion rate of 0.82 mm / s stored in the record "REC-2024-0782" and directly starts the subsequent detection process.

[0101] It is understandable that various methods can be used to measure the similarity between the initial optical profile of a new sample and the profiles recorded in the dynamic experience database. One specific implementation is to use the weighted Euclidean distance based on feature vectors, the calculation formula of which is:

[0102] ;

[0103] in: This represents the calculated similarity score, with values ​​between 0 and 1. This represents the total dimension of the feature vector. This represents the value of the f-th dimension of the initial optical profile feature vector of the new sample. This represents the value of the f-th dimension of the feature vector corresponding to a record in the dynamic experience base. This represents the weight coefficient pre-assigned to the f-th feature dimension, which is determined based on the information entropy of that dimension in distinguishing different sample contours.

[0104] Optionally, after each new record is added, the dynamic experience base will automatically perform cluster analysis on the records in the base, classifying similar optical profiles with corresponding successful detection parameters. When a new sample has a high degree of matching with the overall features of a certain cluster, the "representative" detection path scheme of that cluster can be directly adopted. The representative path scheme is obtained by averaging the node coordinates of all paths within the cluster.

[0105] See Figure 5In the optimization of the detection process for porcine skin collagen content, the performance difference between the rapid path (experience base matching) and the conventional path (replanning) was quantified through multi-dimensional indicators. Specifically, in terms of path planning time, the rapid path (2s) achieved significant time compression compared to the conventional path (18s), thanks to the contour similarity matching mechanism of the dynamic experience base, which skipped the calculation process of path replanning. In terms of total detection time, the efficiency improvement of the rapid path (60s) compared to the conventional path (76s) stemmed from the direct reuse of the path schemes and diffusion rate parameters pre-stored in the experience base. In terms of data completeness, both the rapid path (99%) and the conventional path (97%) maintained high levels, reflecting the high effectiveness of the pre-stored paths in the experience base. In terms of detection error, the rapid path (1.2%) was better than the conventional path (2.5%), which is related to the enhanced detection stability of the "representative" path scheme after cluster analysis of the experience base. The comparison results of various indicators intuitively verified the optimization value of the rapid path driven by the dynamic experience base in terms of detection process efficiency and accuracy.

[0106] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0107] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for detecting collagen content in pig skin, characterized in that, Includes the following steps: Before the pigskin sample comes into contact with the detection medium, an initial optical profile of the pigskin sample is established. The initial optical profile is composed of the surface reflectance distribution and the internal transmittance distribution of the sample. Based on the initial optical profile, a dedicated detection path is generated for the pigskin sample. The dedicated detection path defines the movement trajectory and dwell points of the detection probe on the sample surface for spectral sampling. The detection medium is injected at the starting node of the dedicated detection path, and the path following mechanism is activated simultaneously to keep the movement of the detection probe synchronized with the diffusion front of the detection medium. When the detection probe reaches the end node of the dedicated detection path, the spectral sequence collected at all the stopping nodes on the dedicated detection path is completely recorded. The spectral sequence contains the characteristic absorption information of collagen at different nodes. Cross-node correlation fusion is performed on the spectral sequence to generate a fused spectral feature map of the pigskin sample; The fused spectral feature map is compared with the pre-stored feature-content mapping model to output the collagen content value of the pig skin sample.

2. The method for detecting collagen content in pig skin according to claim 1, characterized in that, After establishing the initial optical profile of the pigskin sample, the method further includes the following steps: By analyzing the spatial gradient changes of reflectance and transmittance in the initial optical profile, high heterogeneity regions and low heterogeneity regions on the surface of the pigskin sample were identified. In the planning of the dedicated detection path, the movement trajectory is made to preferentially traverse highly heterogeneous regions and reduce the density of dwelling nodes in low heterogeneous regions. Based on the identified regional characteristics, an independent sampling integration time is assigned to each stop node on the dedicated detection path, with nodes in highly heterogeneous regions being assigned a longer sampling integration time.

3. The method for detecting collagen content in pig skin according to claim 2, characterized in that, The process of injecting detection medium at the starting node of the dedicated detection path and simultaneously activating the path following mechanism includes the following steps: The actual diffusion rate and direction of the detection medium on the surface of pigskin samples were monitored. Based on the actual diffusion rate and diffusion direction, the moving speed of the detection probe and the local orientation of the dedicated detection path are dynamically fine-tuned to ensure that the detection probe is always located in the area where the detection medium has been fully impregnated for spectral sampling. The dynamic fine-tuning is based on a pre-established medium diffusion model, which relates the influence of pigskin texture density and environmental temperature and humidity on the diffusion process.

4. The method for detecting collagen content in pig skin according to claim 1, characterized in that, After fully recording the spectral sequences collected at all stopping points along the dedicated detection path, the following steps are also included: Check the continuity of the spectral sequence in the time dimension. If a spectral line interruption is found due to probe movement or signal fluctuation, the spectral line repair process is initiated. The spectral line repair process calls the optical information of the corresponding spatial position in the initial optical profile, and performs interpolation reconstruction on the spectral data at the interruption point based on the trend of the spectral data of adjacent nodes.

5. The method for detecting collagen content in pig skin according to claim 4, characterized in that, The cross-node correlation fusion of the spectral sequence specifically includes the following steps: A spatial correlation weight is assigned to the spectral data collected at each stop node on the dedicated detection path. The spatial correlation weight is calculated based on the reciprocal of the Euclidean distance between the stop node and all other nodes on the path. Using the spatial correlation weights, the spectral data of all nodes are weighted and averaged to generate a preliminary fused spectrum; The common background spectrum caused by non-collagen components of the pigskin sample itself is subtracted from the preliminary fusion spectrum. The common background spectrum is established by analyzing the spectral data of collagen-deficient regions in the historical sample library. The result after deducting common background spectra is used as the fused spectral feature map.

6. The method for detecting collagen content in pig skin according to claim 5, characterized in that, It also includes the following steps: After generating the fused spectral feature map, the spectral signal-to-noise ratio and feature peak integrity index of the fused spectral feature map are calculated; If the signal-to-noise ratio of the spectrum is lower than a preset threshold or the feature peak integrity index shows that key features are missing, a re-examination instruction is triggered. The re-inspection command controls the detection equipment to resample a specific sub-region of the pigskin sample according to a supplementary detection path, wherein the supplementary detection path is determined by the region with the most blurred information in the initial optical profile. The calculation of the spectral signal-to-noise ratio and characteristic peak integrity index of the fused spectral feature map specifically includes: Multiple calculation points are randomly selected within the baseline stable region of the fused spectral feature map, and the average signal intensity of the multiple calculation points is calculated as the reference noise level; Identify the characteristic absorption peaks belonging to collagen in the fused spectral feature map, and determine the peak intensity and full width at half maximum (FWHM) of each characteristic absorption peak; The ratio of the peak intensity of the characteristic absorption peak to the reference noise level is used as the signal-to-noise ratio of the characteristic absorption peak. The average signal-to-noise ratio of all the characteristic absorption peaks is taken as the signal-to-noise ratio of the spectrum; Count the number of characteristic absorption peaks actually identified in the fused spectral feature map; Obtain the theoretical total number of characteristic absorption peaks that a predefined collagen must contain; The ratio of the number of actual identified characteristic absorption peaks to the theoretical total number is used as the characteristic peak integrity index.

7. The method for detecting collagen content in pig skin according to claim 6, characterized in that, After the re-inspection command controls the detection equipment to resample a specific sub-region of the pigskin sample according to a supplementary detection path, the following steps are also included: The resampled spectral data is then replaced with the data of the corresponding spatial region in the fused spectral feature map. The new spectrum after replacement is subjected to cross-node correlation fusion and common background spectrum subtraction steps again to generate an updated fused spectral feature map; The fused spectral feature maps before and after the update are simultaneously input into the feature-content mapping model, and the average of the two outputs is taken as the final collagen content value.

8. The method for detecting collagen content in pig skin according to claim 1, characterized in that, The method for establishing the pre-stored feature-content mapping model includes the following steps: A large number of pig skin standard samples with known precise collagen content were collected. An initial optical profile was established for each standard sample, and spectral sequences were acquired according to its dedicated detection path. The cross-node correlation fusion processing is performed on the spectral sequences of all standard samples to construct a standard fused spectral feature map library; A nonlinear iterative optimization algorithm is used to find the optimal mapping function between the standard fused spectral feature map library and known collagen content values. This optimal mapping function is then solidified into the feature-content mapping model, specifically including: The structure of the optimal mapping function is defined as a network model with multi-layer nonlinear transformations; All spectral data in the standard fusion spectral feature map library are used as the input vector of the network model, and the corresponding known collagen content value is used as the target output. Assign initial weight parameters and bias parameters to each layer of the network model; The forward computation process is performed, passing the input vector layer by layer through the network model to obtain the predicted content value under the current parameters; Calculate the total error between the predicted content value and the known collagen content value; Based on the total error, the weight parameters and bias parameters of each layer of the network model are adjusted through a backpropagation process. Repeat the forward calculation process and the backward propagation process until the change in the total error is less than the preset convergence threshold or the maximum number of iterations is reached; The final stabilized network model and all its parameters are then solidified into the feature-content mapping model.

9. The method for detecting collagen content in pig skin according to claim 1, characterized in that, After outputting the collagen content value of the pigskin sample, the following steps are also included: A detection record is generated based on the dedicated detection path used in this test, the actual diffusion rate, and the final output content value. The detection records are added to a dynamic experience base, which is used to optimize the generation rules for dedicated detection paths and the parameters of the medium diffusion model for subsequent samples.

10. The method for detecting collagen content in pig skin according to claim 9, characterized in that, It also includes the following steps: When a new pigskin sample detection task is received, its initial optical profile is matched with the initial optical profiles of all records in the dynamic experience base. If a matching existing record with a similarity exceeding the threshold is found, the dedicated detection path scheme and estimated diffusion rate stored in the existing record are directly used as the initialization parameters for the new sample detection, and the path planning calculation process is skipped.

Images