A medicine decoction piece dispensing post-checking system

Abstract

The application belongs to the technical field of data processing of medicinal decoction pieces, and discloses a medicinal decoction piece rechecking system after dispensing, which comprises a multispectral scanning module, a feature fusion module, an error compensation module, an analysis and comparison module and a decision module. The multispectral scanning module is used for collecting surface image data of six different visual angles in the rotating state of medicinal decoction pieces. The surface image data comprises morphological feature data and chemical component identification feature data. The feature fusion module can combine the surface image data with a stereo matching algorithm, construct a high-precision three-dimensional point cloud model, determine a 128-dimensional feature vector of the overall feature of medicinal decoction pieces, and convert the near-infrared spectrum data calculation result into a pseudo-color heat map superimposed on the surface of the three-dimensional model. The multispectral scanning module and the feature fusion module can comprehensively and accurately represent the characteristics of medicinal decoction pieces, effectively distinguish the same-form different medicinal decoction pieces, and simultaneously perform multi-source data fusion and analysis, thereby improving the accuracy and reliability of the overall rechecking.

Description

Technical Field

[0001] This application belongs to the field of data processing technology for the verification of medicinal decoction pieces, specifically relating to a verification system for medicinal decoction pieces after dispensing. Background Technology

[0002] Prepared medicinal herbs are an important component of traditional Chinese medicine (TCM) and are Chinese medicinal materials that have been processed according to TCM theories and methods, making them directly applicable in clinical practice. The accuracy of their dispensing directly affects the patient's treatment outcome and medication safety. During the dispensing process, various prepared medicinal herbs need to be allocated according to the prescription in specific proportions and then placed or packaged together. This process is carried out by a pharmacist who dispenses the herbs according to the prescription and verifies the dispensing process afterward. Therefore, a post-dispensing verification system for prepared medicinal herbs is necessary.

[0003] However, most current post-dispensing verification systems for medicinal herbs rely solely on RGB image recognition, which cannot distinguish between herbs of the same shape but different textures. Furthermore, the verification mechanism is based on a relatively singular data dimension and lacks multi-source data fusion analysis capabilities, which affects the overall accuracy and reliability of the verification process. Summary of the Invention

[0004] This application provides a post-dispensing verification system for medicinal decoction pieces, aiming to solve the problems that existing technologies cannot distinguish between homogeneous and heterogeneous medicinal decoction pieces, and that the verification mechanism is based on relatively single data dimensions and lacks multi-source data fusion analysis functions, which affects the accuracy and reliability of the overall verification.

[0005] A post-dispensing verification system for medicinal decoction pieces includes a multispectral scanning module, a feature fusion module, an error compensation module, an analysis and comparison module, and a decision module. The multispectral scanning module is used to acquire surface image data from six different perspectives while the medicinal decoction pieces are rotating. The surface image data includes morphological feature data and chemical component identification feature data.

[0006] The feature fusion module can combine surface image data with stereo matching algorithms to construct a high-precision three-dimensional point cloud model, determine the 128-dimensional feature vector of the overall features of the medicinal slices, and convert the near-infrared spectral data calculation results into a pseudo-color heat map superimposed on the surface of the three-dimensional model.

[0007] The error compensation module is used to collect the weight data of the medicinal slices in real time, and perform air pressure compensation and temperature compensation processing on the weight data according to the preset algorithm, output the weight value, and compare it with the theoretical value of the prescription.

[0008] The analysis and comparison module can receive 128-dimensional feature vectors and, through multi-layer cascaded matching and deep learning technology, identify medicinal slices that are similar in shape but different in material, and determine the final type of medicinal slices.

[0009] The decision-making module can analyze prescription components in real time, automatically generate a temporary graph structure containing medicinal material nodes, compatibility relationships and attributes, analyze potential risks, and trigger corresponding alarms.

[0010] Furthermore, the multispectral scanning module includes a rotating stage, a ring-shaped LED cold light source, a visible light camera, and a near-infrared spectrometer. The multispectral scanning module also includes a prescription input unit for importing data on the compatibility of medicinal herbs.

[0011] Furthermore, the feature fusion module includes a preprocessing unit, a construction unit, and a fusion unit;

[0012] The preprocessing unit is used to perform radiometric correction, noise reduction, and lens distortion correction on the surface image data.

[0013] The construction unit is used to perform three-dimensional reconstruction and feature extraction on the acquired surface image data;

[0014] The fusion unit is used to receive morphological feature data extracted by the image processing unit and spectral feature data generated by near-infrared spectral processing, and finally obtain a 128-dimensional feature vector.

[0015] Furthermore, the specific content of the building unit is as follows:

[0016] Generate point cloud model:

[0017] 1) An improved semi-global matching algorithm (SGM) is used based on camera calibration parameters to construct an energy function by aggregating path costs, which is then used to evaluate the quality of the current disparity map D.

[0018]

[0019] Among them, C data (p,d p ) represents the distance of pixel p in parallax d p Brightness similarity under P smooth V is a coefficient. smooth (p, d) p Pixel p in parallax d p The gradient difference is used to constrain the disparity continuity between adjacent pixels.

[0020] 2) Convert the disparity value of each pixel to 3D point coordinates in the world coordinate system. The conversion formula is as follows:

[0021]

[0022] Where (x, y) are the image pixel coordinates, (c x c y f is the pixel position of the camera's principal point. x ,fy These are the focal lengths in the x and y directions, respectively, and Z is the depth value.

[0023] Furthermore, the specific content of the error compensation module includes weight detection, air pressure compensation, temperature compensation, and parameter comparison.

[0024] Furthermore, the analysis and comparison module includes an association unit and a verification unit. The association unit uses the K-nearest neighbor algorithm to calculate the Euclidean distance D between the features of the medicinal slices to be matched and the features of each sample in the standard library. i The calculation formula is as follows:

[0025]

[0026] Where, x j Let y be the j-th dimension characteristic value of the medicinal decoction piece to be tested. ij Let be the j-th eigenvalue of the i-th sample in the standard library, and n be a 128-dimensional eigenvector.

[0027] Furthermore, the verification unit is used to inject the output potential medicinal herb types into a pre-trained convolutional neural network model for secondary verification to determine the final medicinal herb types.

[0028] Furthermore, the specific content of the building unit also includes near-infrared spectral data mapping. By selecting a standard sample set of typical active ingredients, a partial least squares model is established to calculate the target component concentration vector. Then, the target component concentration vector calculation result is spatially registered with the point cloud model, a pseudo-color heat map is rendered in real time, and the chemical composition value of discrete measurement points is extended to the entire surface using the nearest neighbor interpolation algorithm.

[0029] Compared with the prior art, this application has at least the following beneficial effects:

[0030] Based on further analysis and research of existing technical problems, this application acquires surface image data from six different perspectives in the rotating state of medicinal slices using a multispectral scanning module. This data covers morphological and chemical composition identification features. Combined with a feature fusion module, a high-precision three-dimensional point cloud model is constructed and a 128-dimensional feature vector is determined. Furthermore, near-infrared spectral data is converted into a pseudo-color thermal map and superimposed on the model surface. This approach can comprehensively and accurately characterize the properties of medicinal slices, effectively distinguish between homogeneous and heterogeneous medicinal slices, and greatly improve the accuracy of verification.

[0031] Meanwhile, the error compensation module collects weight data in real time and performs air pressure and temperature compensation, comparing it with the theoretical values ​​of the prescription. The analysis and comparison module uses multi-level matching and deep learning technology to identify the types of medicinal slices. The decision module parses the prescription components to generate a temporary graph structure and analyzes potential risks to trigger alarms. This provides rich data support and scientific judgment basis for the review process, significantly enhancing the reliability and stability of the overall review. It can perform multi-source data fusion and analysis, improving the accuracy and reliability of the overall review. Attached Figure Description

[0032] Figure 1 This is a schematic diagram illustrating the application environment of a post-dispensing verification system for medicinal decoction pieces, provided as an embodiment of this application. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments.

[0034] like Figure 1 As shown, the present application provides a post-dispensing verification system for medicinal decoction pieces, which includes a multispectral scanning module, a feature fusion module, an error compensation module, an analysis and comparison module, and a decision module.

[0035] The multispectral scanning module is used to acquire surface image data from six different perspectives while the medicinal slices are rotating, providing rich and detailed information support for subsequent accurate analysis of the appearance quality of the medicinal slices. The surface image data includes morphological feature data and chemical component identification feature data. The multispectral scanning module includes a rotating stage, a ring LED cold light source, a visible light camera, and a near-infrared spectrometer.

[0036] The rotating stage is used to place the medicinal slices to be reviewed. Driven by a motor, it rotates at a preset speed to ensure that the subsequent acquisition process is triggered each time it rotates to a predetermined angle (each 60° is a sampling interval). This design allows the medicinal slices to be exposed to the scanning system at different angles during the rotation, providing a basis for subsequent multi-angle feature acquisition.

[0037] A ring-shaped LED cold light source is positioned around the stage to provide uniform illumination for the medicinal slices. The array of ring-shaped LED cold light sources creates a shadowless, low-heat-radiation, uniform lighting environment. This facilitates subsequent use with a visible light industrial camera to continuously capture surface images from six different angles at a resolution of 0.1 mm / pixel.

[0038] Visible light cameras are used to acquire surface images of medicinal slices from multiple angles, which facilitates the subsequent complete reconstruction of the microscopic morphological features of the medicinal slice surface, including edge curvature, crack orientation, and texture distribution parameters.

[0039] Near-infrared spectrometers are used to collect spectral data from the surface of medicinal slices; they are equipped with fiber optic probes that synchronously scan the sample surface along the same rotational trajectory. Within the 900-1700 nm wavelength range, absorption lines at 50 characteristic wavelengths are collected at preset intervals, focusing on the characteristic peak regions of flavonoids and alkaloids. This facilitates subsequent spatial quantitative characterization of the chemical composition distribution.

[0040] The multispectral scanning module also includes a prescription input unit for importing medicinal herb compatibility data. This data covers the proportions and contraindications of different medicinal herbs. The imported compatibility data provides crucial reference for subsequent multispectral scanning and analysis, ensuring the scanning module can accurately acquire and analyze the morphological and chemical characteristics of the medicinal herbs based on correct compatibility standards.

[0041] The feature fusion module combines surface image data with stereo matching algorithms to construct a high-precision 3D point cloud model, determine the 128-dimensional feature vector of the overall characteristics of the medicinal slices, and convert the near-infrared spectral data calculation results into a pseudo-color heat map superimposed on the surface of the 3D model to achieve visualized correlation analysis between morphology and chemical composition, thereby improving recognition capabilities. The feature fusion module includes a preprocessing unit, a construction unit, and a fusion unit.

[0042] The preprocessing unit is used to perform radiometric correction, noise reduction, and lens distortion correction on surface image data. It can improve the recognition of parameters such as surface texture and edge contour, ensure that feature data and chemical composition identification feature data can be captured with high precision, and reduce the risk of misjudgment caused by imaging defects.

[0043] Radiometric correction analyzes the brightness value of each pixel in the image based on a pre-defined standard illumination model. By comparing the brightness distribution of the actual acquired image with the standard model, the brightness deviation value of each pixel caused by uneven light source is obtained. Then, the image is adjusted as a whole based on the deviation value to achieve a relatively uniform brightness in various regions of the image, thereby eliminating the impact of uneven light source on image quality.

[0044] The denoising process employs a nonlocal mean filtering method, which searches for pixel blocks similar to the current pixel across the entire image. By comparing the similarity between different pixel blocks, blocks with high similarity are assigned higher weights, while those with low similarity are assigned lower weights. Then, a weighted average is calculated for the similar pixel blocks based on these weights, and the resulting average value replaces the value of the current pixel. This method can remove noise while better preserving image details and edge information, making the surface image of the medicinal slices clearer and smoother, highlighting their true morphological features.

[0045] The preprocessing unit stores lens-specific distortion parameters. During lens distortion correction, the coordinates of each pixel in the image are transformed according to the distortion parameters, and then the grayscale value or color information of that pixel is remapped to the correct position. This results in a more realistic image of the surface of the medicinal slices.

[0046] The construction unit is used to perform 3D reconstruction and feature extraction on the acquired surface image data. Details are as follows:

[0047] a) Generate point cloud model

[0048] 1) An improved semi-global matching algorithm (SGM) is used based on camera calibration parameters to construct an energy function by aggregating path costs, which is then used to evaluate the quality of the current disparity map D.

[0049]

[0050] Among them, C data (p, d) p ) represents the distance of pixel p in parallax d p Brightness similarity under P sumooth V is a coefficient. smooth (p, d) p Pixel p in parallax d p The gradient difference is used to constrain the disparity continuity between adjacent pixels.

[0051] 2) Convert the disparity value of each pixel to 3D point coordinates in the world coordinate system. The conversion formula is as follows:

[0052]

[0053] Where (x, y) are the image pixel coordinates, (c x c y f is the pixel position of the camera's principal point. x f y These are the focal lengths in the x and y directions, respectively, and Z is the depth value.

[0054] 3) By voxelized mesh downsampling and outlier removal (based on statistical filtering), a point cloud model with a density of ≥2000 points / cm2 is generated, with the spacing between adjacent points ≤0.05mm, thus fully preserving the microscopic undulation structure of the surface of the medicinal slices.

[0055] The Poisson surface reconstruction algorithm is applied to transform discrete point clouds into continuous triangular mesh models, and hole regions are repaired through normal consistency verification. For the unique thin-sheet structure of traditional Chinese medicine decoction pieces, a curvature-guided mesh subdivision strategy is introduced to automatically increase topological resolution in densely wrinkled areas, achieving precise representation of millimeter-level detailed features.

[0056] b) Near-infrared spectral data mapping

[0057] 1) Establish a partial least squares model using a standard sample set of typical active ingredients, calculate the target ingredient concentration vector, determine the optimal number of principal components through cross-validation, and then substitute the spectrum of the unknown sample into the model to predict its component content. The calculation formula is as follows:

[0058] C = XB + E

[0059] Where X is the spectral matrix, each row of which represents the characteristic response of a sample across all wavelength channels, B is the regression coefficient vector, reflecting the contribution weight of each wavelength to the component concentration, and E is the residual value.

[0060] 2) Spatial registration is performed between the target component concentration vector calculation results and the point cloud model, a pseudo-color heat map is rendered in real time, and the nearest neighbor interpolation algorithm is used to extend the chemical composition values ​​of discrete measurement points to the entire surface.

[0061] The fusion unit receives morphological feature data extracted by the image processing unit and spectral feature data generated by near-infrared spectroscopy processing. Then, a deep learning-based encoder fusion algorithm is used to construct a dual-branch autoencoder structure. One branch processes morphological features and the other processes spectral features. During the encoding process, each branch learns and extracts key features. Attention mechanism is introduced to automatically assign different weights based on the importance of the features to the quality of the medicinal slices. The features encoded by the two branches are then concatenated and then dimensionality reduced and optimized through a common decoder to remove redundant information. Finally, the fusion results in a 128-dimensional feature vector that can comprehensively and accurately represent the overall features of the medicinal slices.

[0062] The error compensation module is used to collect the weight data of the medicinal slices in real time, and performs air pressure compensation and temperature compensation processing on the weight data according to a preset algorithm, outputting the weight value and comparing it with the theoretical value of the prescription. Details are as follows:

[0063] 1) Weight inspection

[0064] The weight data of the medicinal slices are collected in real time by strain gauge load cells, and the ambient air pressure value is obtained by air pressure sensor and temperature sensor respectively, and the current ambient temperature is detected.

[0065] 2) Air pressure compensation

[0066] A buoyancy correction model is established based on Archimedes' principle: Δm = ρa·V / g, where ρa is the current air density, V is the volume of the medicinal slices, and g is the local gravitational acceleration. The current air density ρa can be quickly obtained using a lookup table to determine the ρa value at the corresponding air pressure. When a sudden change in air pressure is detected (such as a change in airflow caused by the sudden opening of doors or windows), the system activates dynamic tracking mode, recalculates the buoyancy under the new air pressure within 50ms, and adds the compensation amount to the original weighing result in real time.

[0067] 3) Temperature compensation

[0068] Temperature sensors (measuring range -40℃ to +125℃, resolution 0.01℃) are symmetrically positioned at the four corners of the weighing platform, forming a differential temperature measurement network. Real-time temperature values ​​at each measuring point are synchronously acquired at a frequency of 10Hz, constructing a two-dimensional temperature field distribution map. A finite element model is established based on the Fourier heat conduction equation to calculate the thermal deformation of the metal components caused by temperature difference, ΔL = α·L0·ΔT, where α is the coefficient of linear expansion of the material and L0 is the original dimension. The power output of the platform's built-in semiconductor cooling / heating module is dynamically adjusted by a PID controller to control the temperature difference of key stress-bearing structures within ±0.5℃. Temperature change curves are recorded simultaneously as a traceability basis. After dual-channel compensation, the weight reading fluctuation range can be controlled within ±0.001g, improving weighing accuracy.

[0069] 4) Parameter Comparison

[0070] The measured net weight data after compensation is input into the comparison unit. The comparison unit can analyze the theoretical compatibility parameters in the electronic prescription and compare the measured value with the theoretical value of the prescription. When the relative deviation between the measured value and the theoretical value of the prescription exceeds ±0.5%, the graded early warning mechanism is immediately triggered and a review suggestion is generated.

[0071] The analysis and comparison module can receive 128-dimensional feature vectors and, through multi-layer cascaded matching and deep learning technology, identify medicinal slices that are similar in shape but different in material, thus determining the final type of medicinal slice and effectively improving the accuracy and traceability of medicinal slice identification.

[0072] The analysis and comparison module includes an association unit and a verification unit. The association unit uses the K-nearest neighbor algorithm, where K is set to 5. This algorithm calculates the Euclidean distance D between the feature of the medicinal decoction piece to be matched and each sample feature in the standard library, based on a large amount of homogeneous and heterogeneous medicinal decoction piece feature data pre-stored in the standard library. i The algorithm calculates and retrieves the five closest samples by Euclidean distance from the standard library, forming a candidate set. The calculation formula is as follows:

[0073]

[0074] Where, x j y represents the j-th dimension feature value of the medicinal slice to be tested (such as morphological feature data and chemical component identification feature data). ij Let be the j-th eigenvalue of the i-th sample in the standard library, and n be a 128-dimensional feature vector. Euclidean distance can intuitively reflect the proximity of two feature vectors in space; the closer the distance, the more similar the features.

[0075] The cosine similarity method was used to compare the candidate sets, focusing on the correlation coefficient between the texture features and spectral curves of the medicinal slices. Cosine similarity measures the degree of similarity between two vectors by calculating the cosine of the angle between them, with a value ranging from -1 to 1. The closer the value is to 1, the higher the similarity.

[0076] Texture features and spectral curves are important characteristics of medicinal decoction pieces, reflecting their appearance and chemical composition. By calculating the cosine similarity between the candidate set and the sample to be matched in terms of the correlation coefficients of texture features and spectral curves, the samples in the candidate set can be reordered, placing samples with higher similarity at the top, thus outputting the types of medicinal decoction pieces and providing a more valuable reference for the final decision.

[0077] The validation unit is used to inject the output potential medicinal herb types into a pre-trained convolutional neural network model for secondary validation. The convolutional neural network model adopts a ResNet-50 residual structure, and the input layer is a tensor concatenated from a grayscale 3D point cloud projection image and a straightened spectral sequence. Pre-trained weights from a publicly available dataset of Chinese medicinal herb images are loaded through transfer learning, and the parameters of the fully connected layers are fine-tuned using labeled real and fake samples. The model outputs a confidence score, which is then weighted and summed with the result from the rule engine to determine the final medicinal herb type.

[0078] The decision-making module can analyze prescription components in real time, automatically generate a temporary graph structure containing medicinal herb nodes, compatibility relationships, and attributes, analyze potential risks, and trigger corresponding alarms. Details are as follows:

[0079] The system receives the final list of medicinal herbs and initiates a natural language processing engine. A bidirectional LSTM+CRF model is used to accurately identify the herb names, dosages, and processing requirements. Based on an ontology framework, the system automatically maps these information to the TCM concept system, generating a temporary knowledge graph containing entity nodes (herbs), relational edges (compatibility types), and attribute tags.

[0080] Graph Convolutional Networks (GCNs) delve into the interaction patterns and potential rules between nodes in a knowledge graph to comprehensively evaluate the current prescription. Based on the reasoning results, the corresponding level of audible and visual alarm devices are activated.

[0081] For example, if the appearance of a single herb is deemed suspicious, a yellow indicator light will flash regularly to alert the operator; when the weight exceeds the limit, an orange alarm will light up and a buzzer will sound; if any incompatibilities are found, a red emergency stop will be triggered to lock the equipment and prevent potentially serious consequences.

[0082] In the aforementioned system for verifying the dispensing of medicinal herbs, a multispectral scanning module collects surface image data from six different perspectives while the herbs are rotating. This data covers morphological and chemical composition identification features. Combined with a feature fusion module, a high-precision three-dimensional point cloud model is constructed, and a 128-dimensional feature vector is determined. Furthermore, near-infrared spectral data is converted into a pseudo-color thermal map and superimposed on the model surface. This system can comprehensively and accurately characterize the properties of medicinal herbs, effectively distinguish between herbs of the same shape but different textures, and greatly improve the accuracy of verification.

[0083] Meanwhile, the error compensation module collects weight data in real time and performs air pressure and temperature compensation, comparing it with the theoretical value of the prescription. The analysis and comparison module uses multi-level matching and deep learning technology to identify the types of medicinal slices. The decision module parses the prescription components to generate a temporary graph structure and analyzes potential risks to trigger alarms, providing rich data support and scientific judgment basis for the review process, significantly enhancing the reliability and stability of the overall review.

[0084] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A system for verifying the dispensing of medicinal decoction pieces, characterized in that, It includes a multispectral scanning module, a feature fusion module, an error compensation module, an analysis and comparison module, and a decision module. The multispectral scanning module is used to acquire surface image data from six different perspectives while the medicinal slices are rotating. The surface image data includes morphological feature data and chemical component identification feature data. The feature fusion module can combine surface image data with stereo matching algorithms to construct a high-precision three-dimensional point cloud model, determine the 128-dimensional feature vector of the overall features of the medicinal slices, and convert the near-infrared spectral data calculation results into a pseudo-color heat map superimposed on the surface of the three-dimensional model. The feature fusion module includes a preprocessing unit, a construction unit, and a fusion unit; The preprocessing unit is used to perform radiometric correction, noise reduction, and lens distortion correction on the surface image data. The construction unit is used to perform three-dimensional reconstruction and feature extraction on the acquired surface image data; The fusion unit is used to receive morphological feature data extracted by the image processing unit and spectral feature data generated by near-infrared spectral processing, and finally obtain a 128-dimensional feature vector. The specific content of the building unit is as follows: Generate point cloud model: 1) An improved semi-global matching algorithm is used based on camera calibration parameters to construct an energy function by aggregating path costs, which is then used to evaluate the quality of the current disparity map D: in, For pixel p in parallax Brightness similarity under, For coefficients, Pixel p in parallax The gradient difference is used to constrain the disparity continuity between adjacent pixels. 2) Convert the disparity value of each pixel to 3D point coordinates in the world coordinate system. The conversion formula is as follows: in, Image pixel coordinates, The pixel position of the camera's principal point. These are the focal lengths in the x and y directions, respectively, and Z is the depth value; The error compensation module is used to collect the weight data of the medicinal slices in real time, and perform air pressure compensation and temperature compensation processing on the weight data according to the preset algorithm, output the weight value, and compare it with the theoretical value of the prescription. The analysis and comparison module can receive 128-dimensional feature vectors and, through multi-layer cascaded matching and deep learning technology, identify medicinal slices that are similar in shape but different in material, and determine the final type of medicinal slices. The decision-making module can analyze prescription components in real time, automatically generate a temporary graph structure containing medicinal material nodes, compatibility relationships and attributes, analyze potential risks, and trigger corresponding alarms.

2. The system for verifying the dispensing of medicinal decoction pieces according to claim 1, characterized in that, The multispectral scanning module includes a rotating stage, a ring-shaped LED cold light source, a visible light camera, and a near-infrared spectrometer. The multispectral scanning module also includes a prescription input unit for importing data on the compatibility of medicinal herbs.

3. The system for verifying the dispensing of medicinal decoction pieces according to claim 1, characterized in that, The error compensation module includes weight detection, air pressure compensation, temperature compensation, and parameter comparison.

4. The system for verifying the dispensing of medicinal decoction pieces according to claim 1, characterized in that, The analysis and comparison module includes an association unit and a verification unit. The association unit uses the K-nearest neighbor algorithm to calculate the Euclidean distance between the features of the medicinal slices to be matched and the features of each sample in the standard library. The calculation formula is as follows: in, Let j be the j-th dimension characteristic value of the medicinal slice to be tested. Let be the j-th eigenvalue of the i-th sample in the standard library, and n be a 128-dimensional eigenvector.

5. The system for verifying the dispensing of medicinal decoction pieces according to claim 4, characterized in that, The verification unit is used to inject the output potential medicinal herb types into a pre-trained convolutional neural network model for secondary verification to determine the final medicinal herb types.

6. The system for verifying the dispensing of medicinal decoction pieces according to claim 1, characterized in that, The specific content of the building unit also includes near-infrared spectral data mapping. By selecting a standard sample set of typical active ingredients, a partial least squares model is established to calculate the target component concentration vector. Then, the target component concentration vector calculation result is spatially registered with the point cloud model, a pseudo-color heat map is rendered in real time, and the chemical composition value of discrete measurement points is extended to the entire surface using the nearest neighbor interpolation algorithm.

Images