Intelligent traditional Chinese medicinal material character identification system based on computer vision

Abstract

The invention relates to the technical field of computer vision and optical measurement, and discloses a traditional Chinese medicinal material character intelligent identification system based on computer vision, which comprises an active optical projection device, an image acquisition device and a central control and processing device, the system projects a polarization modulation frequency conversion structure light field and collects multi-channel polarization data; the polarization resolving and decoupling module separates the reflected light field into mirror reflection and volume scattering components; the internal density analysis module analyzes a density index based on a multi-frequency modulation amplitude contrast attenuation relationship by using the volume scattering component; the surface microscopic character reconstruction module fuses the macroscopic depth resolved by the structured light and the microscopic normal field resolved by the polarization to generate microscopic surface depth distribution and extract roughness features; and the intelligent identification decision module synthesizes density and roughness indexes to construct a standardized fingerprint vector. According to the invention, synchronous detection and objective quantitative identification of internal compactness and surface microtopography of the traditional Chinese medicinal materials are realized.

Description

Technical Field

[0001] This invention relates to the fields of computer vision and optical measurement technology, specifically to a computer vision-based intelligent identification system for the characteristics of traditional Chinese medicinal materials. Background Technology

[0002] The identification of the properties of Chinese medicinal materials is a core means of evaluating their authenticity and quality. Traditional methods rely on professionals to make a comprehensive judgment on the shape, surface texture, and cross-sectional texture of the materials through sensory means. In existing evaluation systems, terms such as "powdery" and "oily" are often used to describe the internal structure of medicinal materials. However, these descriptions are mainly based on human subjective feelings and accumulated experience, lacking unified and objective physical quantitative standards. This results in poor repeatability of identification results and makes it difficult to establish a standardized digital quality evaluation system.

[0003] With the application of computer vision technology, automated identification methods based on image processing have gradually become widespread. However, existing conventional machine vision solutions only utilize the surface color or texture features of two-dimensional images for classification. Since conventional imaging techniques mainly receive directly reflected light from the surface of objects, they are unable to detect the scattering behavior of photons after entering the interior of the medium, and therefore cannot perceive the density of the internal tissue structure of Chinese medicinal materials. When faced with Chinese medicinal materials that are extremely similar in appearance but differ in internal texture (such as differences in tissue density due to growth years or processing methods), technical solutions relying solely on surface visual features are insufficient to achieve accurate differentiation, resulting in limited identification accuracy.

[0004] Furthermore, the microscopic morphology of the surface of Chinese medicinal materials (such as fine wrinkles and pore distribution) is also an important identification feature. Among existing three-dimensional morphology measurement techniques, traditional structured light projection technology, while capable of acquiring macroscopic geometric depth information of the object's surface, is limited by the spatial frequency limit of the projected fringes and the resolution of the optical system, making it difficult to capture high-frequency microscopic texture details. While polarization-based shape reconstruction technology has high sensitivity to changes in the microscopic normals of the object's surface, it lacks absolute physical depth reference information due to limitations in its physical principles, and is prone to low-frequency deformation when reconstructing complex surfaces. Therefore, a single optical measurement method cannot simultaneously achieve the accuracy of macroscopic geometry and the high fidelity of microscopic surface texture, limiting the extraction and analysis of microscopic roughness characteristics of the surface of Chinese medicinal materials. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent identification system for the properties of Chinese medicinal materials based on computer vision. It solves the problems that traditional identification of Chinese medicinal materials relies on subjective experience and lacks objective quantitative standards for internal texture characteristics; conventional machine vision technology, which only uses surface information, cannot effectively distinguish Chinese medicinal materials that look similar but have different internal densities; and single optical measurement methods cannot simultaneously take into account the accuracy of macroscopic geometric depth and the fidelity of microscopic surface texture, thus limiting the extraction of roughness features.

[0006] This invention provides a computer vision-based intelligent identification system for the properties of traditional Chinese medicinal materials. The system includes an active optical projection device, an image acquisition device, an environmental support device, and a central control and processing device. The active optical projection device is configured to project a polarization-modulated, frequency-modulated structured light field onto the surface of the medicinal material to be tested. The image acquisition device is configured to simultaneously acquire multi-channel polarization state data from the reflected light field on the surface of the medicinal material within the same exposure cycle. The central control and processing device establishes a communication connection with the active optical projection device and the image acquisition device via a data bus, and includes a light field control and synchronization module, a polarization calculation and decoupling module, an internal density analysis module, a surface microstructure reconstruction module, and an intelligent identification decision module.

[0007] In terms of light field control and data acquisition, the light field control and synchronization module generates digital stripe patterns and triggers synchronous acquisition. The intelligent identification system for Chinese medicinal materials adopts a multi-frequency phase-shifting strategy, storing a set of spatial frequency parameters, including a low-frequency reference frequency and a high-frequency detection frequency, through a frequency-variable stripe generation unit. The low-frequency reference frequency is mainly used to establish a macroscopic depth reference, while the high-frequency detection frequency is used to excite multiple scattering effects of photons within the medicinal material, providing a physical basis for subsequent internal texture analysis.

[0008] In terms of polarization data processing, the polarization resolution and decoupling module performs physical component separation of the optical field. First, the module performs channel separation and reconstruction on the acquired raw image data, generating radiance images with different polarization analysis directions. Then, it constructs a linear Stokes vector field through linear intensity combination operations. Based on this, the intelligent identification system for Chinese medicinal materials calculates the degree of linear polarization of the surface of the medicinal material to be tested. Using the degree of linear polarization as a weighting factor, it decouples the total radiance field at the pixel level into a specular reflection component that retains polarization characteristics and a volume scattering component that undergoes depolarization.

[0009] For the quantitative analysis of the internal texture (e.g., powdery or oily) of Chinese medicinal materials, the internal density analysis module utilizes volume scattering components for analysis. The module extracts the modulation amplitude spectrum by performing digital phase-shift demodulation on the image sequence composed of volume scattering components using a modulation amplitude extraction unit. Subsequently, the MTF characteristic analysis unit, based on photon transport theory, analyzes the contrast attenuation relationship between the modulation amplitude at the low-frequency reference frequency and the modulation amplitude at the high-frequency detection frequency. Since media with different densities have varying damping capabilities for scattering high-frequency light signals, this contrast attenuation directly reflects the internal density index of the Chinese medicinal materials. Furthermore, the intelligent identification system for Chinese medicinal material characteristics combines an effective area mask to statistically analyze this index, generating a global internal density score characterizing the average density level of the entire sample and an internal density consistency variance characterizing the dispersion of the internal structure distribution.

[0010] For the quantitative analysis of the surface microstructures (such as wrinkling and smoothness) of Chinese medicinal materials, the surface microstructure reconstruction module adopts a multi-dimensional feature fusion strategy. The macroscopic depth calculation unit uses volume scattering components to calculate the truncated phase and performs multi-frequency time phase expansion to obtain a continuous absolute phase. This phase is then converted into a macroscopic depth in physical space using a calibrated phase height mapping coefficient matrix. Simultaneously, the polarization normal fusion unit uses the degree of linear polarization to calculate microscopic normals and construct a polarization gradient vector field. The intelligent identification system for Chinese medicinal material properties fuses the low-frequency geometric information of the macroscopic depth with the high-frequency texture information of the polarization gradient by solving a global optimization problem containing gradient fidelity terms and depth constraints, thereby obtaining the microscopic surface depth distribution. Based on this, the roughness feature extraction unit separates the average surface and calculates the residual of the microscopic surface depth value relative to the average surface, generating a root mean square roughness index.

[0011] In the final identification and decision-making stage, the intelligent identification and decision-making module transforms the aforementioned physical characteristics into mathematical criteria. The multi-dimensional fingerprint vector construction unit normalizes the global internal density score, internal density consistency variance, and root mean square roughness index to generate a standardized multi-dimensional fingerprint vector. The classification and decision-making unit calculates the Euclidean distance feature distance between this vector and the preset target fingerprint center vector to achieve objective identification of the authenticity and quality of the Chinese medicinal materials to be tested.

[0012] This invention provides an intelligent identification system for the properties of traditional Chinese medicinal materials based on computer vision. It has the following beneficial effects: 1. This invention projects a polarization-modulated variable-frequency structured light field using an active optical projection device, and uses a polarization calculation and decoupling module to separate the reflected light field into a specular reflection component that characterizes surface information and a volume scattering component that characterizes internal information. This enables the simultaneous detection of the dual physical properties of internal density and external micromorphology of Chinese medicinal materials, solving the problem that traditional visual methods cannot distinguish between Chinese medicinal materials with similar appearance features but significant differences in internal tissue structure or density.

[0013] 2. This invention utilizes the MTF characteristic analysis unit in the internal density analysis module to analyze the modulation amplitude contrast attenuation relationship between the low-frequency reference frequency and the high-frequency detection frequency based on the photon transmission principle. This allows for the non-destructive quantification of the scattering damping ability of the internal tissue of Chinese medicinal materials to photons, providing objective digital evaluation indicators for traditional morphological terms such as powderiness or oiliness of Chinese medicinal materials.

[0014] 3. This invention solves a global optimization problem containing gradient fidelity terms and depth constraints by using a surface microstructure reconstruction module. It fuses the macroscopic depth obtained by structured light with the microscopic normal field obtained by polarization vision, supplementing the spatial resolution information of the microscopic region of the single structured light technology and the physical depth reference information missing by the single polarization technology. This allows for the construction of surface depth distribution data containing microscopic geometric features and the calculation of roughness features. Attached Figure Description

[0015] Figure 1 This is a block diagram illustrating the system interaction principle between the various modules of this invention. Figure 2 This is a data processing logic block diagram of the polarization calculation and decoupling module of the present invention; Figure 3 This is a schematic diagram illustrating the internal density analysis principle and modulation transfer function characteristics of this invention. Figure 4 This is a schematic diagram of the fusion process of the surface microstructure reconstruction module of the present invention; Figure 5 This is a schematic diagram of fingerprint vector construction and classification decision in the intelligent identification decision module of the present invention. Detailed Implementation

[0016] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] See attached document Figure 1This invention provides a computer vision-based intelligent identification system for the properties of traditional Chinese medicinal materials. At the physical hardware level, it mainly includes: an active optical projection device, an image acquisition device, an environmental support device, and a central control and processing device. These devices are interconnected via data transmission cables and synchronous triggering circuits, forming a closed-loop optical measurement hardware system.

[0018] The active optical projection device is configured to project a polarization-modulated, frequency-modulated structured light field onto the surface of the medicinal material to be tested. This active optical projection device includes a digital micromirror projection unit and a polarization modulation unit. The digital micromirror projection unit employs projection equipment based on digital light processing (DLP) technology or liquid crystal on silicon (LCoS) technology, and has programmable image projection capabilities. The optical axis of the digital micromirror projection unit points towards the sample area to be tested, and is configured to project sinusoidal fringe patterns with different spatial frequencies and different phase offsets according to received control commands.

[0019] A polarization modulation unit is disposed in the projection optical path of the digital micromirror projection unit. In this embodiment, the polarization modulation unit includes a linear polarization filter, which is mechanically fixed in front of the projection lens of the digital micromirror projection unit. The transmission axis of the linear polarization filter forms a first preset angle with the horizontal reference line of the system. After being modulated by the linear polarization filter, the unpolarized structured light emitted by the digital micromirror projection unit is converted into polarized structured light with a specific linear polarization state.

[0020] The image acquisition device is configured to simultaneously acquire multi-channel polarization state data of the reflected light field on the surface of the Chinese medicinal material under test within the same exposure cycle. The image acquisition device includes a focal plane polarization imaging unit and an optical imaging lens. The focal plane polarization imaging unit employs a complementary metal-oxide-semiconductor (CMOS) or charge-coupled device (CCD) image sensor with an integrated micro-polarizer array (MPA).

[0021] The micro-polarization array is directly applied to the photosensitive pixel surface of the image sensor. This micro-polarization array consists of multiple superpixel units arranged periodically in a 2×2 matrix. Each superpixel unit contains four adjacent micro-polarized pixels with polarization detection directions of 0°, 45°, 90°, and 135°, respectively. The image acquisition device is configured to output raw image data, in which adjacent pixels correspond to the four different polarization detection channels.

[0022] The optical axes of the image acquisition device and the active optical projection device are spatially convergent, intersecting at the plane where the medicinal material to be tested is placed. A second preset angle is formed between the optical axes of the image acquisition device and the active optical projection device. The second preset angle The reference angle for triangulation is configured to meet the parallax requirements of structured light 3D reconstruction.

[0023] The environmental support device is configured to provide the physical environment for optical measurements. The environmental support device includes a light-shielding enclosure and a stage. The light-shielding enclosure constructs a closed darkroom space, configured to block external ambient light from entering the measurement area. The stage is located inside the light-shielding enclosure, in the overlapping area of ​​the projection field of view of the active optical projection device and the acquisition field of view of the image acquisition device. The upper surface of the stage is covered with a visible light-absorbing material, configured to support the medicinal material to be measured and absorb light projected onto the background area to reduce background stray light reflection.

[0024] The central control and processing unit establishes bidirectional communication connections with both the active optical projection device and the image acquisition device. The central control and processing unit is configured to send stripe pattern data and projection commands to the active optical projection device, and to send a synchronization trigger signal to the image acquisition device. The central control and processing unit is also configured to receive raw image data output by the image acquisition device, and to store and process the data. The central control and processing unit includes one or more combinations of a general-purpose computer workstation, an embedded processor, or a field-programmable gate array (FPGA).

[0025] The central control and processing device in this embodiment of the invention runs system control and data processing software. This system control and data processing is divided into multiple functional modules at the logical level. Each module works together to complete tasks such as light field projection control, data acquisition synchronization, signal calculation and feature extraction. The system logical functional modules include: light field control and synchronization module, polarization calculation and decoupling module, internal density analysis module, surface microstructure reconstruction module and intelligent identification and decision module.

[0026] The light field control and synchronization module is configured to generate the digital stripe pattern required to drive the active optical projection device and to generate a synchronous trigger signal to control the exposure timing of the image acquisition device. The polarization calculation and decoupling module is configured to receive the raw image data, calculate the Stokes vector field, and separate the reflected light field into specular reflection and volume scattering components based on polarization characteristics. The internal density analysis module is configured to analyze the internal density index of the medicinal material using high-frequency structured light modulation information in the volume scattering component. The surface microstructure reconstruction module is configured to reconstruct the microscopic normal field and roughness characteristics of the medicinal material surface by combining the specular reflection component and polarization information. The intelligent identification decision module is configured to construct a multi-dimensional fingerprint vector by integrating the above physical characteristics and output the final identification result.

[0027] The light field control and synchronization module contains a frequency conversion stripe generation unit, which is configured to generate a series of digitized sinusoidal grating stripe image data based on a preset optical coding protocol, and transmit the sinusoidal grating stripe image data to an active optical projection device for projection.

[0028] The frequency conversion fringe generation unit uses digital phase-shifting fringe technology to generate the light field pattern. For each complete measurement task, the frequency conversion fringe generation unit is configured to construct a pattern containing... A set of stripe sequences with different spatial frequencies, among which It is an integer greater than or equal to 2. For the th element in the sequence... Spatial frequency of the stripes The frequency conversion stripe generation unit is configured to perform a phase shift step of 1000 steps. The phase shift strategy, and accordingly generate a number equal to A sequence of phase-shifted fringe patterns. This phase-shifted fringe pattern is located at pixel coordinates. The digital grayscale value below Defined by the following formula: ; in, Indicates the first The first group frequency Phase shift fringes at pixel coordinates The digital grayscale value at that location; This represents the background light intensity bias, which corresponds to the DC component of the sine wave, and is configured to keep the output grayscale value within the linear dynamic range of the projector. This represents the amplitude of the modulated light intensity, which determines the contrast of the projected stripes; Indicates the first The spatial frequency of a group of fringes, measured in periods per pixel or periods per field of view; Indicates the first The phase shift corresponding to the step phase shift stripe.

[0029] The frequency conversion stripe generation unit is configured to generate stripes based on the number of phase shift steps. Calculate discrete phase offset In this embodiment, the number of phase shift steps... Set to 3 or 4. When hour, The set of values ​​is ;when hour, The set of values ​​is .

[0030] The frequency conversion stripe generation unit stores a set of preset spatial frequency parameters. This set of spatial frequency parameters must contain at least a low-frequency reference frequency. and high frequency detection frequency Low-frequency reference frequency The period configuration is set to cover the entire field of view of the Chinese medicinal material being tested, i.e. A value greater than or equal to the lateral pixel width of the projected field of view, used to establish a phase-blur-free macroscopic depth reference. High-frequency detection frequency. Configured to a frequency higher than the low-frequency reference frequency , specifically, The value range is set between 1 / 10 and 1 / 2 of the projector's Nyquist frequency. This high-frequency detection frequency... This is used to excite the photon multiple scattering effect inside the Chinese medicinal material under test in subsequent image processing, so as to generate a measurable modulation transfer function attenuation signal.

[0031] The frequency conversion stripe generation unit is also configured to process the calculated digital grayscale values. The image is quantized and encoded into an 8-bit or 10-bit unsigned integer grayscale image matrix suitable for the input of the digital micromirror projection unit. The quantized image matrix is ​​then transmitted to the active optical projection device via a video signal interface (such as HDMI or DVI) or a high-speed data bus.

[0032] The light field control and synchronization module also includes a timing synchronization control unit, which is configured to precisely coordinate the pattern refresh action of the active optical projection device and the exposure action of the image acquisition device in the time dimension, so as to ensure that each frame of the acquired image corresponds to one and only one structured light stripe pattern in a stable display state.

[0033] The timing synchronization control unit is connected to the external trigger interface of the image acquisition device via physical signal lines. The timing synchronization control unit is configured to generate a hardware trigger pulse signal with a TTL (transistor to transistor logic) level standard. The image acquisition device is configured to immediately activate the electronic shutter of the image sensor for exposure upon detecting a valid edge of the hardware trigger pulse signal.

[0034] For including Group frequencies and each group of frequencies contains The complete measurement sequence of phase shifts is configured with a timing synchronization control unit to execute a strictly ordered projection, wait, and trigger control loop. For any given phase shift... The first group frequency To display the phase-shifting stripe pattern, the timing synchronization control unit first sends a refresh command to the active optical projection device, driving the digital micromirror element to flip and display the corresponding digital grayscale value. .

[0035] After sending the refresh command, the timing synchronization control unit is configured to perform a micromirror deflection stabilization delay of a preset duration. The micromirror deflection stabilization delay The value is set to a range of 500 microseconds to 5 milliseconds, with the specific value depending on the hardware response characteristics of the digital micromirror projection unit. The purpose of setting this delay is to filter out the transitional light field during the micromirror flipping process, preventing the image acquisition device from recording blurry or torn projection patterns.

[0036] Micromirror deflection stabilization delay At the end of the sequence, the timing synchronization control unit immediately sends a hardware trigger pulse signal. The image acquisition device responds to the hardware trigger pulse signal and remains operational for a duration of [duration missing]. The exposure window. The timing synchronization control unit is configured to ensure the exposure duration. The following time-domain constraint inequalities must be satisfied: ; in, This indicates the time point at which the hardware trigger pulse signal is emitted, which corresponds to the start time when the current stripe pattern enters a stable state; This indicates the exposure time of the image acquisition device, and this parameter is determined by the dynamic range of the reflectance of the surface of the Chinese medicinal material being tested. The digital grayscale value indicates that the active optical projection device has started switching to the next stripe pattern. (or (The time point).

[0037] By applying the above time-domain constraints, the timing synchronization control unit ensures the operation throughout the entire exposure time. Inside, the light field projected onto the surface of the medicinal herb under test remains a static and unique digital grayscale value. This achieves microsecond-level strict synchronization between light field projection and polarization acquisition, eliminating phase calculation errors caused by timing misalignment.

[0038] See attached document Figure 2 The polarization calculation and decoupling module contains a Stokes vector calculation unit, which is configured as the first link in the data processing pipeline. It receives the raw polarization image data output by the image acquisition device and converts the data into a Stokes vector field that characterizes the polarization state of the light field.

[0039] The Stokes vector computing unit first performs channel separation and demosaicing on each received frame of the original image. Because the image acquisition device employs a focal plane micro-polarization array structure, the output single-frame original image appears as a pixel-level interwoven mosaic pattern. The Stokes vector computing unit is configured based on a 2×2 superpixel arrangement rule of the micro-polarization array, and uses bilinear interpolation or bicubic interpolation algorithms to reconstruct the original image into four single-channel polarized radiance images with the same resolution.

[0040] The four single-channel polarized radiance images above correspond to the four polarization analysis directions of 0°, 45°, 90°, and 135°, respectively. Defined in pixel coordinates... At that location, the radiance values ​​of the four channels obtained after reconstruction are respectively , , as well as .in, This represents the light intensity component when the transmission axis is 0°. This represents the light intensity component when the transmission axis is 45°. This represents the light intensity component when the transmission axis is 90°. This represents the light intensity component when the light transmission axis is 135°.

[0041] Based on the acquired radiance values ​​from the four channels, the Stokes vector computation unit is configured to calculate the linear Stokes vector pixel-by-pixel. In this embodiment, since the circular polarization component contributes very little to the identification of the properties of Chinese medicinal materials and is not actively modulated, the system ignores the circular polarization component. Only the first three components are calculated. , , The calculation formula is defined as follows: ; ; ; in, Represents pixel coordinates The total radiance at that location includes the sum of the unpolarized light component and the polarized light component, and is physically equivalent to a grayscale image captured by a traditional unpolarized camera. This represents the intensity difference between the horizontally linearly polarized light component and the vertically linearly polarized light component. This represents the intensity difference between the linearly polarized light component at 45° and the linearly polarized light component at 135°.

[0042] The Stokes vector computation unit is configured to repeatedly perform the above calculation steps for each acquired image in the frequency conversion stripe sequence, output the corresponding Stokes vector field data stream, and display the calculation results. It is then transmitted to the subsequent optical transmission component separation unit.

[0043] The polarization calculation and decoupling module also includes an optical transmission component separation unit, which is configured to receive the output of the Stokes vector calculation unit. Based on the depolarization physical characteristics of light propagating at different media interfaces, the total radiance is decoupled into a specular reflection component characterizing surface physical properties and a volume scattering component characterizing internal material properties.

[0044] The optical transmission component separation unit first calculates the degree of linear polarization (DoLP) of the surface of the medicinal material under test based on the input Stokes vector field. (DoLP) The formula for quantifying the proportion of light intensity that retains a specific polarization state in reflected light is defined as follows: ; in, Represents pixel coordinates The degree of linear polarization at that location ranges from 0 to 1. This represents the total intensity of the linearly polarized light component; This indicates the total radiance at that location; This represents the intensity difference between the horizontally linearly polarized light component and the vertically linearly polarized light component. This represents the intensity difference between the linearly polarized light component at 45° and the linearly polarized light component at 135°.

[0045] The optical transmission component separation unit is configured to perform decoupling operations based on a polarization component model of the bidirectional reflection distribution function (BRDF). This polarization component model is based on the following physical mechanism: the projected polarized structured light mainly maintains its original polarization characteristics during a single specular reflection upon contact with the surface of the medicinal material; while photons that enter the interior of the medicinal material and undergo multiple refractions and scatterings exhibit non-polarization characteristics due to the decoherence of their polarization state caused by random scattering paths.

[0046] Based on the above physical mechanism, the optical transmission component separation unit uses the following formula to separate the specular reflection component. Volume scattering component : ; ; in, This represents the specular reflection component, which includes high-frequency reflection information of the gloss, oiliness, and micro-texture of the surface of Chinese medicinal materials. This represents the volume scattering component, which contains information about diffuse reflection of light after it has traveled through the interior of the medicinal material, reflecting the density of the internal structure and optical turbidity of the material. This indicates the total radiance at that location; Represents pixel coordinates The degree of linear polarization at that location.

[0047] The optical transmission component separation unit is configured to independently perform separation operations on each frame of data in the frequency conversion fringe sequence, generating independent volume scattering pattern sequences and specular reflection pattern sequences. The optical transmission component separation unit then separates the resulting... The data is transmitted to the internal density analysis module for analyzing internal structural features; simultaneously... The data is transmitted to the surface microstructure reconstruction module for analyzing surface texture features. Through this decoupling process, the system achieves a physical-level separation of the internal properties and external optical information of Chinese medicinal materials.

[0048] See attached document Figure 3 In this embodiment of the invention, the internal density analysis module includes a modulation amplitude extraction unit, which is configured to perform frequency domain demodulation operation on the volume scattering component image sequence output by the optical transmission component separation unit to obtain light field contrast distribution data at different spatial frequencies.

[0049] The modulation amplitude extraction unit first receives a sequence of volume scattering component images. Each image in this sequence corresponds to a spatial frequency generated by the optical field control and synchronization module. The next Phase-shifting fringes. Definition For a spatial frequency of And the phase offset is Under the projection conditions, the volume scattering component obtained by decoupling through the optical transmission component separation unit at the pixel coordinates The light intensity value at that location.

[0050] The modulation amplitude extraction unit is configured to apply a digital phase-shift demodulation algorithm to extract amplitude from data containing DC background components and AC modulation components. The amplitude of the AC component is extracted from the sequence. For each selected spatial frequency... The modulation amplitude extraction unit calculates the modulation amplitude distribution at this frequency according to the following formula. : in, Indicates the first Group space frequency The following is a distribution diagram of the modulation amplitude of the scattered light field inside the Chinese medicinal material to be tested, which characterizes the AC signal intensity of the photon density wave of this spatial frequency after transmission in the medium. This represents the total number of phase shift fringes (i.e., the number of phase shift steps) contained in a complete phase shift cycle; this parameter is a fixed positive integer. This represents the global term of the normalized coefficients, configured to eliminate the phase shift step factor. The difference in magnitude affects the final calculated amplitude. Indicates the first The phase shift corresponding to the phase shift fringes; Indicates the first At the group spatial frequency, corresponding to the first During phase-shift fringe projection, at pixel coordinates The intensity value of the volume scattering component light obtained by collecting and separating at the location; and They represent the first The sine and cosine values ​​of the phase-shifted fringes are used as weighting coefficients for orthogonal projection; This represents the weighted sum of the intensity value of the volume scattering component and the phase sine value. This overall term corresponds in a physical sense to the imaginary component of the demodulation vector in the orthogonal coordinate system. This represents the weighted sum of the intensity value of the volume scattering component and the phase cosine value. This overall term physically corresponds to the real component of the demodulation vector in the orthogonal coordinate system.

[0051] The above formula extracts the amplitude of the AC component of the optical field by calculating the arithmetic square root of the sum of the squares of the real and imaginary components. The modulation amplitude extraction unit is configured to process the set of spatial frequency parameters. Perform the above calculations for each frequency in the dataset to generate a set of modulation amplitude spectra. Each pixel value in this modulation amplitude spectrum set The contrast retention of structured light at corresponding frequencies during transmission within the Chinese medicinal materials was quantified. The modulation amplitude extraction unit transmits the calculated modulation amplitude spectrum set to the subsequent modulation transfer function (MTF) analysis unit.

[0052] The internal density analysis module also includes an MTF characteristic analysis unit, which is configured to receive the modulation amplitude spectrum set output by the modulation amplitude extraction unit and calculate the density index characterizing the internal tissue structure of Chinese medicinal materials based on the frequency response difference of the multi-frequency light field in the medium.

[0053] The MTF feature analysis unit first analyzes the received modulation amplitude spectrum set. Extract the low-frequency reference frequency Low-frequency modulation amplitude diagram and corresponding high-frequency detection frequency High-frequency modulation amplitude diagram .in, and The fixed frequency parameters are preset in the frequency conversion stripe generation unit.

[0054] The MTF characteristic analysis unit is configured to perform normalization operations to eliminate the influence of uneven light source intensity and local reflectivity differences on the surface of the medicinal materials on the measurement results. Since low-frequency fringes mainly reflect macroscopic reflectivity, while high-frequency fringes are highly sensitive to contrast attenuation caused by internal scattering, the MTF characteristic analysis unit uses low-frequency data as a benchmark to calculate the internal density index of the medicinal materials under test. The calculation process is based on the following formula: ; in, This indicates the pixel coordinates of the Chinese medicinal material to be tested. The internal density index is a physical quantity used to quantify the scattering damping ability of the internal tissue of Chinese medicinal materials to photon transmission. Its value is positively correlated with the fiber density, starch content and cell structure compactness inside the medicinal materials. This represents the density calibration coefficient, which is a preset positive real constant; This represents the natural logarithm operator, configured to linearize the exponential contrast decay relationship; Indicates the low-frequency reference frequency The calculated low-frequency modulation amplitude diagram is shown below; Indicates the frequency of high-frequency detection The high-frequency modulation amplitude diagram is calculated below; The numerical stability constant is a very small positive real number (e.g., 10^6). -6 The purpose of configuring this numerical stability constant is to prevent the denominator from being zero and to ensure the numerical stability of calculations in extremely low contrast regions (such as shadow areas). This represents the ratio of low-frequency amplitude to high-frequency amplitude, i.e., the contrast attenuation ratio. This ratio directly reflects the steepness of the decrease in modulation transfer function (MTF) as frequency increases. Within traditional Chinese medicine materials, the denser the tissue, the shorter the mean free path of photons, and the more pronounced the depolarization and blurring effects of the high-frequency light field, leading to… Compared to This decreases significantly, thus increasing the ratio.

[0055] The MTF characteristic analysis unit is configured to perform the above calculations on each pixel within the entire field of view, generating a complete two-dimensional internal density distribution map. This map can visually show whether structural variations such as hollowness, powderiness, or lignification exist within the medicinal material. The MTF characteristic analysis unit will generate the internal density index. The matrix is ​​transmitted to the subsequent intelligent identification decision module as one of the core dimensional data for constructing the identification fingerprint vector.

[0056] The internal density analysis module also includes an internal density index (ID) generation unit. This internal density index generation unit is configured to perform statistical aggregation and background removal processing on the two-dimensional internal density index distribution map output by the MTF characteristic analysis unit in order to generate a quantitative scalar index characterizing the overall internal structural characteristics of the Chinese medicinal material to be tested.

[0057] The internal density index generation unit is first configured to generate the density index based on the low-frequency modulation amplitude map. Constructing an effective region mask To eliminate the interference of background noise and non-measurement areas at the edge of the field of view on the statistical results, the internal density index generation unit utilizes a preset background threshold. Binarize and segment the field of view. Effective region mask. The definition logic is: when pixel coordinates place Greater than the background threshold hour, A value of 1 indicates that the pixel belongs to the sample area of ​​Chinese medicinal materials; otherwise... A value of 0 indicates that the pixel belongs to the background area.

[0058] Based on the constructed effective region mask With the input internal density index The internal density index generation unit is configured to calculate the global internal density score at the sample level. The calculation process is based on the following formula: ; The global internal density score of the tested Chinese medicinal material sample is a single real scalar that represents the average statistical characteristics of the internal tissue density of the entire Chinese medicinal material sample. These represent the total pixel resolution of the image sensor in the horizontal and vertical directions, respectively. This represents the pixel coordinates input from the MTF feature parsing unit. The internal density index at the location; This indicates the effective area mask at pixel coordinates. The binary state (0 or 1) at the location; This represents the total number of pixels with a mask value of 1 in the entire field of view, which is the total effective pixel area occupied by the Chinese medicinal material to be tested in the image. This term is used as the denominator to perform spatial averaging operations. This represents the cumulative sum of internal density indices within the valid sample region, multiplied by... All background areas The value is set to zero, thus excluding it from the summation operation.

[0059] After calculating the global mean, the internal density index generation unit is also configured to calculate the consistency variance of the internal density of the sample. This method is used to assess the uniformity of the internal structure distribution of Chinese medicinal materials and to detect the presence of defects such as local hollowness or insect infestation. The calculation is based on the following formula: ; in, This represents the global internal density score of the tested Chinese medicinal material sample, which indicates the average density level of the entire sample. The uniformity variance represents the internal packing density of a sample and is used to quantify the dispersion of the internal structure distribution. This represents the sum of squared density deviations only within the valid sample area.

[0060] in, As a spatial gating operator, it ensures that only pixels located within the actual area of ​​the Chinese medicinal material participate in the fluctuation statistics, while forcing the bias contribution of the background area to zero; this overall term numerically accumulates the total fluctuation energy of all positions within the sample relative to the average compactness, reflecting the overall non-uniformity of the internal tissue of the Chinese medicinal material.

[0061] The internal density index generation unit calculates the global internal density score. and consistency variance The packaged output is transmitted as the final internal feature descriptor to the subsequent intelligent identification and decision-making module.

[0062] See attached document Figure 4 The surface microstructure reconstruction module in this embodiment of the invention includes a macroscopic depth calculation unit. The macroscopic depth calculation unit is configured as a basic step in the three-dimensional reconstruction process, receiving the volume scattering component image sequence output by the optical transmission component separation unit, and reconstructing the macroscopic three-dimensional topological structure of the surface of the Chinese medicinal material to be tested using the multi-frequency phase-shifting interference principle.

[0063] The macroscopic depth calculation unit is first configured for each preset spatial frequency. The corresponding wrap-around phase map is calculated. To avoid phase errors caused by specular highlight regions, the macroscopic depth calculation unit preferentially selects the volume scattering component intensity value after removing the specular reflection component. As input. The macroscopic depth solution unit is based on a four-step phase-shift algorithm (when... (Time) or general The phase-shift least squares algorithm calculates pixel coordinates using the following formula. Cutoff phase value at : ; in, Indicates spatial frequency The truncation phase value below, its numerical range is limited to Within the range; This represents a two-parameter arctangent function, configured to determine the quadrant of the phase based on the signs of the sum of the sine and cosine components. Indicates the number of phase shift steps; Indicates the first The phase shift corresponding to the phase shift fringes; This represents the intensity value of the corresponding volume scattering component; This represents the weighted sum of the intensity value of the volume scattering component and the phase sine value; It represents the weighted sum of the intensity value of the volume scattering component and the phase cosine value.

[0064] The macroscopic depth calculation unit is then configured to perform multi-frequency time-phase expansion operations to eliminate discontinuities caused by phase truncation, using the monotonicity of the low-frequency phase as a guide to move the high-frequency detection frequency... Corresponding truncation phase Unfolding into continuous absolute phases When using a dual-frequency hierarchical expansion strategy, the calculation of the absolute phase follows the recursive formula: ; in, Represents pixel coordinates The absolute phase at that point; Indicates phase cutoff; This represents the low-frequency absolute phase used as a reference (when the lowest frequency period covers the entire field of view). ); and These represent the high-frequency detection frequency and the low-frequency reference frequency, respectively. This represents the rounding operator, configured to correct the integer part of the period series required to determine the high-frequency phase.

[0065] Based on the acquired absolute phase The macroscopic depth calculation unit is configured to convert phase data into physical space height coordinates using a pre-calibrated phase-depth mapping relationship. This conversion process is based on the following linear phase height model (for telecentric optical path systems) or a general triangulation model: ; in, The height coordinates of the surface of the Chinese medicinal material to be tested relative to the reference plane constitute the low-frequency skeleton of the three-dimensional morphology of the Chinese medicinal material. This represents the phase-to-phase-height mapping coefficient matrix obtained from system calibration, used to characterize the amount of physical height change corresponding to a unit phase change; This represents the reference absolute phase field that has been pre-acquired and stored on the zero-height reference plane.

[0066] The macroscopic deep solution unit will calculate the results The data is transmitted to the subsequent micro-texture fusion unit as a boundary constraint for the surface normal integral.

[0067] The surface microstructure reconstruction module also includes a polarization normal fusion unit. This polarization normal fusion unit is configured to fuse the low-frequency macroscopic depth data output by the macroscopic depth calculation unit with the high-frequency normal data derived from the polarization optical properties, so as to reconstruct the three-dimensional morphology of the Chinese medicinal material surface that has both macroscopic geometric accuracy and microscopic texture details.

[0068] The polarization normal fusion unit is first configured to be input-based. and and height coordinates The deambiguity-processed microscopic surface gradient field is calculated. The polarization normal fusion element first utilizes height coordinates... Calculate the reference normal vector Then, the azimuth angle of the micro normal is calculated according to the polarization azimuth angle formula. And using the reference normal to determine the polarization azimuth angle Periodic ambiguity is corrected.

[0069] The polarization normal fusion unit is configured as a lookup table or analytical model based on Fresnel's law of reflection, and the degree of linear polarization output by the optical transmission component separation unit is... Solving the zenith angle of the micronormal Based on the corrected azimuth angle The zenith angle obtained by the solution Polarization normal fusion units construct a polarization gradient vector field that characterizes the orientation of surface micro-texture. The formula for calculating the gradient components is as follows: ; ; in, This represents the local slope of a microscopic surface in the horizontal direction. This represents the local slope of a microscopic surface in the vertical direction; Represents pixel coordinates The angle between the surface normal at a given location and the direction of the line of sight (i.e., the zenith angle); This represents the angle (i.e., azimuth) between the projection of the surface normal onto the imaging plane and the horizontal axis. The tangent value represents the zenith angle and is used to quantify the steepness of the surface tilt. and These represent the cosine and sine values ​​of the azimuth angle, respectively, used to decompose the kurtosis projection into an orthogonal coordinate system. This polarization gradient vector field... It contains high-frequency detailed information about the pores, wrinkles and textures on the surface of Chinese medicinal materials, but does not have an absolute depth benchmark.

[0070] To impose absolute depth constraints on high-frequency gradient information, the polarization normal fusion unit is configured to construct and solve a global energy functional minimization problem, aiming to find the optimal microscopic surface depth distribution. So that it maintains its height coordinates While maintaining consistency, its surface gradient should fit the polarization gradient vector field as closely as possible. The total energy loss function of this optimization problem. The definition is as follows: in, The depth distribution of the microscopic surface to be solved is represented and is a decision variable in the optimization process; This represents the total energy loss function, whose minimum state corresponds to the optimal surface reconstruction result; Indicates the surface to be solved in pixel coordinates The gradient operator at a given point is typically represented in discrete difference form; This represents the polarization gradient vector field, serving as a guiding field for high-frequency shapes; The height coordinates input by the macroscopic depth calculation unit are used as the geometric constraint surface for the low-frequency shape. This represents the regularization weight coefficient, which is a preset positive real constant. This regularization weight coefficient is configured to adjust the balance between the gradient fidelity term and the depth constraint term. The larger the value, the closer the reconstruction result is to the macroscopic depth; The smaller the value, the more the reconstruction result highlights the details of the polarization texture; This represents the gradient fidelity term, used to force the reconstructed local geometric changes of the surface to conform to the physical properties of polarized light measurement; This represents a depth constraint term used to prevent low-frequency drift during integration and ensure the accuracy of the absolute spatial position of the reconstructed surface.

[0071] The polarization normal fusion unit is configured to numerically solve for the minimum value of the above objective function using a Poisson equation solver or the conjugate gradient method, thereby obtaining the final microscopic surface depth distribution. The polarization normal fusion unit will reconstruct the The data is then transmitted to the subsequent intelligent identification and decision-making module for the extraction of microscopic morphological features such as surface roughness and texture directionality.

[0072] The surface microstructure reconstruction module also includes a roughness feature extraction unit, which is configured to perform statistical analysis on the high-precision surface three-dimensional model output by the polarization normal fusion unit to quantify the microstructure undulation characteristics of the surface of the Chinese medicinal material to be tested.

[0073] The roughness feature extraction unit is first configured to receive the depth distribution of the microscopic surface. To accurately separate surface texture from macroscopic surface shape, the roughness feature extraction unit is configured to apply a two-dimensional Gaussian smoothing filter or a polynomial surface fitting algorithm, from... Extract the average surface as the reference. This average surface represents the low-frequency geometric profile of the surface of Chinese medicinal materials.

[0074] The roughness feature extraction unit is then configured to calculate the residual field of the surface height, and combined with a pre-defined effective region mask. The root mean square roughness index, which characterizes the surface microstructure density, is calculated based on the root mean square statistical principle. The calculation process is based on the following formula: ; in, The root mean square coarseness index (RMSE) represents the surface coarseness of the Chinese medicinal material being tested. It is a scalar quantity used to quantify the average dispersion of the micro-texture on the surface of the Chinese medicinal material in the vertical direction. This index is closely related to the degree of epidermal keratinization, water loss state, and processing technology of the Chinese medicinal material. These represent the total pixel resolution of the image sensor in the horizontal and vertical directions, respectively. This represents the effective area mask, when the pixel coordinates are... When located within the effective surface area of ​​the medicinal herb, the value is 1; when the pixel coordinates This value is 0 when the mask is located in the background or edge invalid region. This mask is configured to eliminate background noise interference with roughness calculations. This represents a high-precision surface depth value, including macroscopic shape and microscopic texture, input from the polarization normal fusion unit. Represents the average surface used as a reference; This indicates a two-dimensional traversal and accumulation operation on all pixels within the entire field of view; This represents the surface height residual, which eliminates the influence of macroscopic shape and retains only the microscopic texture undulation height information of the surface; This represents the total number of pixels within the effective region, i.e., the actual projected area (in pixels) involved in the roughness calculation. This term serves as the denominator for performing normalized averaging.

[0075] The above formula, by taking the square root of the variance of the height deviation, yields a value with the dimension of length. The roughness feature extraction unit will calculate the value. The indicators are packaged as key fingerprint features describing the microscopic properties of the surface of Chinese medicinal materials and transmitted to the subsequent intelligent identification and decision-making module.

[0076] See attached document Figure 5 The intelligent identification decision module in this embodiment of the invention includes a multi-dimensional fingerprint vector construction unit. The multi-dimensional fingerprint vector construction unit is configured to receive independent feature indicators generated by each preceding processing module and map them to a unified feature space to generate a standardized multi-dimensional fingerprint vector for subsequent classification and comparison.

[0077] The multidimensional fingerprint vector construction unit is first configured to aggregate the global internal density scores output by the internal density analysis module. Consistency variance of internal packing And the root mean square roughness index output by the surface microstructure reconstruction module. This multidimensional fingerprint vector construction unit combines these physical quantities in a preset order to construct the original feature vector to be processed. To eliminate the impact of differences in physical dimensions and numerical magnitudes on the identification algorithm, the multidimensional fingerprint vector construction unit is configured to construct the original feature vector. Each feature component in the fingerprint is normalized and weighted, and the final multidimensional fingerprint vector is calculated according to the following formula. The first in each feature component : ; in, Represents the first digit in the standardized multidimensional fingerprint vector. Each feature component (scalar value) has a range of values ​​constrained to within a certain range. Within the range; The index number representing the feature component, with a value ranging from 1 to... ,in The total dimension of the fingerprint vector; Represents the original feature vector The first in For example, the correspondence between the measured values ​​of each feature component is set as follows: Corresponding to , Corresponding to , Corresponding to ; Indicates corresponding to the first The identification weight coefficient of each feature component is a preset positive real number used to quantify the importance of the feature in distinguishing genuine and counterfeit Chinese medicinal materials. Indicates corresponding to the first A standard sample database distribution set of each feature component; This indicates that in the standard sample database, the first... The statistical minimum or preset lower bound of each characteristic component; This indicates that in the standard sample database, the first... The statistical maximum value or preset upper bound of each feature component.

[0078] The multidimensional fingerprint vector building unit is configured to compute all components. Arranged in index order, they are combined to form the final multidimensional fingerprint vector. This multidimensional fingerprint vector In mathematical form, it is represented as a A column vector of dimension, i.e. . As a digital identifier for the Chinese medicinal material to be tested, it is transmitted to the subsequent similarity matching unit for comparison and analysis with the benchmark vector in the standard Chinese medicinal material atlas library.

[0079] The intelligent identification decision module also includes a classification decision unit. This classification decision unit is configured to perform pattern recognition and similarity calculation based on the standardized multidimensional fingerprint vector output by the multidimensional fingerprint vector construction unit, so as to output the authenticity identification result or quality grade determination result of the Chinese medicinal material to be tested.

[0080] The classification decision unit is first configured to access a pre-set standard Chinese medicinal herb fingerprint database. This database stores verified target fingerprint center vectors corresponding to genuine Chinese medicinal herbs. The classification and decision unit will use the multidimensional fingerprint vector of the current test sample. With the target fingerprint center vector The two are compared, and the Euclidean distance error in the feature space is calculated. The calculation process is based on the following formula: in, This represents the Euclidean distance error between the fingerprint of the test sample and the standard genuine fingerprint. This value is a non-negative real number. The smaller the value, the closer the test sample is to the standard genuine fingerprint in terms of physical characteristics. A multidimensional fingerprint vector representing the sample to be tested. The first in Each feature component; This represents the target fingerprint center vector retrieved from the standard Chinese medicinal material fingerprint database. The first in The mean of each feature component, which is the expected value obtained through statistical analysis of a large number of genuine samples; This indicates that the squared deviations of all feature dimensions are summed.

[0081] The classification decision unit is then configured to base its decision on the calculated Euclidean distance error. Quantify the confidence probability that the Chinese medicinal materials under test are genuine products. To map unbounded distance values ​​to percentage-based probability values, the classification decision unit employs the following nonlinear mapping formula: ; in, This represents the confidence level probability that the tested Chinese medicinal material is genuine. The decision sensitivity coefficient is a preset positive real constant that determines the rate at which confidence decreases with increasing distance. The larger the value is set, the lower the system's tolerance for feature differences, and the stricter the judgment.

[0082] The classification decision unit is configured to calculate the confidence probability. Compared with the preset identification threshold Comparison. When Greater than or equal to At that time, the classification and judgment unit generates a judgment signal of "genuine" or "qualified"; when Less than At that time, the classification and judgment unit generates a judgment signal of "counterfeit" or "unqualified".

[0083] Finally, the classification and decision unit outputs an identification report containing the judgment result, confidence level value, and feature distance data to the user interface or external data interface, completing the automated comprehensive identification process of the internal density and surface microstructure of Chinese medicinal materials.

Claims

1. A computer vision-based intelligent identification system for the characteristics of traditional Chinese medicinal materials, characterized in that, The intelligent identification system for the properties of Chinese medicinal materials includes an active optical projection device, an image acquisition device, an environmental support device, and a central control and processing device. The active optical projection device is used to project a polarization-modulated frequency-converted structured light field onto the surface of the Chinese medicinal material to be tested. The image acquisition device is used to simultaneously acquire multi-channel polarization state data in the reflected light field on the surface of the Chinese medicinal material to be tested within the same exposure cycle. The environmental support device is used to provide an optical measurement environment; The central control and processing device establishes a communication connection with the active optical projection device and the image acquisition device; The central control and processing device includes an optical field control and synchronization module, a polarization calculation and decoupling module, an internal density analysis module, a surface microstructure reconstruction module, and an intelligent identification and decision-making module. The light field control and synchronization module is used to generate digital stripe patterns to drive the active optical projection device and generate a synchronization trigger signal. The polarization calculation and decoupling module is used to receive the original image data, calculate the Stokes vector field through channel separation and reconstruction, and separate the reflected light field into specular reflection component and volume scattering component. The internal density analysis module is used to analyze the internal density index of Chinese medicinal materials using the volume scattering component. The surface microstructure reconstruction module is used to reconstruct the microscopic normal field and roughness characteristics of the surface of Chinese medicinal materials by combining the specular reflection component and polarization information. The intelligent identification decision module is used to construct a multi-dimensional fingerprint vector by integrating the internal density index and roughness features and output the identification result.

2. The intelligent identification system for the properties of traditional Chinese medicinal materials based on computer vision according to claim 1, characterized in that, The optical field control and synchronization module includes a frequency conversion stripe generation unit; The frequency conversion stripe generation unit stores a set of spatial frequency parameters including a low-frequency reference frequency and a high-frequency detection frequency, and performs a multi-step phase shift strategy for each frequency in the set of spatial frequency parameters to generate a phase shift stripe pattern sequence. The low-frequency reference frequency is used to establish a macroscopic depth reference, and the high-frequency detection frequency is used to excite the internal photon multiple scattering effect.

3. The intelligent identification system for the properties of traditional Chinese medicinal materials based on computer vision according to claim 2, characterized in that, The polarization calculation and decoupling module includes a Stokes vector calculation unit and an optical transmission component separation unit; The Stokes vector calculation unit is used to reconstruct the original image data into radiance images with different polarization directions and calculate the linear Stokes vector through linear combination of light intensity. The optical transmission component separation unit is used to calculate the degree of linear polarization of the surface of the Chinese medicinal material to be tested based on the linear Stokes vector, and to decouple the total radiance into the specular reflection component and the volume scattering component using the degree of linear polarization.

4. The intelligent identification system for the properties of traditional Chinese medicinal materials based on computer vision according to claim 3, characterized in that, The internal density analysis module includes a modulation amplitude extraction unit, which is used to receive an image sequence composed of the volume scattering components and perform digital phase shift demodulation operation for each spatial frequency to extract a modulation amplitude spectrum determined by the weighted summation result of the volume scattering component light intensity value and the phase shift sine and cosine values.

5. The intelligent identification system for the properties of traditional Chinese medicinal materials based on computer vision according to claim 4, characterized in that, The internal density analysis module includes an MTF characteristic analysis unit, which is used to receive the modulation amplitude spectrum and calculate the internal density index, which quantifies the scattering damping ability of the internal tissue of Chinese medicinal materials to photon transmission, based on the contrast attenuation relationship between the modulation amplitude at the low-frequency reference frequency and the modulation amplitude at the high-frequency detection frequency.

6. The intelligent identification system for the properties of traditional Chinese medicinal materials based on computer vision according to claim 5, characterized in that, The internal density analysis module includes an internal density index generation unit, which is used to statistically analyze the internal density index in conjunction with an effective region mask, and generate a global internal density score that characterizes the average density level of the entire sample and an internal density consistency variance that characterizes the dispersion of the internal structure distribution.

7. The intelligent identification system for the properties of traditional Chinese medicinal materials based on computer vision according to claim 3, characterized in that, The surface microstructure reconstruction module includes a macroscopic depth calculation unit. The macroscopic depth calculation unit is used to calculate the truncated phase at different spatial frequencies using the volume scattering components, perform multi-frequency time phase expansion operations to obtain continuous absolute phase, and use a calibrated phase-phase-height mapping coefficient matrix to convert the absolute phase into the macroscopic depth of physical space.

8. The intelligent identification system for the properties of traditional Chinese medicinal materials based on computer vision according to claim 7, characterized in that, The surface microstructure reconstruction module includes a polarization normal fusion unit; The polarization normal fusion unit is used to calculate the microscopic normal using the linear polarization degree and construct the polarization gradient vector field. By solving a global optimization problem containing gradient fidelity terms and depth constraint terms, the macroscopic depth and the polarization gradient vector field are fused to obtain the microscopic surface depth distribution. The gradient fidelity term is used to make the reconstructed surface approximate the polarization gradient vector field, and the depth constraint term is used to make the reconstructed surface approximate the macroscopic depth.

9. The intelligent identification system for the properties of traditional Chinese medicinal materials based on computer vision according to claim 8, characterized in that, The surface microstructure reconstruction module includes a roughness feature extraction unit, which is used to separate the average surface from the microscopic surface depth distribution and generate a root mean square roughness index based on the root mean square value of the residual between the microscopic high-precision surface depth value and the average surface height value within the effective area.

10. The intelligent identification system for the properties of traditional Chinese medicinal materials based on computer vision according to claim 9, characterized in that, The intelligent identification decision module includes a multi-dimensional fingerprint vector construction unit and a classification decision unit; The multidimensional fingerprint vector construction unit is used to perform normalization processing on the global internal density score, the internal density consistency variance and the root mean square roughness index to generate a standardized multidimensional fingerprint vector. The classification and decision unit is used to calculate the feature distance between the standardized multidimensional fingerprint vector and the preset target fingerprint center vector using Euclidean distance metric, and output the authenticity identification result of the Chinese medicinal material to be tested based on the feature distance.

Images