A recognition device and method for soybean milk soybean particle screening equipment

Abstract

This application provides an identification device and method for soybean grain screening equipment in soybean milk production. The device scans and images the soybean surface. A polarization beam splitter in the camera's optical path separates the reflected light from the soybean surface into parallel polarized light and orthogonally polarized light with mutually perpendicular polarization directions, acquiring parallel and orthogonal polarized images. The parallel and orthogonal polarized images are adaptively fused to generate a polarization-fused image that suppresses surface reflection interference and enhances structural information. Texture periodic features characterizing the regularity of the soybean skin texture and distortion energy features characterizing the degree of surface undulation are extracted from the polarization-fused image. The texture periodic features and distortion energy features are input into a pre-constructed feature discrimination model for defect identification. Based on the defect identification results, it is determined whether defects exist on the soybean surface. Using the scheme of this application, polarization-fused imaging can suppress surface reflection interference, thereby improving the accuracy of soybean surface defect detection.

Description

Technical Field

[0001] This application relates to the field of intelligent identification technology, and more specifically, to an identification device and method for soybean granule screening equipment for soybean milk. Background Technology

[0002] With the continuous improvement of automation in the food industry, machine vision technology is increasingly widely used in the sorting of agricultural products. Soybeans, as the core raw material for soybean product processing, have surface defects that directly affect the quality and safety of the final product. Traditional manual sorting methods can no longer meet the dual requirements of efficiency and precision for large-scale production. In recent years, visual recognition devices based on intelligent line scan cameras have been gradually applied to soybean screening equipment. Through high-speed imaging, online detection of surface defects is achieved, which has powerfully promoted the intelligent development of soybean sorting technology.

[0003] However, existing visual recognition devices in soybean milk screening equipment mostly use ordinary line-scan cameras with uniform light sources for imaging and detection. In practical applications, the surface of fresh soybeans is usually covered with a natural waxy layer. This waxy layer easily produces strong and irregular specular reflections under light, resulting in large areas of highlight in the acquired soybean surface images. These highlight areas completely obscure the inherent texture and subtle defects of the soybean surface, appearing as grayscale saturation and local overexposure in the image. This causes defect recognition algorithms based on grayscale gradients or texture analysis to fail in these areas, leading to the missed detection of many critical defects such as cracks and dents. Even more challenging is that the intensity and position of specular reflection fluctuate randomly with changes in the soybean conveying posture. Traditional image preprocessing methods such as thresholding or filtering denoising are insufficient to fundamentally eliminate this dynamic interference, severely limiting the accuracy and stability of the detection system. Therefore, how to effectively suppress the reflective interference from the soybean surface and achieve high-precision surface defect detection has become a major challenge in the intelligent upgrading of current soybean milk screening equipment. Summary of the Invention

[0004] This application provides an identification device and method for soybean screening equipment for soybean milk, which can realize polarization fusion imaging to suppress surface reflection interference, thereby improving the accuracy of soybean surface defect detection.

[0005] In a first aspect, this application provides a method for identifying soybean particles in a soybean milk screening device, comprising: The soybeans to be tested are dispersed into a sequence of individual beans by a single-layer vibrating feeding mechanism and transported to the testing station. The surface of the moving soybeans is scanned and imaged using a laser light source arranged coaxially with an intelligent line scan camera. A polarization beam splitter is set in the camera's receiving optical path to separate the light reflected from the bean surface into parallel polarized light and orthogonal polarized light with mutually perpendicular polarization directions, and to acquire parallel polarization images and orthogonal polarization images. Adaptive fusion of the parallel polarization image and the orthogonal polarization image is performed to generate a polarization fused image that suppresses surface reflection interference and enhances structural information. Frequency domain features are extracted from the polarization fusion image to obtain texture periodic features that characterize the regularity of the bean skin texture and distortion energy features that characterize the degree of surface undulation of the bean. The texture periodic features and the distortion energy features are input into a pre-constructed feature discrimination model for defect identification, and then the presence of defects on the surface of the bean is determined based on the defect identification results.

[0006] Preferably, the scanning and imaging of the surface of a moving bean using a laser light source arranged coaxially with the intelligent line scan camera specifically includes: The laser source is activated to emit a linear beam, which is then guided by a coaxial optical assembly to a position that coincides with the optical axis of the intelligent line scan camera and projected onto the material passage area of ​​the inspection station. The line frequency parameters of the intelligent line scan camera are set according to the conveying speed of the single-layer vibrating feeder to keep the image acquisition frequency matched with the movement speed of the beans. As the beans pass through the inspection station, the intelligent line scan camera continuously collects the reflected light signals from the surface of the beans, line by line, generating a continuous two-dimensional image data stream.

[0007] Preferably, a polarization beam splitter is set in the camera's receiving optical path to separate the light reflected from the bean surface into parallel polarized light and orthogonal polarized light with mutually perpendicular polarization directions, and the acquisition of parallel polarization images and orthogonal polarization images specifically includes: A polarizing beam splitter is set at the front end of the lens of an intelligent line scan camera, and the incident and reflected light is separated into a first optical path and a second optical path by the polarizing beam splitter. A first polarizer is set in the first optical path so that its polarization direction is consistent with the output polarization direction of the laser source, and the transmitted light signal is projected onto the first image sensor to obtain a parallel polarized image; A second polarizer is set in the second optical path so that its polarization direction is perpendicular to the output polarization direction of the laser source, and the transmitted light signal is projected onto the second image sensor to obtain an orthogonally polarized image.

[0008] Preferably, adaptively fusing the parallel polarization image and the orthogonal polarization image to generate a polarization-fused image that suppresses surface reflection interference and enhances structural information specifically includes: For each pixel, the specular reflection index is calculated based on the gray value of the pixel in the parallel polarization image and the gray value in the orthogonal polarization image, which characterizes the degree of specular reflection affecting the pixel. Based on the orthogonal polarization image, the gray-level variance of each pixel within its preset neighborhood window is calculated as the texture complexity index of the local region where the pixel is located. For each pixel, based on the comparison between the specular reflection index of all pixels in the neighborhood window of that pixel and a preset threshold, it is determined whether the pixel is located in a region affected by strong specular reflection. Based on the judgment result and the texture complexity index, the weight coefficient of the pixel during fusion is determined. For pixels that are determined to be located in areas affected by strong specular reflection, their weight coefficient is set to a first preset value so that the gray values ​​of the orthogonal polarization image are fully used during fusion. For pixels that are not determined to be located in areas affected by strong specular reflection, the weight coefficient is dynamically allocated according to their texture complexity index so that the contribution ratio of the parallel polarization image is greater in areas with higher texture complexity. Based on the determined weighting coefficients, the gray values ​​of the pixel in the parallel polarization image and the gray values ​​in the orthogonal polarization image are weighted and averaged and fused, and the fusion result is used as the pixel value at the corresponding position in the polarization fused image. By iterating through all pixels, a polarization-fused image is generated that suppresses surface reflection interference and enhances structural information.

[0009] Preferably, determining whether the pixel is located in a region affected by strong specular reflection specifically includes: counting the proportion of pixels in the neighborhood window where the pixel's specular reflection feature value is greater than a preset threshold to the total number of pixels in the window; if the proportion exceeds the preset proportion threshold, then the pixel is determined to be located in a region affected by strong specular reflection.

[0010] Preferably, inputting the texture periodicity features and the distortion energy features into a pre-constructed feature discrimination model for defect identification specifically includes: Multiple standard bean samples with no surface defects were pre-acquired, and their corresponding texture periodic features and distortion energy features were extracted to construct a standard feature dataset. Based on the standard feature dataset, the feature distribution center and distribution boundary range are determined, and a feature discrimination model for distinguishing normal and abnormal regions is established. The texture periodic features and distortion energy features extracted from the bean grains to be tested are input into the feature discrimination model to obtain the corresponding defect identification results.

[0011] Preferably, the single-layer vibrating feeding mechanism includes a vibrating conveying trough with a single-layer conveying surface and a vibration driving device connected thereto. The vibration driving device is used to disperse and arrange the beans along a predetermined direction and convey them to the detection station in a single-bean interval state by controlling the vibration frequency and amplitude.

[0012] Secondly, this application provides an identification device for a soybean granule sieving equipment for soybean milk, comprising: The scanning module is used to disperse the soybeans to be tested into a sequence of individual beans through a single-layer vibrating feeding mechanism and transport them to the detection station. It uses a laser light source arranged coaxially with the intelligent line scan camera to scan and image the surface of the moving soybeans. The feature processing module is used to set up a polarization beam splitter in the camera receiving optical path to separate the light reflected from the surface of the bean into parallel polarized light and orthogonal polarized light with polarization directions perpendicular to each other, and to acquire parallel polarization images and orthogonal polarization images. The feature processing module is also used to adaptively fuse the parallel polarization image and the orthogonal polarization image to generate a polarization fused image that suppresses surface reflection interference and enhances structural information. The feature processing module is also used to extract frequency domain features from the polarization fusion image to obtain texture periodic features that characterize the regularity of the bean skin texture and distortion energy features that characterize the degree of surface undulation of the bean. The identification module is used to input the texture periodic features and the distortion energy features into a pre-constructed feature discrimination model for defect identification, and then determine whether there are defects on the surface of the bean based on the defect identification results.

[0013] Thirdly, this application provides a computer device, the computer device including a memory and a processor, the memory storing code, the processor being configured to acquire the code and execute the above-described identification method for a soybean granule sieving device for soy milk.

[0014] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described identification method for a soybean granule sieving device for soybean milk.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: In this embodiment, polarization fusion imaging can suppress reflective interference from the bean surface. First, a single-layer vibrating feeding mechanism disperses the bean grains to be tested into a sequence of individual grains and transports them to the detection station. A laser light source coaxially arranged with an intelligent line scan camera scans and images the surface of the moving bean grains, enabling continuous and stable high-speed online imaging. This ensures that the surface of each bean grain is uniformly illuminated and its trajectory is stable, providing high-quality raw image data for subsequent polarization analysis. Second, a polarization beam splitter is set in the camera's receiving optical path to separate the reflected light from the bean surface into parallel polarized light and orthogonal polarized light with mutually perpendicular polarization directions. Parallel polarized images and orthogonal polarized images are acquired, which can physically separate the specular reflection component from the diffuse reflection component, avoiding irreversible loss of reflective information in single-channel imaging and providing complementary raw image pairs for subsequent adaptive fusion. Then, the parallel polarized images and orthogonal polarized images are adaptively fused to generate suppressed reflective interference. This invention employs a polarization-fusion image method that suppresses surface reflection interference and enhances structural information. By adaptively fusing the generated polarization-fusion image, pixel-level intelligent weight allocation is used to completely suppress areas of strong specular reflection while dynamically preserving detailed information in parallel polarized light based on local texture complexity, significantly enhancing the inherent texture and uneven structural features of the bean surface. Finally, frequency domain features are extracted from the polarization-fusion image to obtain texture periodic features characterizing the regularity of the bean skin texture and distortion energy features characterizing the degree of surface undulation. These texture periodic features and distortion energy features are then input into a pre-constructed feature discrimination model for defect identification. This transforms the structural information in the fused image into stable frequency domain quantification indicators. Combined with a Mahalanobis distance discrimination model based on the statistical distribution of normal samples, accurate identification and reliable judgment of surface defects are achieved. In summary, this application's solution can achieve polarization-fusion imaging to suppress surface reflection interference, thereby improving the accuracy of bean surface defect detection. Attached Figure Description

[0016] Figure 1 This is a schematic diagram illustrating an application scenario of an identification method for a soybean grain screening device in soybean milk, according to some embodiments of this application. Figure 2 This is an exemplary flowchart of an identification method for a soybean grain screening device for soybean milk, according to some embodiments of this application; Figure 3 This is a schematic diagram illustrating the process of defect identification using a feature discrimination model according to some embodiments of this application; Figure 4 This is a schematic diagram of the structure of an identification device for a soybean granule screening equipment for soybean milk, according to some embodiments of this application; Figure 5 This is a schematic diagram of the structure of a computer device that implements an identification method for a soybean sieving device for soybean milk, according to some embodiments of this application. Detailed Implementation

[0017] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] refer to Figure 1 As shown in the figure, this figure is a schematic diagram of an application scenario of an identification method for a soybean sieving device for soy milk, according to some embodiments of this application. The figure includes three main components: a data acquisition device, a server, and a data storage device. The data acquisition device is used to scan and image the surface of the soybeans in motion, acquire image data of the soybean surface, and then send the acquired image data to the server through a communication network. The server runs the execution code for the identification method for the soybean sieving device for soy milk, and finally the server stores the processed identification result in the data storage device.

[0019] refer to Figure 2 As shown, this figure is an exemplary flowchart of an identification method for a soybean grain screening device for soybean milk, according to some embodiments of this application. The identification method for a soybean grain screening device for soybean milk mainly includes the following steps: In step 101, the soybeans to be tested are dispersed into a sequence of individual beans by a single-layer vibrating feeding mechanism and transported to the detection station. The surface of the moving soybeans is scanned and imaged using a laser light source arranged coaxially with the intelligent line scan camera.

[0020] It should be noted that the single-layer vibrating feeding mechanism in this application includes a vibrating conveying trough with a single-layer conveying surface and a vibrating driving device connected thereto. The vibrating driving device is used to disperse and arrange the beans along a predetermined direction and convey them to the detection station in a single-bean interval state by controlling the vibration frequency and amplitude. The detection station in this application includes an imaging detection area located at the end of the single-layer vibrating feeding mechanism. The imaging detection area is provided with a light-transmitting support surface and an optical mounting structure corresponding to the intelligent line scanning camera and the laser light source, which is used to keep the movement trajectory of the beans stable and within the predetermined imaging focal plane when the beans pass through.

[0021] In some embodiments, the dispersion of the soybeans to be tested into a sequence of individual beans and their transport to the testing station via a single-layer vibrating feeding mechanism can be achieved as follows: First, the soybeans to be tested are placed into a storage hopper located above the starting end of the vibrating conveyor trough. The discharge port of the storage hopper is set to a height slightly higher than the thickness of a single soybean, allowing the soybeans to initially fall onto the surface of the vibrating conveyor trough in the form of a thinner layer. Then, the vibration drive device is activated. This device typically employs an electromagnetic vibrator or a vibrating motor with an eccentric block. By adjusting the vibration frequency and amplitude, directional micro-vibration is generated, causing the soybeans to jump forward along the conveyor trough under the action of this vibration. In this process, the surface of the conveying trough is typically designed to be flat or slightly concave, with a width that allows only a single bean to pass through. An anti-slip coating can be applied to the trough surface to increase friction. During the vibration conveying process, the beans gradually adjust their posture due to continuous inertial and frictional forces, causing stacked beans to automatically slide off or separate. Ultimately, upon reaching the testing station, they form a stable flow of beans arranged individually with intervals between them. Finally, by adjusting the vibration parameters and the inclination angle of the conveying trough, the conveying speed of the beans is matched to the line frequency acquisition speed of the subsequent intelligent line scan camera, thus achieving continuous and stable delivery of the beans to be tested one by one to the testing station.

[0022] In some embodiments, scanning and imaging the surface of a moving bean using a laser light source arranged coaxially with the intelligent line scan camera can be achieved through the following steps: The laser source is activated to emit a linear beam, which is then guided by a coaxial optical assembly to a position that coincides with the optical axis of the intelligent line scan camera and projected onto the material passage area of ​​the inspection station. The line frequency parameters of the intelligent line scan camera are set according to the conveying speed of the single-layer vibrating feeder to keep the image acquisition frequency matched with the movement speed of the beans. As the beans pass through the inspection station, the intelligent line scan camera continuously collects the reflected light signals from the surface of the beans, line by line, generating a continuous two-dimensional image data stream.

[0023] It should be noted that the coaxial optical component in this application is an optical device used to make the laser beam coincide with the optical axis of the camera; the line frequency parameter is a set value that controls the number of lines scanned per second by the intelligent line scan camera, used to match the acquisition frequency with the material conveying speed.

[0024] In practical applications, firstly, the laser source emits a linear beam. The coaxial optical assembly guides this beam to a position coinciding with the optical axis of the intelligent line scan camera, projecting it onto the material passage area of ​​the inspection station. This can be achieved as follows: A line laser is selected as the laser source to emit a linear beam. This beam is incident on the coaxial optical assembly, which uses a semi-transparent, semi-reflective prism for coaxial illumination. The prism reflects the beam, causing its propagation direction to change and precisely align it with the optical axis of the intelligent line scan camera in the same straight line direction. The aligned beam is then projected vertically downwards onto the material passage area of ​​the inspection station. This material passage area is set to be along the vertical... In the elongated strip-shaped area extending in the conveying direction, its length covers the width of the conveying trough, and its width matches the focal depth range of the linear beam. This ensures that the surface of each bean passing through this area is uniformly illuminated by the linear beam. Finally, the linear beam is precisely projected onto the material passage area as the illumination condition for the bean to be tested. Secondly, the line frequency parameters of the intelligent line scan camera are set according to the conveying speed of the single-layer vibrating feeder, ensuring that the image acquisition frequency matches the bean movement speed. This can be achieved by installing a rotary encoder on the drive shaft or side of the conveying trough of the single-layer vibrating feeder. This encoder generates a signal proportional to the conveying speed as the conveying mechanism operates. The pulse signal is input to the external trigger interface of the intelligent line scan camera as a line frequency trigger source. Simultaneously, based on the number of pulses per encoder revolution and the transmission ratio of the conveyor mechanism, the distance the bean moves corresponding to each pulse is calculated. This allows the camera to set its internal frequency division coefficient, ensuring that the camera captures one line of image data for each pulse output by the encoder, and that the distance the bean moves within that pulse interval is exactly equal to the physical size corresponding to a single pixel of the image sensor. This achieves precise synchronization between the acquisition frequency and the movement speed. Then, as the bean passes through the detection station, the intelligent line scan camera continuously acquires the reflected light signal from the bean surface line by line, generating a continuous... The two-dimensional image data stream can be implemented in the following way: The camera starts continuous scanning based on the line frequency signal triggered by the encoder. Each time its linear array image sensor receives a trigger pulse, it completes the exposure of a line of photosensitive pixels, converting the intensity of reflected light on the surface of the bean at that line position into an analog electrical signal. After being converted into digital pixel values ​​by the internal analog-to-digital conversion circuit, the signal is output. As the bean moves forward continuously, the camera sequentially collects data from each line. These continuous lines of data are arranged and stitched from top to bottom in the image processing unit according to the acquisition time order, thereby generating a complete two-dimensional grayscale image. This image completely records the grayscale distribution information of each position on the surface of the bean.

[0025] In step 102, a polarization beam splitter is set in the camera receiving optical path to separate the light reflected from the surface of the bean into parallel polarized light and orthogonal polarized light with mutually perpendicular polarization directions, and parallel polarization images and orthogonal polarization images are acquired.

[0026] In some embodiments, a polarization beam splitter is set in the camera receiving optical path to separate the light reflected from the surface of the bean into parallel polarized light and orthogonal polarized light with mutually perpendicular polarization directions, and the parallel polarization image and orthogonal polarization image are acquired. This can be achieved by the following steps: A polarizing beam splitter is set at the front end of the lens of an intelligent line scan camera, and the incident and reflected light is separated into a first optical path and a second optical path by the polarizing beam splitter. A first polarizer is set in the first optical path so that its polarization direction is consistent with the output polarization direction of the laser source, and the transmitted light signal is projected onto the first image sensor to obtain a parallel polarized image; A second polarizer is set in the second optical path so that its polarization direction is perpendicular to the output polarization direction of the laser source, and the transmitted light signal is projected onto the second image sensor to obtain an orthogonally polarized image.

[0027] It should be noted that the polarizing beam splitter in this application is an optical element used to decompose reflected light from the surface of a bean into two beams of light with mutually perpendicular propagation directions according to their polarization states; the first optical path is the propagation path of reflected light transmitted through the polarizing beam splitter, with its polarization direction parallel to the polarization direction emitted from the laser source; the second optical path is the propagation path of reflected light reflected by the polarizing beam splitter, with its polarization direction perpendicular to the polarization direction emitted from the laser source; the parallel polarization image is image data acquired through the first optical path, used to characterize the specular reflection component and gloss distribution of the bean surface; the orthogonal polarization image is image data acquired through the second optical path, used to suppress specular reflection interference and characterize the diffuse reflection component and inherent texture distribution of the bean surface.

[0028] In practical applications, firstly, a polarizing beam splitter is installed at the front of the lens of the intelligent line scan camera. The polarizing beam splitter separates the incident and reflected light into a first optical path and a second optical path. This can be achieved as follows: The polarizing beam splitter can be installed behind the lens of the intelligent line scan camera and in front of the image sensor, aligning its optical axis with the lens's optical axis. When the reflected light from the surface of the bean is converged by the lens and incident on the polarizing beam splitter, the optical thin film interface inside the prism separates the incident light into a first optical path in the transmission direction and a second optical path in the reflection direction. The propagation directions of the two light rays are at a 90-degree angle, and their polarization directions are perpendicular to each other. The transmitted light is... The reflected light is orthogonally polarized; secondly, a first polarizer is set in the first optical path so that its polarization direction is consistent with the output polarization direction of the laser source, and the transmitted light signal is projected onto the first image sensor to obtain a parallel polarized image. This can be achieved by the following method: a first polarizer is installed on the output light of the first optical path, and the transmission direction of the first polarizer is rotated to be parallel to the output polarization direction of the laser source. The light rays of the first optical path separated by the polarizing beam splitter are already parallel polarized light. After the first polarizer further filters out possible stray light, it is projected onto the photosensitive surface of the first image sensor, and the first image... The sensor converts the optical signal into an electrical signal and generates image data. This image mainly contains the specular reflection component of the bean surface, so the image generated by the first image sensor can be used as the parallel polarized image in this application. Then, a second polarizer is set in the second optical path, making its polarization direction perpendicular to the output polarization direction of the laser source, and the transmitted light signal is projected onto the second image sensor to obtain an orthogonal polarized image. This can be achieved by installing a second polarizer on the output light of the second optical path, and rotating the transmission direction of the polarizer to make it perpendicular to the output polarization direction of the laser source. According to the polarization principle, the specularly reflected light will maintain its original polarization. The direction of the light is effectively blocked by the polarizer perpendicular to the direction of the light. Due to the depolarization effect, the polarization direction of the diffuse light tends to be randomized. Therefore, there is always a component that is consistent with the transmission direction of the second polarizer and can pass through. This allows the orthogonal polarization channel to mainly receive diffuse light from the lower layer of the bean surface, effectively suppressing the interference of surface mirror reflection and highlighting the inherent color and texture information of the bean. The light signal after being selected by the second polarizer is projected onto the photosensitive surface of the second image sensor. The second image sensor converts the light signal into an electrical signal and generates image data. Finally, the light signal of the second optical path is converted into an orthogonal polarization image as image data characterizing the surface texture structure.

[0029] In step 103, the parallel polarization image and the orthogonal polarization image are adaptively fused to generate a polarization fused image that suppresses surface reflection interference and enhances structural information.

[0030] In some embodiments, adaptively fusing the parallel polarization image and the orthogonal polarization image to generate a polarization-fused image that suppresses surface reflection interference and enhances structural information can be achieved through the following steps: For each pixel, the specular reflection index is calculated based on the gray value of the pixel in the parallel polarization image and the gray value in the orthogonal polarization image, which characterizes the degree of specular reflection affecting the pixel. Based on the orthogonal polarization image, the gray-level variance of each pixel within its preset neighborhood window is calculated as the texture complexity index of the local region where the pixel is located. For each pixel, based on the comparison between the specular reflection index of all pixels in the neighborhood window of that pixel and a preset threshold, it is determined whether the pixel is located in a region affected by strong specular reflection. Based on the judgment result and the texture complexity index, the weight coefficient of the pixel during fusion is determined. For pixels that are determined to be located in areas affected by strong specular reflection, their weight coefficient is set to a first preset value so that the gray values ​​of the orthogonal polarization image are fully used during fusion. For pixels that are not determined to be located in areas affected by strong specular reflection, the weight coefficient is dynamically allocated according to their texture complexity index so that the contribution ratio of the parallel polarization image is greater in areas with higher texture complexity. Based on the determined weighting coefficients, the gray values ​​of the pixel in the parallel polarization image and the gray values ​​in the orthogonal polarization image are weighted and averaged and fused, and the fusion result is used as the pixel value at the corresponding position in the polarization fused image. By iterating through all pixels, a polarization-fused image is generated that suppresses surface reflection interference and enhances structural information.

[0031] It should be noted that the specular reflection index in this application is a quantitative indicator used to characterize the degree of specular reflection influence on each pixel; the preset neighborhood window is a fixed-size region used to determine the local pixel range centered on the current pixel; the grayscale variance is a statistical measure used to measure the dispersion of grayscale values ​​of all pixels within the preset neighborhood window; the texture complexity index is a numerical value used to reflect the texture complexity of the local area where the pixel is located; the preset threshold is a critical value used to determine whether the specular reflection index is significant; the strong specular reflection influence area refers to the area in the local area where the pixel is located where the specular reflection phenomenon is severe; the weighting coefficient is a numerical value used to control the contribution ratio of parallel polarization image and orthogonal polarization image in the fusion process; the first preset value is a pre-set fixed weight value; and the polarization fused image is an output image used to suppress specular reflection interference and enhance surface structure information.

[0032] In practical applications, firstly, for each pixel, the specular reflection index, which characterizes the degree of specular reflection affecting the pixel, is calculated based on the pixel's grayscale value in the parallel polarization image and its grayscale value in the orthogonal polarization image. This can be achieved as follows: For each pixel, obtain its grayscale value in the parallel polarization image and its grayscale value in the orthogonal polarization image, calculate the absolute value of the difference between the two grayscale values, and then divide it by the sum of the two grayscale values ​​and a very small positive constant. The resulting ratio is used as the specular reflection index of the pixel, where the very small positive constant is used to avoid the denominator being zero. Secondly, using the orthogonal polarization image as a reference, calculate the grayscale variance of each pixel within its preset neighborhood window, which is used as the local variance of the pixel's grayscale value. The texture complexity index of a local region can be implemented as follows: Using an orthogonal polarization image as a reference, for each pixel, a preset neighborhood window is determined centered on it. The average grayscale value of all pixels within this window is calculated. Then, the sum of the squares of the differences between the grayscale value of each pixel and the average value is calculated and divided by the total number of pixels within the window. The result is used as the grayscale variance of that pixel, and this grayscale variance is used as the texture complexity index of the local region where that pixel is located. It should be further noted that the neighborhood window is typically set to a 5×5 pixel or 7×7 pixel square area. Next, for each pixel, based on the comparison between the specular reflection index of all pixels within the neighborhood window and a preset threshold, the texture complexity index of that pixel is determined. Whether a pixel is located in a region heavily influenced by specular reflection can be determined as follows: It should be noted that the preset threshold is a critical value used to determine whether the specular reflection index is significant. Typically, the average value is used as the preset threshold to distinguish between regions dominated by specular reflection and regions dominated by diffuse reflection. For each pixel, the number of pixels with a specular reflection index greater than the preset threshold within its preset neighborhood window is counted. The proportion of these pixels to the total number of pixels in the window is calculated. If this proportion exceeds a preset proportion threshold, the pixel is determined to be located in a region heavily influenced by specular reflection. It should be further noted that this preset proportion threshold can be 30%, and can also be adjusted based on historical data. This is merely an example and is not intended to limit the specific scope of the invention. Then, based on the judgment result and the texture complexity index, the weight coefficient of the pixel during fusion is determined. For pixels determined to be located in areas affected by strong specular reflection, their weight coefficient is set to a first preset value so that the gray values ​​of the orthogonal polarization image are fully used during fusion. For pixels not determined to be located in areas affected by strong specular reflection, the weight coefficient is dynamically allocated according to their texture complexity index so that the contribution ratio of the parallel polarization image is greater in areas with higher texture complexity. This can be achieved in the following way: For pixels determined to be located in areas affected by strong specular reflection, their weight coefficient is set to a first preset value, which is set to 0, so that the gray values ​​of the orthogonal polarization image are fully used during fusion.For pixels not identified as being located in areas affected by strong specular reflection, their texture complexity index is normalized to obtain a normalized value between 0 and 1. This normalized value is used as the weight coefficient for that pixel, so that regions with higher texture complexity have larger weight coefficients, i.e., the contribution ratio of the parallel polarization image is larger. Further, based on the determined weight coefficient, a weighted average fusion is performed on the gray values ​​of the pixel in the parallel polarization image and the gray values ​​in the orthogonal polarization image. The fusion result is used as the pixel value at the corresponding position in the polarization fused image. This can be achieved as follows: Based on the determined weight coefficient, the gray values ​​of each pixel in the parallel polarization image and the gray values ​​in the orthogonal polarization image are weighted averaged. A weighted average fusion calculation is performed, specifically as follows: the gray value of the orthogonal polarization image is multiplied by 1 minus the difference of the weighting coefficients, and then the gray value of the parallel polarization image is multiplied by the weighting coefficients. The result is used as the pixel value at the corresponding position in the polarization fusion image. Finally, all pixels are traversed to generate a polarization fusion image that suppresses surface reflection interference and enhances structural information. This can be achieved as follows: traversing all pixels in the image in row-first, column-second order, calculating new pixel values ​​sequentially, and storing the calculated pixel values ​​in the corresponding positions of the new image matrix. After all pixels have been processed, the generated complete two-dimensional image is used as the polarization fusion image that suppresses surface reflection interference and enhances structural information.

[0033] In step 104, frequency domain features are extracted from the polarization fusion image to obtain texture periodic features that characterize the regularity of the bean skin texture and distortion energy features that characterize the degree of surface undulation of the bean.

[0034] In some embodiments, frequency domain feature extraction of the polarization fusion image to obtain texture periodic features characterizing the regularity of bean skin texture and distortion energy features characterizing the degree of surface undulation of bean can be achieved by the following steps: The polarization fusion image is subjected to frequency domain transformation processing to convert the polarization fusion image from the spatial domain to the frequency domain to obtain a spectrum map; In the spectrum, identify the frequency component peak corresponding to the main energy concentration region, and extract the frequency value corresponding to the peak as a texture periodic feature characterizing the regularity of the epidermal texture; The total energy of the preset high-frequency regions in the spectrum is statistically analyzed, and the total energy is used as the distortion energy feature characterizing the degree of surface undulation of the bean.

[0035] It should be noted that the frequency component peak value in this application is the frequency point with the largest energy amplitude in the main energy concentration area of ​​the spectrum diagram, which is used to characterize the period length of the dominant texture on the bean surface; the texture regularity is a quantitative index reflecting the uniformity and periodicity of the texture arrangement on the bean surface; the texture periodicity feature is a numerical feature extracted from the spectrum diagram to characterize the periodic repetition frequency of the texture on the bean surface; the distortion energy feature is the sum of the energy in the preset high-frequency area of ​​the spectrum diagram, which is used to quantify the severity of grayscale abrupt changes caused by unevenness defects on the bean surface.

[0036] In practical applications, firstly, the polarization-fused image undergoes frequency domain transformation to convert it from the spatial domain to the frequency domain to obtain a spectrum. This can be achieved as follows: The polarization-fused image can be used as input data, and a two-dimensional discrete Fourier transform algorithm is employed to transform the entire image, resulting in a complex matrix composed of real and imaginary parts. The modulus of this complex matrix is ​​then calculated point by point to obtain the amplitude spectrum matrix. To facilitate observation and subsequent processing, a logarithmic transformation is performed on the amplitude spectrum to compress the dynamic range, followed by a centering shift operation. This shifts the low-frequency components to the center of the spectrum and the high-frequency components to the periphery, thus obtaining a standard-form spectrum. The central region of this spectrum represents... The overall outline and gently changing areas of the image represent high-frequency information such as details, textures, and edges. Secondly, identifying the frequency component peaks corresponding to the main energy concentration areas in the spectrogram and extracting the frequency values ​​corresponding to these peaks as texture periodic features characterizing the regularity of the epidermal texture can be achieved as follows: Starting from the center point of the spectrogram, scan pixel by pixel outwards in the radial direction to find the pixel with the largest energy amplitude. Since regular textures form energy rings or peaks at specific radial distances in the frequency domain, the radial distance corresponding to this maximum energy point is the main frequency value of the texture. This radial distance is converted into a frequency value and output as the texture periodic feature. For normal soybeans with regular and uniform surface texture, their spectrograms typically show clear and concentrated energy rings, with stable corresponding texture periodic feature values. For soybeans with disordered surface textures or unevenness, the main energy peak is often diffused or shifted, leading to abnormal feature values. Ultimately, the frequency value corresponding to the main energy peak is used as the texture periodic feature characterizing the regularity of the skin texture. Then, the sum of energy in the preset high-frequency region of the spectrogram is statistically analyzed, and this sum of energy is used as the distortion energy feature characterizing the degree of surface undulation of the soybean. This can be achieved in the following way: Based on the pixel resolution of the intelligent line scan camera and the typical size of the soybean, a frequency threshold is predetermined. This threshold corresponds to the frequency in the spatial domain that can reflect... The minimum scale of the defect is determined by identifying all pixels with a radial distance greater than the threshold in the obtained spectrum. The annular region formed by these pixels is taken as the preset high-frequency region. The energy amplitude of all pixels in this region is summed to obtain the total energy. Since surface unevenness defects can cause drastic changes in local grayscale in the image, these changes are manifested as a significant enhancement of high-frequency components in the frequency domain. Therefore, the larger the distortion energy feature value, the more uneven the bean surface and the more serious the defect. Conversely, the total high-frequency energy of normal beans with smooth and flat surfaces is usually maintained at a low level. Finally, the total energy of the preset high-frequency region is taken as the distortion energy feature characterizing the degree of surface undulation of the bean.

[0037] In step 105, the texture periodic features and the distortion energy features are input into a pre-constructed feature discrimination model for defect identification, and then the presence of defects on the surface of the bean is determined based on the defect identification results.

[0038] In some embodiments, reference Figure 3 As shown in the figure, this is a flowchart illustrating defect identification using a feature discrimination model in some embodiments of this application. In this embodiment, the defect identification of the texture periodic features and the distortion energy features by inputting them into a pre-constructed feature discrimination model can be achieved through the following steps: In step 1051, images of multiple standard bean samples with no surface defects are pre-acquired, and the corresponding texture periodic features and distortion energy features are extracted to construct a standard feature dataset. In step 1052, the feature distribution center and distribution boundary range are determined based on the standard feature dataset, and a feature discrimination model for distinguishing normal regions from abnormal regions is established. In step 1053, the texture periodic features and distortion energy features obtained from the extracted soybean grains are input into the feature discrimination model to obtain the corresponding defect identification results.

[0039] It should be noted that the feature distribution center in this application is a statistical benchmark point reflecting the typical characteristic state of normal soybean grains; the distribution boundary range is a critical region of the feature space that reflects the dispersion of the characteristic distribution of normal soybean grains and is used to distinguish between normal and abnormal regions; the feature discrimination model is a mathematical model used to determine whether a soybean grain belongs to the normal range based on its feature value.

[0040] In practical applications, firstly, images of multiple standard soybean samples with no surface defects are pre-acquired, and the corresponding texture periodic features and distortion energy features are extracted respectively. The standard feature dataset can be constructed as follows: A certain number of standard soybeans, manually screened and confirmed to be free of any surface defects, are selected. These soybeans are then sequentially transported to the detection station via the aforementioned single-layer vibrating feeding mechanism. Using coaxial laser scanning imaging, polarization beam splitting, polarization difference analysis, and frequency domain feature extraction steps, the texture periodic feature value and distortion energy feature value of each standard soybean are obtained. The two feature values ​​of each soybean are then used as a single feature set. A standard feature dataset is constructed from the feature vectors of all standard beans, forming a two-dimensional feature vector. This dataset serves as the basis for establishing a discriminant model. The feature distribution center and boundary range are determined based on this dataset, and a feature discriminant model is established to distinguish between normal and abnormal regions. This can be achieved as follows: Statistical analysis is performed on all two-dimensional feature vectors in the standard feature dataset. The mean vector is calculated as the feature distribution center, and the covariance matrix is ​​calculated to describe the dispersion of the feature distribution. A Mahalanobis distance threshold is determined based on a preset confidence level as the distribution boundary range. This threshold ensures that a corresponding proportion of points in the standard samples lie within this boundary. The mean vector, covariance matrix, and Mahalanobis distance threshold are then used to construct a feature discriminant model. This model can calculate the Mahalanobis distance from any feature vector to the distribution center and compare it with the threshold to determine whether the feature vector belongs to the normal region. Finally, the feature discriminant model is used as the basis for defect identification. It should be further noted that the preset confidence level in this application can be set as follows: As a preferred embodiment, the confidence level is determined based on statistical distribution theory, specifically based on the samples in the standard feature dataset. Based on the characteristic distribution of texture periodicity and distortion energy features, assuming that they follow a two-dimensional normal distribution, the square of the Mahalanobis distance from the sample feature vector to the distribution center follows a chi-square distribution with 2 degrees of freedom. Based on this assumption, a confidence level is set according to the required detection sensitivity. This confidence level is usually between 95% and 99%. Then, the critical value corresponding to 2 degrees of freedom is obtained by looking up the chi-square distribution table as the Mahalanobis distance threshold, so that the proportion of standard samples within this threshold is exactly equal to the set confidence level. Thus, the region in the feature space within this threshold range is defined as the normal region, and the region outside the range is defined as the abnormal region.Finally, the texture periodicity features and distortion energy features extracted from the bean grains to be tested are input into the feature discrimination model to obtain the corresponding defect identification results. This can be achieved in the following way: For each bean grain that has passed through the inspection station, its texture periodicity features and distortion energy features are extracted to form a feature vector to be tested. This feature vector is input into the pre-established feature discrimination model, and the Mahalanobis distance from it to the feature distribution center is calculated. If the Mahalanobis distance is less than or equal to a preset distribution boundary threshold, the bean grain is determined to have no defects on its surface, and the identification result is normal. If the distance is greater than the preset distribution boundary threshold, the bean grain is determined to have unevenness defects on its surface, and the identification result is abnormal.

[0041] In some embodiments, determining whether a bean surface has a defect based on the defect identification result can be achieved in the following way: First, obtain the defect identification result output by the feature discrimination model. This result is usually represented in numerical form, such as the Mahalanobis distance value from the feature vector of the bean to be tested to the center of the standard feature distribution. Second, compare the Mahalanobis distance value with a pre-set Mahalanobis distance threshold. If the Mahalanobis distance value is less than or equal to the threshold, it is determined that the bean surface has no defect. If the Mahalanobis distance value is greater than the threshold, it is determined that the bean surface has an unevenness defect. The threshold comparison operation is a basic logical judgment method in the field of digital signal processing and can be implemented by comparison instructions in a microprocessor or a hardware comparator circuit. This process is the same as the threshold comparison judgment method commonly used in industrial inspection, and both are common knowledge in the field. Finally, generate a corresponding judgment conclusion based on the comparison result and use the conclusion as the triggering basis for the subsequent rejection control signal. In the above way, the abstract numerical value output by the model is transformed into a clear judgment result of whether a defect exists, realizing a complete closed loop from feature analysis to practical application.

[0042] On the other hand, in some embodiments, this application provides an identification device for a soybean granule screening equipment for soybean milk, with reference to... Figure 4 The figure is a schematic diagram of the structure of an identification device for a soybean sieving equipment for soy milk, according to some embodiments of this application. The identification device 400 for the soybean sieving equipment for soy milk includes: a scanning module 401, a feature processing module 402, and an identification module 403, which are described below: Scanning module 401, in this application, is mainly used to disperse the bean grains to be tested into a sequence of individual grains through a single-layer vibrating feeding mechanism and transport them to the detection station, and to scan and image the surface of the moving bean grains using a laser light source arranged coaxially with the intelligent line scan camera. Feature processing module 402, in this application, is used to set a polarization beam splitter in the camera receiving optical path to separate the light reflected from the surface of the bean into parallel polarized light and orthogonal polarized light with polarization directions perpendicular to each other, and to acquire parallel polarization image and orthogonal polarization image; In this application, the feature processing module 402 is also used to adaptively fuse the parallel polarization image and the orthogonal polarization image to generate a polarization fused image that suppresses surface reflection interference and enhances structural information. In this application, the feature processing module 402 is also used to extract frequency domain features from the polarization fusion image to obtain texture periodic features that characterize the regularity of the bean skin texture and distortion energy features that characterize the degree of surface undulation of the bean. The identification module 403 in this application is mainly used to input the texture periodic features and the distortion energy features into a pre-constructed feature discrimination model for defect identification, and then determine whether there are defects on the surface of the bean based on the defect identification results.

[0043] In addition, this application also provides a computer device, the computer device including a memory and a processor, the memory storing code, the processor being configured to acquire the code and execute the above-described identification method for a soybean granule sieving device for soy milk.

[0044] In some embodiments, reference Figure 5 The figure is a schematic diagram of the structure of a computer device implementing a method for identifying soybean particles in a soybean milk screening device, according to some embodiments of this application. The identification method for soybean particles in a soybean milk screening device described in the above embodiments can be achieved through... Figure 5 The computer device shown is used to implement this, and the computer device 500 includes at least one processor 501, a communication bus 502, a memory 503, and at least one communication interface 504.

[0045] Processor 501 can be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0046] The communication bus 502 can be used to transmit information between the aforementioned components.

[0047] Memory 503 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CDROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory 503 may exist independently and be connected to processor 501 via communication bus 502. Memory 503 may also be integrated with processor 501.

[0048] The memory 503 stores program code for executing the scheme of this application, and its execution is controlled by the processor 501. The processor 501 executes the program code stored in the memory 503. The program code may include one or more software modules. The identification method for the soybean granule screening device in the above embodiment can be implemented by the processor 501 and one or more software modules in the program code in the memory 503.

[0049] Communication interface 504 uses any transceiver-like device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0050] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single CPU) processor or a multi-core (multi CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0051] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0052] In addition, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described identification method for a soybean granule sieving device for soybean milk.

[0053] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0054] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A recognition method for a soybean milk soybean particle screening device, applied to a recognition device of a soybean milk soybean particle screening device, the recognition device comprising an intelligent line scan camera, characterized in that, The method includes: The soybeans to be tested are dispersed into a sequence of individual beans by a single-layer vibrating feeding mechanism and transported to the testing station. The surface of the moving soybeans is scanned and imaged using a laser light source arranged coaxially with an intelligent line scan camera. A polarization beam splitter is set in the camera's receiving optical path to separate the light reflected from the bean surface into parallel polarized light and orthogonal polarized light with mutually perpendicular polarization directions, and to acquire parallel polarization images and orthogonal polarization images. Adaptive fusion of the parallel polarization image and the orthogonal polarization image is performed to generate a polarization fused image that suppresses surface reflection interference and enhances structural information. Frequency domain features are extracted from the polarization fusion image to obtain texture periodic features that characterize the regularity of the bean skin texture and distortion energy features that characterize the degree of surface undulation of the bean. The texture periodic features and the distortion energy features are input into a pre-constructed feature discrimination model for defect identification, and then the presence of defects on the surface of the bean is determined based on the defect identification results.

2. The method of claim 1, wherein, The scanning and imaging of the surface of a moving bean using a laser light source arranged coaxially with an intelligent line scan camera specifically includes: The laser source is activated to emit a linear beam, which is then guided by a coaxial optical assembly to a position that coincides with the optical axis of the intelligent line scan camera and projected onto the material passage area of ​​the inspection station. The line frequency parameters of the intelligent line scan camera are set according to the conveying speed of the single-layer vibrating feeder to keep the image acquisition frequency matched with the movement speed of the beans. As the beans pass through the inspection station, the intelligent line scan camera continuously collects the reflected light signals from the surface of the beans, line by line, generating a continuous two-dimensional image data stream.

3. The method of claim 1, wherein, A polarization beam splitter is set up in the camera's receiving optical path to separate the light reflected from the bean surface into parallel polarized light and orthogonal polarized light with mutually perpendicular polarization directions, and to acquire parallel polarization images and orthogonal polarization images. Specifically, this includes: A polarizing beam splitter is set at the front end of the lens of an intelligent line scan camera, and the incident and reflected light is separated into a first optical path and a second optical path by the polarizing beam splitter. A first polarizer is set in the first optical path so that its polarization direction is consistent with the output polarization direction of the laser source, and the transmitted light signal is projected onto the first image sensor to obtain a parallel polarized image; A second polarizer is set in the second optical path so that its polarization direction is perpendicular to the output polarization direction of the laser source, and the transmitted light signal is projected onto the second image sensor to obtain an orthogonally polarized image.

4. The method as described in claim 1, characterized in that, Adaptive fusion of the parallel polarization image and the orthogonal polarization image to generate a polarization-fused image that suppresses surface reflection interference and enhances structural information specifically includes: For each pixel, the specular reflection index is calculated based on the gray value of the pixel in the parallel polarization image and the gray value in the orthogonal polarization image, which characterizes the degree of specular reflection affecting the pixel. Based on the orthogonal polarization image, the gray-level variance of each pixel within its preset neighborhood window is calculated as the texture complexity index of the local region where the pixel is located. For each pixel, based on the comparison between the specular reflection index of all pixels in the neighborhood window of that pixel and a preset threshold, it is determined whether the pixel is located in a region affected by strong specular reflection. Based on the judgment result and the texture complexity index, the weight coefficient of the pixel during fusion is determined. For pixels that are determined to be located in areas affected by strong specular reflection, their weight coefficient is set to a first preset value so that the gray values ​​of the orthogonal polarization image are fully used during fusion. For pixels that are not determined to be located in areas affected by strong specular reflection, the weight coefficient is dynamically allocated according to their texture complexity index so that the contribution ratio of the parallel polarization image is greater in areas with higher texture complexity. Based on the determined weighting coefficients, the gray values ​​of the pixel in the parallel polarization image and the gray values ​​in the orthogonal polarization image are weighted and averaged and fused, and the fusion result is used as the pixel value at the corresponding position in the polarization fused image. By iterating through all pixels, a polarization-fused image is generated that suppresses surface reflection interference and enhances structural information.

5. The method as described in claim 4, characterized in that, The determination of whether a pixel is located in a region affected by strong specular reflection specifically includes: counting the proportion of pixels in the neighborhood window where the pixel's specular reflection feature value is greater than a preset threshold to the total number of pixels in the window; if the proportion exceeds the preset proportion threshold, then the pixel is determined to be located in a region affected by strong specular reflection.

6. The method as described in claim 1, characterized in that, The defect identification process specifically includes inputting the texture periodicity features and the distortion energy features into a pre-constructed feature discrimination model: Multiple standard bean samples with no surface defects were pre-acquired, and their corresponding texture periodic features and distortion energy features were extracted to construct a standard feature dataset. Based on the standard feature dataset, the feature distribution center and distribution boundary range are determined, and a feature discrimination model for distinguishing normal and abnormal regions is established. The texture periodic features and distortion energy features extracted from the bean grains to be tested are input into the feature discrimination model to obtain the corresponding defect identification results.

7. The method as described in claim 1, characterized in that, The single-layer vibrating feeding mechanism includes a vibrating conveying trough with a single-layer conveying surface and a vibration driving device connected thereto. The vibration driving device is used to disperse and arrange the beans along a predetermined direction and convey them to the detection station in a single-bean interval state by controlling the vibration frequency and amplitude.

8. An identification device for soybean grain screening equipment in soybean milk production, characterized in that, include: The scanning module is used to disperse the soybeans to be tested into a sequence of individual beans through a single-layer vibrating feeding mechanism and transport them to the detection station. It uses a laser light source arranged coaxially with the intelligent line scan camera to scan and image the surface of the moving soybeans. The feature processing module is used to set up a polarization beam splitter in the camera receiving optical path to separate the light reflected from the surface of the bean into parallel polarized light and orthogonal polarized light with polarization directions perpendicular to each other, and to acquire parallel polarization images and orthogonal polarization images. The feature processing module is also used to adaptively fuse the parallel polarization image and the orthogonal polarization image to generate a polarization fused image that suppresses surface reflection interference and enhances structural information. The feature processing module is also used to extract frequency domain features from the polarization fusion image to obtain texture periodic features that characterize the regularity of the bean skin texture and distortion energy features that characterize the degree of surface undulation of the bean. The identification module is used to input the texture periodic features and the distortion energy features into a pre-constructed feature discrimination model for defect identification, and then determine whether there are defects on the surface of the bean based on the defect identification results.

9. A computer device comprising a memory and a processor, the memory storing code, characterized in that, The processor is configured to acquire the code and execute the identification method for a soybean granule sieving device for soy milk as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the identification method for a soybean granule screening device for soybean milk as described in any one of claims 1 to 7.

Images