Intelligent Identification Method and System for Broken Spherical Grain Particles

Abstract

This invention relates to the field of grain quality testing technology, and discloses an intelligent identification method and system for broken spherical grains. It aims to solve the problems of poor efficiency and accuracy in existing methods. The main steps include: continuously and sequentially conveying the spherical grains to be tested through an imaging area without overlap; applying transmitted light to each passing grain from above and below within the imaging area, and acquiring a transmission image of the grain grain under the transmitted light from above; using an image instance segmentation algorithm to segment instance regions of individual grains from the continuously acquired transmission images; analyzing the grayscale distribution characteristics in the transmission image for each instance region; identifying areas with increased translucency due to localized thickness reduction based on the grayscale distribution characteristics, and determining whether the grain grain is broken. This invention improves the efficiency and accuracy of broken grain detection through intelligent imaging technology and is suitable for the intelligent identification of quality indicators of spherical grains such as rapeseed and soybeans.

Description

Technical Field

[0001] This invention relates to the field of grain quality testing technology, specifically to an intelligent identification method and system for broken spherical grain grains. Background Technology

[0002] In the grain purchasing, processing, and storage logistics process, the integrity of spherical or near-spherical grain kernels (such as rapeseed, soybeans, peanuts, peas, mung beans, and corn kernels) is one of the key quality indicators for assessing their quality grade. Broken kernels not only affect the commercial appearance and pricing of grain kernels, but also damage their protective outer layer, making them prone to mold and pests during storage, and accelerating fat oxidation, which seriously affects storage stability and the quality of processed products. For example, broken rapeseed kernels directly affect the oil yield, oil quality, storage safety, and processing efficiency of rapeseed.

[0003] Currently, the detection of broken grain kernels mainly relies on the following two methods:

[0004] The first type is manual visual inspection. Inspectors manually sift out broken particles by visual observation. This method has inherent drawbacks such as low efficiency, high labor intensity, strong subjectivity of inspection results, and poor repeatability among different inspectors, making it difficult to meet the large-scale, rapid, and standardized quality inspection needs of modern grain processing and distribution.

[0005] The second category is automated inspection methods based on machine vision. Existing machine vision systems mostly employ single-view reflected light imaging technology. This involves shining light over the grain grain and acquiring the reflected image of the grain surface, then using image processing algorithms to analyze features such as color, texture, or shape to identify anomalies. However, this approach has significant limitations: it can only effectively capture visible damage on the grain surface. Broken defects in spherical grain grains often manifest as localized thickness reduction (such as shallow surface defects, missing corners, or broken edges), and these defects are weakly characterized under reflected light, making them blind spots for detection. Especially for grains with a certain degree of translucency, the reduced physical thickness in the broken area leads to localized increased translucency, but this crucial physical signal cannot be acquired under reflected light imaging mode, resulting in poor detection accuracy for broken grains.

[0006] Application publication number CN110412045A discloses a rapid detection device and method for internal cracks in corn seeds using transmission light. This method utilizes transmitted light to detect density differences or voids within the grains, and is effective in identifying obvious internal cracks, insect infestations, and other defects. However, when this technology is directly applied to the detection of broken spherical grain kernels, there are still significant shortcomings, specifically:

[0007] "Broken grains" is a specific term in agricultural quality standards, encompassing various forms such as fragmentation, defects, cracks, and flattening. Their common physical essence is a reduction in thickness and mass loss, either locally or overall, which differs significantly from a single crack defect in both morphology and cause. Existing transmission light methods primarily target the identification of linear defects like cracks, with image processing and judgment logic designed around the extraction of linear features. However, broken grains exhibit complex and diverse morphologies, including large-area missing corners and fragmentation, and their image features are characterized by irregular shapes. Algorithm models designed for cracks struggle to effectively cover and accurately determine breakage defects caused by thickness loss, failing to identify the diverse forms of broken grains and resulting in poor detection accuracy. Summary of the Invention

[0008] This invention aims to solve the problems of poor efficiency and accuracy in existing broken grain detection methods, and proposes an intelligent identification method and system for spherical grain broken grains.

[0009] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0010] In a first aspect, the present invention provides a method for intelligent identification of broken spherical grain kernels, the method comprising:

[0011] The spherical grains to be tested are continuously, one by one, and without overlap, conveyed through the imaging area; in the imaging area, transmitted light is applied to the individual grain grains passing through from above and below, and a transmission image of the grain grains under transmitted light is acquired from above;

[0012] An image instance segmentation algorithm is used to segment the instance region of a single grain grain from continuously acquired transmission images;

[0013] For each grain grain instance area, the grayscale distribution characteristics in its transmission image are analyzed; based on the grayscale distribution characteristics, areas where light transmittance is enhanced due to local thickness reduction are identified, and it is determined whether the grain grain is a broken grain.

[0014] Furthermore, after segmenting the instance regions of individual grains, the method further includes:

[0015] Calculate the center coordinates of each instance region;

[0016] Based on the center coordinates of the instance region in multiple consecutive transmission images, it is determined whether they are the same grain grain, and repeated images of the same grain grain are filtered to ensure that each grain grain is detected only once.

[0017] Furthermore, the method for determining whether the grain particles are broken particles specifically includes:

[0018] Calculate the overall average gray value and / or gray standard deviation of the instance region corresponding to the grain grain in the transmission image;

[0019] When the overall average gray value of the instance area is higher than the first preset threshold, and / or the gray standard deviation is higher than the second preset threshold, the grain is determined to be a broken grain.

[0020] Furthermore, the method for determining whether the grain particles are broken particles specifically includes:

[0021] If a local connected region exists in the instance region of the transmission image corresponding to a grain grain, then the average gray value of the local connected region is calculated.

[0022] When the difference between the average gray value of the local connected region and the overall average gray value of the instance region is greater than a third preset threshold, the grain particle is determined to be a broken particle.

[0023] Furthermore, the method also includes:

[0024] The percentage of broken grains is calculated based on the total number of grain particles in the spherical grains to be tested and the number of grain particles identified as broken grains.

[0025] Furthermore, the method also includes:

[0026] After determining that the grain particles are broken, estimate the average remaining thickness of the grain particles after they are broken.

[0027] The maximum projected area of ​​the grain grain is determined based on the corresponding instance area, and the total volume of the complete grain is calculated based on the maximum projected area and the reference thickness of the standard grain.

[0028] Based on the total volume, average remaining thickness, reference thickness, and maximum projected area of ​​the intact grain, the broken mass percentage of the grain grain is calculated using the following formula:

[0029] ;

[0030] in, Indicates the first The percentage of crushed particles by their crushed mass. Indicates the reference thickness. Indicates the average remaining thickness. Indicates the first The maximum projected area of ​​each broken particle This represents the total volume of the intact particle.

[0031] Furthermore, the average remaining thickness is calculated as follows:

[0032] A physical optical model is constructed that correlates image grayscale data with physical thickness parameters. The overall average grayscale value of the instance region of the transmission image corresponding to the grain particle is substituted into the physical optical model to obtain the average remaining thickness of the grain particle after it is broken. The physical optical model is as follows:

[0033] ;

[0034] in, This represents the overall average gray value of the instance region in the transmission image. This represents the reference grayscale value, obtained through direct imaging calibration without particle occlusion. This represents the average extinction coefficient of the grain being tested. The average remaining thickness, Represents the natural constant.

[0035] Furthermore, the method also includes:

[0036] The total mass percentage of broken grains in the tested spherical grains is calculated based on the mass percentage of each broken grain. The calculation formula is as follows:

[0037] ;

[0038] in, This indicates the percentage of total mass of crushed particles. Indicates the first The percentage of crushed particles by their crushed mass. Indicates the first The maximum projected area of ​​each broken particle This indicates the number of grain particles identified as broken. Indicates the first The maximum projected area of ​​a grain grain This indicates the total number of grain particles in the spherical grain kernels being tested.

[0039] Furthermore, the spherical grains are any one of rapeseed, soybean, peanut, pea, mung bean, or corn kernels.

[0040] In a second aspect, the present invention provides an intelligent identification system for broken spherical grain kernels, used to implement the intelligent identification method for broken spherical grain kernels described in the first aspect, the system comprising:

[0041] An image acquisition device is used to continuously, one grain at a time and without overlap transport spherical grains to be tested through an imaging area; in the imaging area, transmitted light is applied to the passing individual grain grains from above and below, and a transmission image of the grain grains under transmitted light is acquired from above;

[0042] An image recognition device is used to segment instance regions of individual grain particles from continuously acquired transmission images using an image instance segmentation algorithm; for each instance region of a grain particle, the gray-scale distribution characteristics in its transmission image are analyzed; based on the gray-scale distribution characteristics, regions with enhanced light transmittance due to local thickness reduction are identified, and the grain particle is determined to be a broken particle.

[0043] The beneficial effects of this invention are as follows: The intelligent identification method and system for broken spherical grain kernels provided by this invention, based on intelligent imaging technology, captures internal cracks and thickness defects invisible under reflected light by employing transmission imaging and analyzing areas in the image where local thickness reduction leads to increased light transmittance. This achieves high-precision and high-reliability identification of broken kernels, effectively reducing the false negative rate. Compared to existing technologies that use a single algorithm targeting cracks, this invention, based on the physical principle of "thickness-transmittance-grayscale characteristics," can effectively identify broken kernels of various forms, such as fragments, defects, and flattening, improving the identification accuracy of broken kernels. Furthermore, this invention achieves full automation from sample introduction, imaging, analysis to statistics, requiring no manual intervention. Its detection efficiency is far higher than manual methods, enabling intelligent identification of grain kernel quality indicators. It is applicable to spherical grain kernels such as rapeseed and soybeans. Attached Figure Description

[0044] Figure 1 A flowchart illustrating the intelligent identification method for broken spherical grain kernels provided in this embodiment;

[0045] Figure 2 This is a schematic diagram of the image acquisition device provided in the embodiment;

[0046] Figure 3 A schematic diagram of the intelligent identification system for broken spherical grain kernels provided in this embodiment;

[0047] Explanation of reference numerals in the attached figures:

[0048] 1-Vibrating feeder; 2-High-transmittance rotating tray; 3-Camera; 4-Upper light source module; 5-Lower light source module; 6-Sample collection device. Detailed Implementation

[0049] While existing technologies can detect specific cracks using transmitted light, they suffer from limitations in detecting a single target and inability to identify fragmented particles with diverse shapes. Therefore, this invention addresses this issue. By analyzing regions of enhanced light transmittance in the transmitted image caused by localized thickness reduction, various types of fragmented particles can be accurately and efficiently identified, achieving efficient and reliable detection of broken spherical grain grains.

[0050] Specifically, the core of this invention lies in leveraging intelligent imaging technology and the inherent semi-transparent properties of spherical grains. Through transmission imaging, the key defect of physical thickness reduction caused by breakage is transformed into easily detectable gray-scale anomalies in the transmission image, enabling precise identification. The technical approach spans three levels: physical optics, image processing, and defect identification. First, when uniformly transmitted light passes through the grain, according to the Lambert-Beer law, the local thickness reduction caused by breakage leads to an exponential increase in the intensity of the emitted light. Subsequently, the camera captures the enhanced light signal as a digital image, making the broken area appear as a bright area with significantly higher local gray-scale than the normal area. Finally, by analyzing the gray-scale distribution characteristics of the image (such as overall statistics and / or local connected regions), these bright areas caused by "thickness reduction - increased light transmittance" can be specifically identified, thus reliably determining the broken grain. This invention transforms changes in internal physical properties into visible and quantifiable image information, achieving high-precision detection of thickness-related defects that are difficult to detect with reflected light imaging.

[0051] The technical solutions in this embodiment 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.

[0052] Figure 1 A flowchart illustrating an intelligent identification method for broken spherical grain kernels is shown. Please refer to [link / reference]. Figure 1 The method includes the following steps:

[0053] Step 1: The spherical grains to be tested are continuously, one by one, and without overlap, transported through the imaging area; in the imaging area, transmitted light is applied to the individual grain grains passing through from above and below, and a transmission image of the grain grains under transmitted light is acquired from above.

[0054] In this embodiment, a transmission image is acquired using an image acquisition device. Please refer to [link / reference]. Figure 2 The image acquisition device includes a vibrating feeder 1, a high-transmittance rotating tray 2, a camera 3, an upper light source module 4, a lower light source module 5, and a sample collection device 6.

[0055] In practical applications, the vibrating feeder 1, in conjunction with the high-transmittance rotating carrier disk 2, enables continuous, one-by-one, and non-overlapping conveying of grains through the imaging area. Within the imaging area, coaxially arranged upper and lower light source modules 4 and 5 (e.g., LED surface light sources) provide a uniform transmitted light field. An upper camera 3 (e.g., a global shutter CMOS camera) synchronously acquires the transmitted images of the grain grains and sends these images to an image recognition device for broken grain detection. After the transmitted images are acquired, the grain grains are conveyed to a sample collection device 6 for recovery.

[0056] In this embodiment, the spherical grain is any one of rapeseed, soybean, peanut, pea, mung bean or corn kernel.

[0057] This embodiment utilizes the inherent translucent properties of spherical grains. According to the Lambert-Beer law, when transmitted light passes through a material, the light intensity attenuation is negatively exponentially related to the penetration thickness. Therefore, a reduction in local thickness (i.e., broken areas) leads to a significant increase in transmitted light intensity. By employing transmission imaging and analyzing the areas in the image where increased translucency is caused by local thickness reduction, internal cracks and thickness defects invisible under reflected light can be captured. This enables high-precision and high-reliability identification of broken particles, effectively reducing the false negative rate.

[0058] Step 2: Use an image instance segmentation algorithm to segment the instance region of a single grain grain from the continuously acquired transmission images.

[0059] This step is used to separate the foreground (individual grains) from the background in the transmission image and generate an independent pixel-level mask (instance region) for each unconnected grain region, so as to facilitate subsequent single-grain statistics and analysis.

[0060] In practical applications, image instance segmentation algorithms, such as deep learning models like Mask R-CNN or YOLOv8-seg, can be used to accurately segment the outline of each grain of food from a complex background.

[0061] Building upon this, to further improve the accuracy of online statistics, this step also includes:

[0062] Calculate the center coordinates of each instance region; based on the center coordinates of the instance regions in multiple consecutive frames of transmission images, determine whether they are the same grain grain, and filter duplicate images of the same grain grain so that each grain grain is detected only once.

[0063] In the above steps, instance segmentation ensures that the granularity of the analysis reaches the single-particle level. Combined with repetition filtering, it completely solves the technical problem caused by the same particle being captured multiple times in high-speed continuous imaging, making the result of the proportion of broken particles truly reliable.

[0064] Step 3: For each grain grain instance area, analyze the grayscale distribution characteristics in its transmission image; based on the grayscale distribution characteristics, identify the areas where light transmittance is enhanced due to local thickness reduction, and determine whether the grain grain is a broken grain.

[0065] The essence of broken particles is thickness reduction, which directly manifests as an abnormal grayscale distribution in transmission images. Based on grayscale distribution characteristics, this embodiment can use one or more of the following two methods to detect broken particles:

[0066] The first method is the overall statistical method:

[0067] Calculate the overall average gray value and / or gray standard deviation of the instance area corresponding to the grain particle in the transmission image; when the overall average gray value of the instance area is higher than a first preset threshold, and / or the gray standard deviation is higher than a second preset threshold, the grain particle is determined to be a broken particle.

[0068] Specifically, broken grains have increased overall or localized light transmittance, resulting in an average gray value higher than normal grains. Simultaneously, because damaged and intact areas coexist, their internal gray value consistency is poor, manifesting as an increased gray standard deviation. Based on this, this embodiment can determine whether a grain grain is broken based on an overall average gray value exceeding a first preset threshold and / or a gray standard deviation exceeding a second preset threshold. The first and second preset thresholds can be determined based on experimental gray value data of intact grain grains, and this embodiment does not impose any limitations on this.

[0069] The second method is the local feature method:

[0070] If a local connected region exists in the instance area of ​​the transmission image corresponding to a grain particle, the average gray value of the local connected region is calculated. When the difference between the average gray value of the local connected region and the overall average gray value of the instance area is greater than a third preset threshold, the grain particle is determined to be a broken particle.

[0071] Specifically, by identifying whether there are isolated local connected regions in the instance area and calculating the average gray value of the local connected region, if the difference between its value and the overall background gray value of the particle exceeds a third preset threshold, the local area is determined to be a bright broken area formed by severe thickness reduction.

[0072] The overall statistical method is highly efficient and suitable for particles that are entirely broken or have dense cracks; the local feature method is extremely sensitive to defects such as localized damage and missing corners. Both methods can be used individually or in combination, ensuring high coverage and accuracy of the judgment logic for particles with various shapes of breakage.

[0073] In addition, in practical applications, a deep learning classification model for the grain to be tested can be pre-trained directly. The gray-scale distribution features corresponding to the grain are input into the deep learning classification model, and the classification result of the grain is directly output.

[0074] In this embodiment, after completing the qualitative determination, quantitative analysis is also included:

[0075] The percentage of broken grains is calculated based on the total number of grain particles in the spherical grains to be tested and the number of grain particles identified as broken grains.

[0076] Specifically, after each grain kernel is tested for broken grains, the ratio of the number of broken grain kernels to the total number of grain kernels is the percentage of broken grains. This provides a quick and intuitive quality assessment indicator that directly matches manual testing standards.

[0077] In this embodiment, the method further includes:

[0078] Step 4: After determining that the grain particles are broken, estimate the average remaining thickness of the grain particles after breaking. The calculation method is as follows:

[0079] A physical optical model is constructed that correlates image grayscale data with physical thickness parameters. The overall average grayscale value of the instance region of the transmission image corresponding to the grain particle is substituted into the physical optical model to obtain the average remaining thickness of the grain particle after it is broken. The physical optical model is as follows:

[0080] ;

[0081] in, This represents the overall average gray value of the instance region in the transmission image. This represents the reference grayscale value, obtained through direct imaging calibration without particle occlusion. This represents the average extinction coefficient of the grain being tested. The average remaining thickness, Represents the natural constant.

[0082] The theoretical basis for this step is the Lambert-Beer law. This law describes how the intensity of light propagating in a uniform medium decreases exponentially with increasing penetration distance (i.e., thickness). For spherical grains with translucent properties, the overall average gray value of its transmission image... Proportional to the intensity of the emitted light, a physical optical model can be constructed based on this. When a particle breaks, its average remaining thickness... It will decrease, resulting in its overall average gray value. Increase. This model allows us to obtain the measured grayscale values. The physical thickness parameters of the particles are determined by inversion.

[0083] In practical applications, a blank image without particles is acquired, and its average gray level is calculated as a reference gray level value. Furthermore, by measuring a batch of standard samples with known thicknesses, the average extinction coefficient was fitted. For each instance region identified as a fragmented particle, its overall average gray value is extracted. By substituting the physical optics model, the average remaining thickness after fragmentation can be obtained. .

[0084] Step 5: Determine the maximum projected area of ​​the grain particles according to the corresponding instance area, and calculate the total volume of the complete particles based on the maximum projected area and the reference thickness of the standard particles.

[0085] This step is based on a geometric model approximation. The spherical grain grain is approximated as a cylinder or oblate spheroid, and its volume can be calculated from its maximum projected area and a characteristic thickness. For grains of the same variety, a reference thickness for a standard grain can be defined.

[0086] In practical applications, the precise pixel-level mask for the grain particles has already been obtained during instance segmentation in step 2. The total number of pixels in this mask represents the maximum projected pixel area of ​​the grain particle. Multiplying this by the actual physical area of ​​a single pixel yields the true maximum projected area. This refers to the total area of ​​the outline projected by the grain particles onto the imaging plane perpendicular to the camera's optical axis. Based on the selected geometric model, using... and Perform the calculations. For example, in the case of a cylindrical model, we have: .in, It represents the volume that a grain of grain should have if it were intact, and is the benchmark for calculating mass loss.

[0087] Step 6: Based on the total volume, average remaining thickness, reference thickness, and maximum projected area of ​​the intact grain, calculate the broken mass percentage of the grain grain. The calculation formula is as follows:

[0088] ;

[0089] in, Indicates the first The percentage of crushed particles by their crushed mass. Indicates the reference thickness. Indicates the average remaining thickness. Indicates the first The maximum projected area of ​​each broken particle This represents the total volume of the intact particle.

[0090] This step is based on the direct proportionality between volume and mass. Assuming uniform particle density, the mass percentage equals the volume percentage. The damaged portion caused by breakage is approximated as a region with a height of... The base area is The column. Therefore, the volume of the missing part. Crushed mass percentage It is the ratio of the damaged volume to the intact volume.

[0091] In practical applications, the calculated average remaining thickness will be... Maximum projected area and the total volume of the intact particles and known reference thickness Substituting these values ​​into the formula, we can obtain the percentage of crushed grain mass. .

[0092] By following the steps above, the percentage of crushed grain particles can be accurately calculated, thereby achieving precise quantification of the degree of defects.

[0093] In this embodiment, the method further includes:

[0094] Step 7: Calculate the total mass percentage of broken grains in the tested spherical grains based on the mass percentage of each broken grain. The calculation formula is as follows:

[0095] ;

[0096] in, This indicates the percentage of total mass of crushed particles. Indicates the first The percentage of crushed particles by their crushed mass. Indicates the first The maximum projected area of ​​each broken particle This indicates the number of grain particles identified as broken. Indicates the first The maximum projected area of ​​a grain grain This indicates the total number of grain particles in the spherical grain kernels being tested.

[0097] Specifically, for spherical grains of the same variety and with uniform density, the mass of a single grain is directly proportional to its volume. For grains with similar shapes, there is a stable positive correlation between their volume and their maximum projected area. Therefore, the maximum projected area can effectively serve as a proxy variable for grain mass. Grains with larger areas contribute more to the total mass of the entire batch.

[0098] Based on this, this embodiment calculates the proportion of total mass of broken particles by comparing the weighted sum of mass losses of all broken particles with the estimated total mass of all grain particles to be tested. Specifically, when calculating the weighted sum of mass losses of all broken particles, the mass loss contributed by each broken particle is its own degree of breakage. Compared to its own size The product of the product and the weighted average is calculated by weighting the mass of each broken grain. The estimated total mass of all grain grains under test is represented by the total projected area. The proportion of total mass of broken grains directly reflects the actual economic loss caused by breakage. This indicator can provide accurate data feedback for seed breeding, harvesting parameter adjustment, and processing technology optimization.

[0099] In summary, the intelligent identification method for broken spherical grains provided in this embodiment utilizes the semi-transparent nature of spherical grains to transform difficult-to-observe physical thickness defects into easily identifiable grayscale anomalies in the image through transmission imaging. This fundamentally solves the technical bottleneck of traditional reflected light imaging's insensitivity to internal cracks and thickness direction defects. Instance segmentation ensures the accuracy of single-grain analysis, and combined with a coordinate-based duplicate particle filtering mechanism, it effectively overcomes the statistical distortion problem caused by repeated imaging of the same particle in online detection, ensuring the accuracy of statistical results in high-throughput detection environments. Furthermore, this embodiment achieves a leap from qualitative judgment to quantitative evaluation, not only providing the basic indicator of the proportion of broken grains but also innovatively mapping grayscale features to thickness parameters by establishing an optical physical model. This allows for the calculation of the proportion of broken quality that accurately reflects actual economic losses, achieving intelligent identification of grain quality indicators and providing a direct and objective pricing basis for commercial grading of grains.

[0100] Based on the above technical solution, this embodiment also proposes an intelligent identification system for broken spherical grain kernels, used to implement the intelligent identification method for broken spherical grain kernels described in the embodiment. Please refer to [link to relevant documentation]. Figure 3 The system includes:

[0101] An image acquisition device is used to continuously, one grain at a time and without overlap transport spherical grains to be tested through an imaging area; in the imaging area, transmitted light is applied to the passing individual grain grains from above and below, and a transmission image of the grain grains under transmitted light is acquired from above;

[0102] An image recognition device is used to segment instance regions of individual grain particles from continuously acquired transmission images using an image instance segmentation algorithm; for each instance region of a grain particle, the gray-scale distribution characteristics in its transmission image are analyzed; based on the gray-scale distribution characteristics, regions with enhanced light transmittance due to local thickness reduction are identified, and the grain particle is determined to be a broken particle.

[0103] It is understood that since the intelligent identification system for broken spherical grains described in this embodiment is a system for implementing the intelligent identification method for broken spherical grains described in the embodiment, the system disclosed in the embodiment is relatively simple to describe because it corresponds to the method disclosed in the embodiment. For relevant parts, please refer to the description of the method, and it will not be repeated here.

Claims

1. A method for intelligent identification of broken spherical grain kernels, characterized in that, The method includes: The spherical grains to be tested are continuously, one by one, and without overlap, conveyed through the imaging area; in the imaging area, transmitted light is applied to the individual grain grains passing through from above and below, and a transmission image of the grain grains under transmitted light is acquired from above; An image instance segmentation algorithm is used to segment the instance region of a single grain grain from continuously acquired transmission images; For each grain grain instance area, the grayscale distribution characteristics in its transmission image are analyzed; based on the grayscale distribution characteristics, areas where light transmittance is enhanced due to local thickness reduction are identified, and it is determined whether the grain grain is a broken grain. After determining that the grain particles are broken, estimate the average remaining thickness of the grain particles after they are broken. The maximum projected area of ​​the grain grain is determined based on the corresponding instance area, and the total volume of the complete grain is calculated based on the maximum projected area and the reference thickness of the standard grain. Based on the total volume, average remaining thickness, reference thickness, and maximum projected area of ​​the intact grain, the broken mass percentage of the grain grain is calculated using the following formula: ； in, Indicates the first The percentage of crushed particles by their crushed mass. Indicates the reference thickness. Indicates the average remaining thickness. Indicates the first The maximum projected area of ​​each broken particle This represents the total volume of the intact particle.

2. The intelligent identification method for broken spherical grain kernels according to claim 1, characterized in that, After segmenting the instance regions of individual grains, the method further includes: Calculate the center coordinates of each instance region; Based on the center coordinates of the instance region in multiple consecutive transmission images, it is determined whether they are the same grain grain, and repeated images of the same grain grain are filtered to ensure that each grain grain is detected only once.

3. The intelligent identification method for broken spherical grain kernels according to claim 1, characterized in that, The method for determining whether the grain particles are broken includes: Calculate the overall average gray value and / or gray standard deviation of the instance region corresponding to the grain grain in the transmission image; When the overall average gray value of the instance area is higher than the first preset threshold, and / or the gray standard deviation is higher than the second preset threshold, the grain is determined to be a broken grain.

4. The intelligent identification method for broken spherical grain kernels according to claim 1, characterized in that, The method for determining whether grain particles are broken particles also includes: If a local connected region exists in the instance region of the transmission image corresponding to a grain grain, then the average gray value of the local connected region is calculated. When the difference between the average gray value of the local connected region and the overall average gray value of the instance region is greater than a third preset threshold, the grain particle is determined to be a broken particle.

5. The intelligent identification method for broken spherical grain kernels according to claim 1, characterized in that, The method further includes: The percentage of broken grains is calculated based on the total number of grain particles in the spherical grains to be tested and the number of grain particles identified as broken grains.

6. The intelligent identification method for broken spherical grain kernels according to claim 1, characterized in that, The average remaining thickness is calculated as follows: A physical optical model is constructed that correlates image grayscale data with physical thickness parameters. The overall average grayscale value of the instance region of the transmission image corresponding to the grain particle is substituted into the physical optical model to obtain the average remaining thickness of the grain particle after it is broken. The physical optical model is as follows: ； in, This represents the overall average gray value of the instance region in the transmission image. This represents the reference grayscale value, obtained through direct imaging calibration without particle occlusion. This represents the average extinction coefficient of the grain being tested. The average remaining thickness, Represents the natural constant.

7. The intelligent identification method for broken spherical grain kernels according to claim 1, characterized in that, The method further includes: The total mass percentage of broken grains in the tested spherical grains is calculated based on the mass percentage of each broken grain. The calculation formula is as follows: ； in, This indicates the percentage of total mass of crushed particles. Indicates the first The percentage of crushed particles by their crushed mass. Indicates the first The maximum projected area of ​​each broken particle This indicates the number of grain particles identified as broken. Indicates the first The maximum projected area of ​​a grain grain This indicates the total number of grain particles in the spherical grain kernels being tested.

8. The intelligent identification method for broken spherical grain kernels according to claim 1, characterized in that, The spherical grains are any one of rapeseed, soybeans, peanuts, peas, mung beans, or corn kernels.

9. An intelligent identification system for broken spherical grain kernels, characterized in that, For implementing the intelligent identification method for broken spherical grain kernels as described in any one of claims 1 to 8, the system comprises: An image acquisition device is used to continuously, one grain at a time and without overlap transport spherical grains to be tested through an imaging area; in the imaging area, transmitted light is applied to the passing individual grain grains from above and below, and a transmission image of the grain grains under transmitted light is acquired from above; An image recognition device is used to segment instance regions of individual grain particles from continuously acquired transmission images using an image instance segmentation algorithm; for each instance region of a grain particle, the gray-scale distribution characteristics in its transmission image are analyzed; based on the gray-scale distribution characteristics, regions with enhanced light transmittance due to local thickness reduction are identified, and the grain particle is determined to be a broken particle.

Images