Method and system for automated optical inspection of visible foreign matter in midazolam buccal mucosal solution

Abstract

The application provides a method and system for automatically optically detecting visible foreign matters in midazolam buccal mucosa solution, and relates to the technical field of drug detection. The application controls the flow of midazolam buccal mucosa solution in a transparent flow cell, dynamically observes and identifies suspended particles in the solution by means of an optical imaging system. Then, the polarized light source of the optical imaging system is controlled to rotate relative to the transparent flow cell, and optical response data of each suspended particle at multiple rotation angles is collected. Next, the change information of light intensity with the relative rotation angle is determined according to the data, and the color distribution information is synchronously extracted to generate key feature data representing the optical response characteristics of the particles. Finally, the key feature data is compared with the preset standard feature data of the raw materials and excipients, so that the suspended particles are determined as raw material crystals, excipient precipitates or visible foreign matters, and the automatic and accurate classification of the suspended particles in the midazolam buccal mucosa solution is realized.

Description

Technical Field

[0001] This application relates to the field of drug detection technology, and in particular to an automated optical detection method and system for visible foreign matter in a midazolam buccal mucosa solution. Background Technology

[0002] Midazolam buccal mucosal solution is a key drug for treating acute seizures in children, and the detection of visible foreign matter is a core aspect of drug quality control. This automated optical detection method can accurately identify suspended particles in the solution, providing assurance for drug production quality control and clinical medication safety, and has broad application prospects in the field of pediatric medication.

[0003] Currently, the automatic optical detection of visible foreign matter in such pharmaceutical solutions often employs the light obscuration method. This method is based on the change in light intensity caused by particles blocking light, enabling the automatic counting and particle size detection of insoluble particles in the solution. It is widely used in the quality screening of injectable and other pharmaceutical solutions.

[0004] Optical obscuration methods struggle to distinguish specific types of particles, failing to differentiate between suspended particles that are active pharmaceutical ingredient crystals, excipient precipitates, or other visible foreign matter. This can easily lead to misinterpretations and affect the accurate assessment of drug purity and safety. Therefore, existing technologies suffer from insufficient ability to identify visible foreign matter in complex components. Summary of the Invention

[0005] The purpose of this application is to provide an automated optical detection method and system for visible foreign matter in midazolam buccal mucosa solution, in order to solve the problem of insufficient ability to identify visible foreign matter in complex components in the prior art.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides an automated optical detection method for visible foreign matter in a midazolam buccal mucosa solution, comprising:

[0007] The flow of midazolam buccal mucosa solution was controlled in a transparent flow cell, and the solution under flow was dynamically observed using an optical imaging system to identify suspended particles in the midazolam buccal mucosa solution.

[0008] The polarized light source in the optical imaging system is controlled to rotate relative to the transparent flow cell. During the relative rotation, optical response data of each suspended particle is collected at multiple rotation angles.

[0009] Based on the optical response data of each suspended particle, the information on the change of light intensity with the relative rotation angle is determined, and the color distribution information is extracted simultaneously.

[0010] By combining the change information and the color distribution information, key feature data of the suspended particles are generated. The key feature data is used to characterize the optical response characteristics of the suspended particles during relative rotation.

[0011] Based on the comparison results between the key characteristic data and the preset standard characteristic data of active pharmaceutical ingredients and excipients, the suspended particles are determined to be active pharmaceutical ingredient crystals, excipient precipitates, or visible foreign matter.

[0012] Optionally, after determining that the suspended particles are active pharmaceutical ingredient crystals, excipient precipitates, or visible foreign matter, the method further includes:

[0013] The timestamp information of suspended particles identified as active pharmaceutical ingredient crystals, excipient precipitates, and visible foreign matter is recorded respectively to generate time series data of at least three types of particles;

[0014] Calculate the cross-correlation function between the time series data of different types of particles to determine at least one particle type as a leading indicator;

[0015] When the growth rate of the number of particles in the leading indicator exceeds a preset rate, a warning signal representing the risk of precipitation of the corresponding type of particles is generated.

[0016] Optionally, after determining that the suspended particles are active pharmaceutical ingredient crystals, excipient precipitates, or visible foreign matter, the method further includes:

[0017] For each type of suspended particulate matter, time-series data on the change in quantity over time is generated;

[0018] Analyze the time series data to determine the change pattern of the corresponding type of suspended particles, and predict the change in the number of suspended particles within a preset time period based on the change pattern;

[0019] When the quantity change exceeds a preset threshold, a graded early warning signal is generated, which represents the change pattern and quantity change.

[0020] Optionally, the step of combining the change information and the color distribution information to generate key feature data of the suspended particles includes:

[0021] Periodic components are extracted from the change information to determine a first feature parameter, which is used to quantify the regularity of the periodic components.

[0022] The distribution pattern of the color distribution information in a preset color space is analyzed to determine the second feature parameter, which is used to characterize the statistical properties of the distribution pattern.

[0023] The first feature parameter and the second feature parameter are combined to generate a joint feature vector, and a scale value is generated based on the relationship between the components in the joint feature vector.

[0024] The joint feature vector is combined with the scaling value to form key feature data.

[0025] Optionally, analyzing the distribution pattern of the color distribution information in a preset color space to determine the second feature parameter includes:

[0026] Based on the frequency of occurrence in the color distribution information, a weighted color point set is determined in a preset color space;

[0027] Calculate the spatial distribution ellipsoid of the weighted color point set, and determine the principal axis direction of the spatial distribution ellipsoid;

[0028] Based on the distribution range in the color distribution information, obtain the first dimension parameter of the spatial distribution ellipsoid along the main axis direction and the second dimension parameter perpendicular to the main axis direction;

[0029] A second feature parameter is generated based on the ratio of the first size parameter to the second size parameter.

[0030] Optionally, generating a scale value based on the relationship between the components within the joint feature vector includes:

[0031] The difference between the first feature parameter and the second feature parameter in the joint feature vector is used as the first relationship factor.

[0032] Obtain the angle between the principal axis direction of the spatial distribution ellipsoid corresponding to the second feature parameter and the preset plane, and use it as the second relationship factor;

[0033] The first relation factor and the second relation factor are multiplied to generate an intermediate value;

[0034] The intermediate value is nonlinearly adjusted to generate the scale value, and the degree of nonlinear adjustment is determined by the first feature parameter.

[0035] Optionally, determining the change in light intensity with relative rotation angle based on the optical response data of each of the suspended particles, and simultaneously extracting color distribution information, includes:

[0036] For each suspended particle, the target image of the suspended particle at each rotation angle is extracted from the optical response data, and the light intensity value of each target image at the corresponding rotation angle is determined;

[0037] Arrange the light intensity values ​​according to the rotation angle to generate information characterizing the change of light intensity with the rotation angle;

[0038] Simultaneously, the color features of each target image are analyzed to generate color distribution information, the color features including the frequency and distribution range of pixel color values.

[0039] Optionally, the dynamic observation of the solution in a flowing state to identify suspended particles in the midazolam buccal mucosa solution includes:

[0040] Dynamic observation was conducted on the buccal mucosa solution of midazolam in a flowing state to obtain image data;

[0041] Analyze the image data to determine the initial region where the difference between the brightness value and the background exceeds a preset threshold;

[0042] The initial region is filtered to obtain candidate regions whose geometric features meet preset conditions;

[0043] The candidate regions are tracked in the image data, and the candidate regions that have shifted are identified as suspended particles.

[0044] Optionally, determining the suspended particles as active pharmaceutical ingredient crystals, excipient precipitates, or visible foreign matter based on the comparison results between the key feature data and preset standard feature data of active pharmaceutical ingredients and excipients includes:

[0045] The key feature data corresponding to each suspended particle is compared with the preset standard feature data of active pharmaceutical ingredient and standard feature data of excipient.

[0046] When the key feature data matches the standard feature data of the active pharmaceutical ingredient, the corresponding suspended particles are identified as active pharmaceutical ingredient crystals.

[0047] When the key feature data matches the standard feature data of the excipient, the corresponding suspended particles are identified as excipient precipitates.

[0048] When the key feature data does not match the standard feature data of the active pharmaceutical ingredient and the standard feature data of the excipient, the corresponding suspended particles are identified as visible foreign matter.

[0049] Secondly, this application provides an automated optical detection system for visible foreign matter in a midazolam buccal mucosa solution, comprising:

[0050] The observation module is used to control the flow of midazolam buccal mucosa solution in a transparent flow cell and to use an optical imaging system to dynamically observe the solution in the flow state in order to identify suspended particles in the midazolam buccal mucosa solution.

[0051] The acquisition module is used to control the relative rotation between the polarized light source in the optical imaging system and the transparent flow cell. During the relative rotation, the module acquires optical response data of each suspended particle at multiple rotation angles.

[0052] The extraction module is used to determine the change information of light intensity with relative rotation angle based on the optical response data of each suspended particle, and simultaneously extract the color distribution information;

[0053] A generation module is used to combine the change information and the color distribution information to generate key feature data of the suspended particles. The key feature data is used to characterize the optical response characteristics of the suspended particles during relative rotation.

[0054] The determination module is used to determine the suspended particles as active pharmaceutical ingredient crystals, excipient precipitates, or visible foreign matter based on the comparison results between the key feature data and the preset standard feature data of active pharmaceutical ingredients and excipients.

[0055] The automated optical detection method for visible foreign matter in midazolam buccal mucosa solution provided in this application controls the flow of midazolam buccal mucosa solution in a transparent flow cell and dynamically observes it through an optical imaging system. This captures suspended particles in real time during the flow, avoiding missed detections due to particle deposition and achieving accurate particle identification. Controlling the relative rotation of the polarized light source with the transparent flow cell and collecting multi-angle optical response data allows for more comprehensive information on the optical characteristics of the particles, overcoming the limitations of single-angle detection. Determining light intensity angle changes and extracting color distribution information based on the optical response data reveals the unique optical properties of the particles, providing effective data support for subsequent classification. Combining these two types of information to generate key feature data integrates dispersed optical information into discriminative core features, facilitating subsequent comparison and judgment. Comparing the key feature data with preset standard data accurately distinguishes particle types, clarifying whether suspended particles are active pharmaceutical ingredient crystals, excipient precipitates, or visible foreign matter, ensuring the specificity and reliability of the detection results.

[0056] Furthermore, after classifying and determining the particulate matter, this method records the timestamps of the three types of particulate matter to generate time series. By calculating the cross-correlation function between different types of sequences, it identifies a certain type of particulate matter that can serve as a leading indicator. When the growth rate of this particulate matter exceeds a threshold, it proactively generates an early warning signal characterizing the risk of precipitation. This method elevates detection from single-point identification to the process detection level. By identifying the temporal correlation between leading indicators and subsequent precipitates, it can provide early warning of potential precipitation trends, achieving a leap from passive detection to proactive early warning. This provides crucial evidence for timely adjustments to production processes and feedforward control of product quality. Attached Figure Description

[0057] To more clearly illustrate the technical solutions of the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 A flowchart illustrating an automated optical detection method for visible foreign matter in a midazolam buccal mucosa solution provided in this application embodiment;

[0059] Figure 2 This application provides a schematic diagram of the optical response curve of an automated optical detection method for visible foreign matter in a midazolam buccal mucosa solution.

[0060] Figure 3 A schematic flowchart of another automated optical detection method for visible foreign matter in a midazolam buccal mucosa solution provided in this application embodiment;

[0061] Figure 4 This is a schematic diagram of an automated optical detection system for visible foreign matter in a midazolam buccal mucosa solution, provided as an embodiment of this application. Detailed Implementation

[0062] In the quality testing of midazolam buccal mucosal solution, existing optical obscuration methods can only count and detect the size of suspended particles, but cannot distinguish the specific type of particles. This makes it difficult to determine whether the particles are active pharmaceutical ingredient crystals, excipient precipitates, or other visible foreign matter, easily leading to misinterpretations of test results. Consequently, this affects the accurate assessment of drug purity and safety, posing a potential threat to drug quality control.

[0063] To address the aforementioned issues, this invention proposes an automated optical detection method for visible foreign matter in midazolam buccal mucosa solution. The core of this method is to achieve accurate classification by dynamically observing and acquiring multi-angle optical data to uncover the unique optical characteristics of microparticles. Specifically, the drug solution is first allowed to flow in a flow cell to dynamically identify microparticles. Then, a polarized light source is rotated relative to the flow cell to collect multi-angle optical response data of the microparticles. Information on changes in light intensity angle and color distribution is integrated to generate key features, which are then compared with preset standard data to determine the microparticle type. This method overcomes the limitation of the photoresist method in distinguishing microparticle types, fundamentally solving the problem of misjudgment and improving the accuracy of visible foreign matter detection in pharmaceuticals, thus providing strong protection for drug quality and medication safety.

[0064] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0065] The core of this application is to provide an automated optical detection method for visible foreign matter in a midazolam buccal mucosa solution, and a flowchart of one specific embodiment is shown below. Figure 1 As shown, the method includes:

[0066] S101. Control the flow of midazolam buccal mucosa solution in a transparent flow cell, and use an optical imaging system to dynamically observe the solution in the flow state in order to identify suspended particles in the midazolam buccal mucosa solution.

[0067] The transparent flow cell is a closed channel made of transparent material through which the solution to be tested flows stably. Its core function is to constrain the solution and provide a clear and stable optical observation window. The optical imaging system typically consists of a light source, a lens, and an image sensor. Its function is like a high-speed microscope, continuously capturing dynamic images of the solution within the flow cell.

[0068] S101 specifically includes:

[0069] S1011. Dynamically observe the midazolam buccal mucosa solution in a flowing state and acquire image data.

[0070] S1012. Analyze the image data to determine the initial region where the difference between the brightness value and the background exceeds a preset threshold.

[0071] Brightness value refers to the lightness or darkness of each pixel in the image, typically ranging from 0 to 255, with higher values ​​indicating greater brightness. Geometric features include mathematical parameters describing the shape of a region, such as area, perimeter, and roundness. The preset threshold is a pre-defined numerical limit used to distinguish foreground objects from the background; this threshold is calculated by statistically analyzing the background brightness distribution of the image.

[0072] S1013. Filter the initial region to obtain candidate regions whose geometric features meet preset conditions.

[0073] S1014. Track the position of the candidate region in the image data, and determine the candidate region that has been displaced as a suspended particle.

[0074] In one specific implementation, a pumping device continuously delivers a midazolam buccal mucosal solution to a transparent flow cell at a stable flow rate, ensuring smooth flow of the solution within a closed channel and preventing uneven particle distribution or deposition due to flow rate fluctuations. Simultaneously, the optical imaging system activates a dynamic observation mode, using the coordinated operation of a light source, lens, and image sensor to continuously capture images of the flowing solution like a high-speed microscope, acquiring high-frame-rate raw image data and fully recording the dynamic trajectories of particles in the solution.

[0075] The collected image data is processed in layers: First, the image is scanned frame by frame using an image processing algorithm to calculate the average brightness of each pixel region and compare it with the overall background brightness. When the brightness difference of a certain region exceeds a preset threshold, the region is marked as the initial region where potential suspended particles exist, thus achieving the initial screening of the target.

[0076] Next, morphological screening is performed on the initial regions, and the geometric feature parameters of each region are calculated, including area, perimeter, and circularity. The formula for calculating circularity is as follows:

[0077]

[0078] Where e represents the circularity. ≈3.14, where S is the area of ​​the initial region and C is the perimeter of the initial region.

[0079] By setting a reasonable parameter range, noise interference that does not meet the size and shape requirements is eliminated, and candidate regions are obtained.

[0080] Then, through multi-frame image analysis, motion trajectory tracking is performed on the candidate regions. The change in center coordinates of the same candidate region in consecutive image frames is calculated, and only regions with significant displacement are identified as real suspended particles.

[0081] As an example, the average background brightness is set to 50 (grayscale range of 0-255), and the preset difference threshold is 30. During image processing, if the average brightness of a certain window area is detected to be 85, the brightness difference is 35. Since 35 is greater than the preset difference threshold of 30, this area is marked as the initial area A.

[0082] Next, geometric features of the initial region A are calculated and filtered. The area threshold is set to 5-500 pixels, and the circularity threshold is set to >0.7.

[0083] The initial region A was calculated to have an area of ​​80 pixels and a perimeter of 35 pixels. (Circularity calculation) Since 80 ∈ [5, 500] and 0.82 > 0.7, region A meets the conditions and is retained as a candidate region.

[0084] Then, the motion characteristics of candidate region A are verified by analyzing its positional changes in consecutive image frames. The Euclidean distance displacement of the region's center coordinates is calculated, with a threshold of greater than 3 pixels for the displacement distance. The center coordinates of candidate region A in three consecutive frames are (100, 150), (102, 149), and (105, 148), respectively. The displacement between adjacent frames is calculated as follows:

[0085] First to second frame Pixel;

[0086] Second to third frames Pixel.

[0087] Region A was identified as a suspended particle because of a displacement greater than 3 pixels.

[0088] This application employs a three-stage processing flow: initial screening based on brightness differences, fine screening based on geometric features, and displacement verification. This effectively eliminates noise interference and static artifacts, accurately identifying real suspended particles and laying the foundation for subsequent classification analysis.

[0089] S102. Control the polarization light source in the optical imaging system to rotate relative to the transparent flow cell. During the relative rotation, for each suspended particle, acquire the optical response data of the suspended particle at multiple rotation angles.

[0090] Relative rotation refers to the relative circular motion around the optical axis between the polarization source and the transparent flow cell. Optical response data consists of image sequences containing the polarization optical properties of suspended particles, acquired at different relative rotation angles.

[0091] In one specific implementation, three adaptive adjustments are first made based on the characteristics of the identified suspended particles:

[0092] Firstly, when the particle size is small, the magnification is automatically increased to ensure image clarity;

[0093] Secondly, shorten the exposure time for fast-moving particles to avoid motion blur;

[0094] Third, the intensity of polarized light is dynamically adjusted based on the initial optical properties of the particles to ensure that image data with the best signal-to-noise ratio is obtained.

[0095] Next, after completing the adaptive parameter adjustment, the polarized light source and the transparent flow cell are controlled to rotate relative to each other. During the rotation, image acquisition is triggered simultaneously at multiple specific angular positions for each identified suspended particle, acquiring a set of optical response data showing the optical response of the particle at different polarization angles, providing complete optical information for subsequent feature analysis.

[0096] As an example, after completing the identification of suspended particles, the first step is to automatically optimize the parameters of the detected particle B: based on the 120-pixel projected area of ​​the particle, the optical magnification is automatically increased to 1.5 times, the exposure time is shortened to 1 / 500 second based on its moving speed of 2.24 pixels / frame, and the polarized light source intensity is adjusted to 60% for its high brightness characteristics of 180 grayscale value.

[0097] After completing these adaptive adjustments, the system drives the polarization light source and the flow cell to rotate relative to each other at 45° intervals. Polarization images of particle B are acquired at eight key angular positions, from 0°, 45°, 90° to 315°, and finally optical response data that fully records the optical response characteristics of the particle at different polarization angles are obtained.

[0098] S103. Based on the optical response data of each suspended particle, determine the information on the change of light intensity with the relative rotation angle, and simultaneously extract the color distribution information.

[0099] Among these, the variation information consists of curves or datasets reflecting the optical anisotropy characteristics of particles. The color distribution information includes the statistical characteristics and distribution patterns of color values ​​in the color space.

[0100] S103 specifically includes:

[0101] S1031. For each suspended particle, extract the target image of the suspended particle at each rotation angle from the optical response data, and determine the light intensity value of each target image at the corresponding rotation angle.

[0102] Among them, the light intensity value refers to the quantized value representing the overall brightness of the particle region extracted from the target image.

[0103] In one specific implementation, an image segmentation algorithm is used to accurately extract the region of interest containing only the target particles from the optical response data corresponding to each rotation angle. A combined segmentation method based on edge detection and region growing is employed to ensure complete extraction of the target region. Then, the average brightness of all pixels in each target image is calculated using the following formula:

[0104]

[0105] in, Indicates rotation angle The light intensity value is given below, where N is the total number of pixels in the target image. Let be the brightness value (range 0-255) of the i-th pixel in the grayscale space. This calculation excludes the influence of the background region and focuses only on the optical properties of the particle itself.

[0106] S1032. Arrange the light intensity values ​​according to the rotation angle to generate information characterizing the change of light intensity with the rotation angle.

[0107] In one specific implementation, the calculated light intensity values ​​at each rotation angle are arranged in ascending order of angle to construct a dataset of corresponding angular light intensities. To further analyze the optical anisotropy characteristics, cubic spline interpolation is used to fit the discrete data to generate a continuous optical response curve. This curve clearly shows the periodicity of light intensity changes and the distribution of extreme points of particles at different polarization angles, providing crucial optical fingerprint information for subsequent material identification.

[0108] S1033. Synchronize and analyze the color features of each target image to generate color distribution information, wherein the color features include the frequency of occurrence and distribution range of pixel color values.

[0109] In one specific implementation, multidimensional statistical analysis is performed on each target image in the RGB color space. First, the numerical distribution range of each color channel is calculated to determine the main color intervals; then, a three-dimensional color histogram is constructed to statistically analyze the frequency of occurrence of each color value; finally, the peak color value with the highest frequency and its distribution characteristics are extracted. The color distribution range is quantified by calculating the numerical range and standard deviation of each channel, while the color frequency is characterized by a normalized histogram to ensure the comparability of statistical results between different images.

[0110] As an example, second image data were acquired for suspended particles B at eight relative rotation angles (0°, 45°, 90°, 135°, 180°, 225°, 270°, 315°).

[0111] First, in step S1031, the target image region at each angle is extracted through image segmentation, and its light intensity value is calculated. For example, at a 0° angle, the target image contains 1520 pixels, and the sum of the brightness values ​​of each pixel is 281200. According to the formula... The calculation yields: Similarly, the light intensity values ​​at other angles are calculated, and the final sequence is: [185, 168, 142, 155, 190, 175, 148, 160].

[0112] Next, in step S1032, these values ​​are arranged in angular order, and a continuous optical response curve is generated through cubic spline interpolation, such as... Figure 2 As shown, the curve indicates that the light intensity reaches a peak of 190 near 180° and a trough of 142 near 90°, exhibiting a clear periodic variation.

[0113] In step S1033, the color features of all target images are analyzed: the distribution of RGB channel values ​​is statistically analyzed, with the red channel mainly concentrated in the range of 185-215, the green channel concentrated in the range of 95-125, and the blue channel concentrated in the range of 75-105; the color histogram shows that the peak color with the highest frequency is (198, 112, 92), accounting for 18.5% of the total pixels. These data together constitute a complete optical feature description of particle B.

[0114] The above example is only one implementation of this application. In practical applications, the rotation angle interval and color statistics method can be adjusted as needed.

[0115] This application achieves comprehensive characterization of the multidimensional optical features of suspended particles by simultaneously extracting information on changes in light intensity and color distribution, providing rich criteria for accurately distinguishing different types of substances.

[0116] S104. Combining change information and color distribution information, key feature data of suspended particles are generated. The key feature data is used to characterize the optical response characteristics of suspended particles during relative rotation.

[0117] Among them, the key feature data is a multidimensional dataset formed by integrating optical response characteristics and color characteristics, which serves as the fundamental basis for subsequent material classification. The periodic component refers to the component in the light intensity variation information that exhibits regular fluctuations, reflecting the degree of optical anisotropy of the particles.

[0118] S104 specifically includes:

[0119] S1041. Extract periodic components from the change information to determine the first characteristic parameter, which is used to quantify the regularity of the periodic components.

[0120] In one specific implementation, periodic features are extracted from the light intensity variation information through spectral analysis. First, a Fourier transform is performed on the discrete light intensity sequence to identify the main frequency components. Then, the energy concentration of these components is calculated as a measure of regularity. The specific calculation of the first feature parameter is based on the normalized spectral entropy value, and the calculation formula is as follows:

[0121]

[0122] in, The first characteristic parameter, Let M be the normalized power of the k-th frequency component, and M be the total number of frequency components. The closer this parameter is to 1, the more regular the periodicity.

[0123] S1042. Analyze the distribution pattern of color distribution information in the preset color space to determine the second characteristic parameter. The second characteristic parameter is used to characterize the statistical properties of the distribution pattern.

[0124] Specifically, S1042 includes: determining a weighted color point set in a preset color space based on the frequency of occurrence in the color distribution information; calculating the spatial distribution ellipsoid of the weighted color point set and determining the principal axis direction of the spatial distribution ellipsoid; obtaining a first dimension parameter of the spatial distribution ellipsoid along the principal axis direction and a second dimension parameter perpendicular to the principal axis direction based on the distribution range in the color distribution information; and generating a second feature parameter based on the ratio of the first dimension parameter and the second dimension parameter.

[0125] The weighted color point set is a statistical sample set formed by weighting each color value according to its frequency of occurrence in the RGB color space. The spatial distribution ellipsoid is a geometric model obtained through principal component analysis that characterizes the distribution range and direction of the color point set in the three-dimensional color space. The principal axis direction refers to the longest extension direction of the spatial distribution ellipsoid, corresponding to the main direction of color distribution variation.

[0126] In one specific implementation, a weighted set of color points is first constructed based on the frequency of occurrence in the color distribution information, with the weight of each color value proportional to its frequency. Then, principal component analysis is used to calculate the covariance matrix of this weighted set, and its eigenvalues ​​and eigenvectors are solved. The eigenvector corresponding to the largest eigenvalue is the principal axis direction of the spatial distribution ellipsoid.

[0127] Next, the first dimension parameter is calculated by taking the square root of the largest eigenvalue for the distribution range along the principal axis; the second dimension parameter is calculated by taking the square root of the smallest eigenvalue for the distribution range perpendicular to the principal axis.

[0128] Finally, the ratio of the two size parameters is used as the second characteristic parameter, as shown in the following formula:

[0129]

[0130] in, The second characteristic parameter, and Let represent the largest and smallest eigenvalues ​​of the covariance matrix, respectively.

[0131] S1043. Combine the first feature parameter and the second feature parameter to generate a joint feature vector, and generate a scale value based on the relationship between the components in the joint feature vector.

[0132] Specifically, in S1043, generating a scale value based on the relationship between the components within the joint feature vector includes: using the difference between the first feature parameter and the second feature parameter within the joint feature vector as a first relationship factor; obtaining the angle between the principal axis direction of the spatial distribution ellipsoid corresponding to the second feature parameter and a preset plane as a second relationship factor; multiplying the first relationship factor and the second relationship factor to generate an intermediate value; and performing a nonlinear adjustment on the intermediate value to generate a scale value, the degree of which is determined by the first feature parameter.

[0133] The joint eigenvector is a mathematical representation formed by combining feature parameters of different dimensions. The first relational factor is used to quantify the degree of numerical difference between the first and second feature parameters. The second relational factor characterizes the spatial relationship between the main direction of color distribution and the reference plane. The scale value is a comprehensive index reflecting the degree of correlation between features.

[0134] In one specific implementation, a joint feature vector of the first feature parameter and the second feature parameter is first constructed. Then, the scale value is calculated based on the relationships between the vector's internal components. The specific calculation process includes:

[0135] The absolute difference between the first feature parameter and the second feature parameter is calculated as the first relationship factor. The angle between the principal axis direction of the determined spatial distribution ellipsoid and a preset reference plane (such as the RG plane in the RGB color space) is obtained as the second relational factor. Multiply the two factors to obtain the median value. Finally, the intermediate value Nonlinear adjustment is performed to generate scale values The degree of nonlinear adjustment is determined by the first characteristic parameter. The determined formula is as follows:

[0136]

[0137]

[0138] in, Let be the principal axis direction vector of the spatially distributed ellipsoid. Here is the normal vector of the preset reference plane, and k is the adjustment coefficient.

[0139] S1044. Combine the joint feature vector with the scale value as key feature data.

[0140] Specifically, the joint feature vector and the scale value are combined to form the final key feature data. Construct a complete three-dimensional feature description.

[0141] As an example, the periodic components are first extracted via step S1041. For instance, spectral analysis is performed on the light intensity sequence of particle B, and the normalized powers of each frequency component are calculated to be [0.35, 0.25, 0.20, 0.15, 0.05]. Substituting these values ​​into the formula, the first characteristic parameter is calculated as follows:

[0142]

[0143] Next, the color distribution pattern is analyzed in step S1042. Based on the color distribution information of particle B, a weighted color point set is constructed, where the weight of color value (198, 112, 92) is 0.185. The covariance matrix is ​​calculated through principal component analysis to obtain the eigenvalues. , The calculated second characteristic parameter is:

[0144]

[0145] Simultaneously, the principal axis direction vector is obtained. .

[0146] Then, scale values ​​are generated via step S1043. First, the degree of difference is calculated. Due to the preset reference plane normal vector Calculate the included angle Radius. Then, taking the adjustment factor k=1, the calculation is... .

[0147] Finally, key feature data were obtained through S1044 combination. .

[0148] This application establishes a feature system that can accurately characterize the essential properties of suspended particles by constructing key feature data that integrates periodicity, color distribution, and their intrinsic correlation, providing a reliable technical foundation for high-precision material classification.

[0149] S105. Based on the comparison results between the key feature data and the preset standard feature data of active pharmaceutical ingredients and excipients, the suspended particles are determined to be active pharmaceutical ingredient crystals, excipient precipitates or visible foreign matter.

[0150] The active pharmaceutical ingredient (API) standard characteristic data is a database of characteristics established through extensive testing of known midazolam crystalline samples, containing typical optical response properties of the API. Similarly, the excipient standard characteristic data is a reference database established through analysis of various excipient precipitates in the formulation.

[0151] S105 specifically includes:

[0152] S1051. Compare the key feature data corresponding to each suspended particle with the preset standard feature data of the active pharmaceutical ingredient and the standard feature data of the excipient.

[0153] S1052. When the key characteristic data matches the standard characteristic data of the active pharmaceutical ingredient, the corresponding suspended particles are identified as active pharmaceutical ingredient crystals.

[0154] S1053. When the key characteristic data matches the standard characteristic data of the excipient, the corresponding suspended particles are identified as excipient precipitates.

[0155] S1054. When the key characteristic data does not match the standard characteristic data of the active pharmaceutical ingredient and the standard characteristic data of the excipient, the corresponding suspended particles are identified as visible foreign matter.

[0156] In one specific implementation, a similarity measurement method in a multi-dimensional feature space is first used for precise comparison. Specifically, the degree of similarity is quantified by calculating the distance between the key feature data of the particle to be tested and each standard feature dataset. The formula for calculating the distance value is as follows:

[0157]

[0158] Where D is the distance value. Let i be the i-th component of the key feature data of the particle to be measured. Let D be the i-th corresponding component of the standard feature data, and n be the number of feature dimensions, where n=3. The smaller the distance value D, the higher the similarity and the greater the probability of a match.

[0159] Next, independent matching thresholds were set for the active pharmaceutical ingredient (API) and excipients, denoted as the API matching threshold. Matching threshold with excipients These two thresholds were determined through statistical analysis of a large number of known samples, reflecting the natural distribution range of their respective categories in the feature space.

[0160] The following steps, S1052-S1054, are then performed:

[0161] First, compare the distance between the microparticles to be tested and the standard characteristic data of the active pharmaceutical ingredient. ,like ≤ If so, it is immediately determined to be crystallized raw material drug; otherwise, the next step is to determine its composition.

[0162] Compare the distance between the test particles and the standard characteristic data of the excipients. ,like ≤ If so, it is determined to be excipient precipitate;

[0163] Only when > and > Only when both conditions are met can particles be classified as visible foreign matter.

[0164] This hierarchical decision structure ensures clear differentiation among the three types of particles while avoiding misjudgments that might occur due to threshold overlap. In practical applications, and The value of is usually determined based on the statistical characteristics of the feature distribution of each category. For example, it can be taken as the 95th percentile of the distance from the feature vector of that category to the cluster center. This setting method can ensure a high recognition rate for normal samples while effectively eliminating the interference of outliers.

[0165] For example, using the key feature data of particle B For example, preset raw material standard characteristic data Standard characteristic data of auxiliary materials Matching threshold .

[0166] In step S1051, the distance to the standard data of the active pharmaceutical ingredient is calculated:

[0167] Calculate the distance from the standard data for auxiliary materials:

[0168] Next, in step S1052, because < The matching conditions are met, therefore particle B is identified as a raw material crystal.

[0169] If the characteristic data is [0.38, 1.18, 0.28], the calculated values ​​are... , ,because < If so, it is determined to be excipient precipitate through step S1053.

[0170] If the particle is D, and the characteristic data is [0.10, 3.60, 0.05], the calculated values ​​are... , If both distance values ​​are greater than the threshold, then the object is determined to be a visible foreign object in step S1054.

[0171] The above example demonstrates the complete judgment process. In practical applications, the matching threshold can be adjusted according to the detection accuracy requirements. This application achieves objective and accurate identification of suspended particulate types through quantified distance calculation and threshold comparison.

[0172] This application establishes a standard feature database and adopts a distance measurement method to achieve objective and accurate determination of the type of suspended particulate matter, providing a reliable technical basis for drug quality assessment.

[0173] Optionally, after S105, the method further includes: generating time series data on the quantity change over time for each type of suspended particles; analyzing the time series data to determine the change pattern of the corresponding type of suspended particles, and predicting the quantity change of suspended particles within a preset time period based on the change pattern; and generating a graded early warning signal when the quantity change exceeds a preset threshold, wherein the graded early warning signal characterizes the change pattern and quantity change.

[0174] Time series data refers to the sequence of quantities of various types of suspended particles recorded in chronological order, reflecting the dynamic changes of different types of particles in solution. Change patterns are the regular characteristics identified through trend analysis of time series data, including key information such as growth trends and fluctuation cycles. Graded early warning signals are warnings categorized into different levels based on the severity of the change patterns, providing differentiated decision-making basis for production quality control.

[0175] In one specific implementation, after determining the type of a single suspended particle, the process detection stage begins. First, the quantity of each type of particle is counted at set time intervals to construct a time-series dataset for three types of particles: active pharmaceutical ingredient crystals, excipient precipitates, and visible foreign matter. Subsequently, moving average and trend analysis methods are used to process this time-series data to identify patterns of change, including short-term fluctuations and long-term trends.

[0176] Based on the identified patterns of change, a linear regression model is used to predict the quantity changes of various particles within a specific future time period. The prediction model comprehensively considers the trend, periodicity, and randomness components of historical data to generate predicted values ​​with confidence intervals. When the predicted value exceeds a preset threshold, the system generates a graded warning signal based on the degree of exceedance: Level 1 warning indicates a minor anomaly, Level 2 warning indicates a moderate anomaly, and Level 3 warning indicates a severe anomaly. Each warning level corresponds to specific handling suggestions and response time limits.

[0177] Taking excipient precipitates as an example, the recorded values ​​over eight consecutive detection cycles (5 minutes per cycle) were: [2, 3, 5, 8, 12, 18, 25, 35]. After eliminating random fluctuations using the moving average method, the sequence was found to exhibit an accelerating growth trend.

[0178] The linear regression model was used to predict the quantity changes over the next three cycles, yielding predicted values ​​of [48, 65, 88]. With a preset threshold of 40, the system generated a level-three warning signal because all predicted sequences exceeded the limit and grew rapidly, indicating that the risk of excipient precipitation had reached a severe level and immediate process adjustment measures were required.

[0179] The time intervals, warning level classifications, and thresholds in the above examples can all be adjusted according to specific process requirements.

[0180] The aforementioned method, belonging to a quality inspection method of this application, typically performs independent quantity statistics and threshold alarms for each type of particle. However, this approach has significant limitations. It assumes that the generation of different types of particles is an independent random event. In reality, there may be an inherent causal relationship or temporal induction relationship between active pharmaceutical ingredient crystals, excipient precipitates, and foreign matter. This isolated detection mode cannot capture the dynamic interactions between different types of particles, causing the system to only respond passively after a problem appears, lacking the ability to identify potential risks early.

[0181] To overcome this deficiency, this method further proposes that, after S105, it also includes: recording the timestamp information of suspended particles identified as active pharmaceutical ingredient crystals, excipient precipitates, and visible foreign matter, respectively, to generate time series data of at least three types of particles; calculating the cross-correlation function between the time series data of different types of particles to determine at least one type of particle as a leading indicator; and generating an early warning signal characterizing the precipitation risk of the corresponding type of particles when the growth rate of the number of leading indicator particles exceeds a preset rate.

[0182] Based on the classification of individual particles, it has established a risk warning mechanism based on time-series correlation analysis.

[0183] Among them, the leading indicator refers to the leading indicator whose changing trend in multiple related time series can predict the changes in other series in advance. In this scheme, it specifically refers to the particle category that appears earlier than other types in time.

[0184] In one specific implementation, such as Figure 3 As shown, each time suspended particles are detected, not only their type is recorded, but also the exact timestamp of their appearance is precisely recorded. These data are categorized into three independent time series: active pharmaceutical ingredient crystals, excipient precipitates, and visible foreign matter. Each series records the frequency of occurrence of that type of particle in chronological order.

[0185] Then, the cross-correlation functions between each of these three types of time series are calculated. The cross-correlation function is calculated using the following formula:

[0186]

[0187] in, and These represent discrete time series of two different types of particles, where n is the time index and k is the time offset (lag or lead). The analysis involves cross-correlation functions. The peak value and its corresponding k value can be used to determine the order of change between the two types of particles. The particle type with the largest positive correlation and a negative time offset is identified as the leading indicator.

[0188] For example, if If the maximum positive peak value is achieved at k=-2, it indicates that the changing trend of type A particles leads that of type B particles by an average of 2 time periods. Therefore, type A particles are a leading indicator of type B particles.

[0189] After determining the leading indicator, the growth rate of this type of particle population is continuously monitored. The growth rate is calculated using a sliding window method, based on the population changes over the most recent monitoring periods. The calculation formula is as follows:

[0190]

[0191] Where r is the growth rate, This refers to the number of leading indicator particles within the current detection cycle. This represents the number of particles in the previous detection period. This formula quantifies the drastic change in quantity.

[0192] When the growth rate *r* of the leading indicator particles continuously exceeds a preset rate threshold, the system immediately generates an early warning signal. This signal not only indicates the current risk status but also includes detailed information such as the risk type, development trend, and expected scope of impact, providing quality control personnel with sufficient decision-making basis.

[0193] Assuming the system counted the number of three types of particles over 10 consecutive detection cycles (each cycle being 1 minute), the following time series data was generated:

[0194] Active pharmaceutical ingredient crystallization sequence : [1, 1, 2, 2, 3, 5, 8, 13, 20, 30];

[0195] Excipient precipitate sequence : [2, 2, 5, 8, 12, 18, 25, 35, 45, 55];

[0196] Visible foreign object sequence [0, 1, 1, 1, 2, 2, 3, 4, 5, 7].

[0197] The excipient precipitate sequence was discovered by calculating the cross-correlation function. Crystallization sequence of active pharmaceutical ingredient The correlation coefficient reached a peak of 0.97 at a time offset of k=-2, which indicates that the excipient precipitate is a leading indicator of the crystallization of the active pharmaceutical ingredient, and its changes lead by an average of 2 cycles.

[0198] Then, the growth rate of this leading indicator is examined according to the formula:

[0199]

[0200] The most recent period was calculated ( =55, =45) growth rate When this value exceeds the preset threshold of 20%, the system immediately generates an early warning signal, indicating that the risk of raw material crystallization will increase significantly.

[0201] This application enables early identification and warning of quality risks by exploring the inherent temporal correlations between different types of particles, providing a critical time window for timely adjustments to production processes.

[0202] Figure 4 This is a schematic diagram illustrating a specific embodiment of an automated optical detection system for visible foreign matter in a midazolam buccal mucosa solution provided in this application. (Refer to...) Figure 4 The system may include:

[0203] The observation module 41 is used to control the flow of midazolam buccal mucosa solution in a transparent flow cell and to use an optical imaging system to dynamically observe the solution in the flow state in order to identify suspended particles in the midazolam buccal mucosa solution.

[0204] The acquisition module 42 is used to control the polarization light source in the optical imaging system to rotate relative to the transparent flow cell. During the relative rotation, it acquires optical response data of each suspended particle at multiple rotation angles.

[0205] Extraction module 43 is used to determine the change information of light intensity with relative rotation angle based on the optical response data of each suspended particle, and simultaneously extract the color distribution information;

[0206] The generation module 44 is used to combine the change information and the color distribution information to generate key feature data of the suspended particles. The key feature data is used to characterize the optical response characteristics of the suspended particles during relative rotation.

[0207] The determination module 45 is used to determine the suspended particles as active pharmaceutical ingredient crystals, excipient precipitates, or visible foreign matter based on the comparison results between the key feature data and the preset standard feature data of active pharmaceutical ingredient and excipient.

[0208] The automated optical detection system for visible foreign matter in midazolam buccal mucosa solution according to this application is used to implement the aforementioned automated optical detection method for visible foreign matter in midazolam buccal mucosa solution. Therefore, the specific implementation of the automated optical detection system for visible foreign matter in midazolam buccal mucosa solution can be found in the embodiment section of the automated optical detection method for midazolam buccal mucosa solution above. The specific implementation can be referred to the description of the corresponding embodiments, and will not be repeated here.

[0209] This application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the above-described automatic optical detection method for visible foreign matter in a midazolam buccal mucosa solution.

[0210] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described automatic optical detection method for visible foreign matter in a midazolam buccal mucosa solution.

[0211] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0212] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above embodiments of the automatic optical detection method for visible foreign matter in a midazolam buccal mucosa solution.

[0213] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0214] The above provides a detailed description of the automated optical detection method and system for visible foreign matter in a midazolam buccal mucosa solution. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of these embodiments are merely illustrative of the method and its core concepts. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. An automated optical detection method for visible foreign matter in a midazolam buccal mucosa solution, characterized in that, include: The flow of midazolam buccal mucosa solution was controlled in a transparent flow cell, and the solution under flow was dynamically observed using an optical imaging system to identify suspended particles in the midazolam buccal mucosa solution. The polarized light source in the optical imaging system is controlled to rotate relative to the transparent flow cell. During the relative rotation, optical response data of each suspended particle is collected at multiple rotation angles. Based on the optical response data of each suspended particle, the information on the change of light intensity with the relative rotation angle is determined, and the color distribution information is extracted simultaneously. By combining the change information and the color distribution information, key feature data of the suspended particles are generated. The key feature data is used to characterize the optical response characteristics of the suspended particles during relative rotation. Based on the comparison results between the key feature data and the preset standard feature data of active pharmaceutical ingredients and excipients, the suspended particles are determined to be active pharmaceutical ingredient crystals, excipient precipitates or visible foreign matter. The process of combining the change information with the color distribution information to generate key feature data of the suspended particles includes: Periodic components are extracted from the change information to determine a first feature parameter, which is used to quantify the regularity of the periodic components. The distribution pattern of the color distribution information in a preset color space is analyzed to determine the second feature parameter, which is used to characterize the statistical properties of the distribution pattern. The first feature parameter and the second feature parameter are combined to generate a joint feature vector, and a scale value is generated based on the relationship between the components in the joint feature vector. The joint feature vector is combined with the scaling value to form key feature data.

2. The method according to claim 1, characterized in that, After determining that the suspended particles are active pharmaceutical ingredient crystals, excipient precipitates, or visible foreign matter, the method further includes: The timestamp information of suspended particles identified as active pharmaceutical ingredient crystals, excipient precipitates, and visible foreign matter is recorded respectively to generate time series data of at least three types of particles; Calculate the cross-correlation function between the time series data of different types of particles to determine at least one particle type as a leading indicator; When the growth rate of the number of particles in the leading indicator exceeds a preset rate, a warning signal representing the risk of precipitation of the corresponding type of particles is generated.

3. The method according to claim 1, characterized in that, After determining that the suspended particles are active pharmaceutical ingredient crystals, excipient precipitates, or visible foreign matter, the method further includes: For each type of suspended particulate matter, time-series data on the change in quantity over time is generated; Analyze the time series data to determine the change pattern of the corresponding type of suspended particles, and predict the change in the number of suspended particles within a preset time period based on the change pattern; When the quantity change exceeds a preset threshold, a graded early warning signal is generated, which represents the change pattern and quantity change.

4. The method according to claim 1, characterized in that, The analysis of the color distribution information in the preset color space to determine the second feature parameter includes: Based on the frequency of occurrence in the color distribution information, a weighted color point set is determined in a preset color space; Calculate the spatial distribution ellipsoid of the weighted color point set, and determine the principal axis direction of the spatial distribution ellipsoid; Based on the distribution range in the color distribution information, obtain the first dimension parameter of the spatial distribution ellipsoid along the main axis direction and the second dimension parameter perpendicular to the main axis direction; A second feature parameter is generated based on the ratio of the first size parameter to the second size parameter.

5. The method according to claim 1, characterized in that, The step of generating a scale value based on the relationship between the components within the joint feature vector includes: The difference between the first feature parameter and the second feature parameter in the joint feature vector is used as the first relationship factor. Obtain the angle between the principal axis direction of the spatial distribution ellipsoid corresponding to the second feature parameter and the preset plane, and use it as the second relationship factor; The first relation factor and the second relation factor are multiplied to generate an intermediate value; The intermediate value is nonlinearly adjusted to generate the scale value, and the degree of nonlinear adjustment is determined by the first feature parameter.

6. The method according to claim 1, characterized in that, The step of determining the change in light intensity with relative rotation angle based on the optical response data of each suspended particle, and simultaneously extracting color distribution information, includes: For each suspended particle, the target image of the suspended particle at each rotation angle is extracted from the optical response data, and the light intensity value of each target image at the corresponding rotation angle is determined; Arrange the light intensity values ​​according to the rotation angle to generate information characterizing the change of light intensity with the rotation angle; Simultaneously, the color features of each target image are analyzed to generate color distribution information, the color features including the frequency and distribution range of pixel color values.

7. The method according to claim 1, characterized in that, The dynamic observation of the solution in a flowing state to identify suspended particles in the midazolam buccal mucosa solution includes: Dynamic observation was conducted on the buccal mucosa solution of midazolam in a flowing state to obtain image data; Analyze the image data to determine the initial region where the difference between the brightness value and the background exceeds a preset threshold; The initial region is filtered to obtain candidate regions whose geometric features meet preset conditions; The candidate regions are tracked in the image data, and the candidate regions that have shifted are identified as suspended particles.

8. The method according to claim 1, characterized in that, The step of determining whether the suspended particles are drug crystals, excipient precipitates, or visible foreign matter based on the comparison results between the key feature data and preset standard feature data of active pharmaceutical ingredients and excipients includes: The key feature data corresponding to each suspended particle is compared with the preset standard feature data of active pharmaceutical ingredient and standard feature data of excipient. When the key feature data matches the standard feature data of the active pharmaceutical ingredient, the corresponding suspended particles are identified as active pharmaceutical ingredient crystals. When the key feature data matches the standard feature data of the excipient, the corresponding suspended particles are identified as excipient precipitates. When the key feature data does not match the standard feature data of the active pharmaceutical ingredient and the standard feature data of the excipient, the corresponding suspended particles are identified as visible foreign matter.

9. An automated optical detection system for visible foreign matter in a midazolam buccal mucosa solution, characterized in that, include: The observation module is used to control the flow of midazolam buccal mucosa solution in a transparent flow cell and to use an optical imaging system to dynamically observe the solution in the flow state in order to identify suspended particles in the midazolam buccal mucosa solution. The acquisition module is used to control the relative rotation between the polarized light source in the optical imaging system and the transparent flow cell. During the relative rotation, the module acquires optical response data of each suspended particle at multiple rotation angles. The extraction module is used to determine the change information of light intensity with relative rotation angle based on the optical response data of each suspended particle, and simultaneously extract the color distribution information; A generation module is used to combine the change information and the color distribution information to generate key feature data of the suspended particles. The key feature data is used to characterize the optical response characteristics of the suspended particles during relative rotation. The determination module is used to determine the suspended particles as active pharmaceutical ingredient crystals, excipient precipitates, or visible foreign matter based on the comparison results between the key feature data and the preset standard feature data of active pharmaceutical ingredients and excipients. The process of combining the change information with the color distribution information to generate key feature data of the suspended particles includes: Periodic components are extracted from the change information to determine a first feature parameter, which is used to quantify the regularity of the periodic components. The distribution pattern of the color distribution information in a preset color space is analyzed to determine the second feature parameter, which is used to characterize the statistical properties of the distribution pattern. The first feature parameter and the second feature parameter are combined to generate a joint feature vector, and a scale value is generated based on the relationship between the components in the joint feature vector. The joint feature vector is combined with the scaling value to form key feature data.

Images