Machine vision-based online detection device for integrity of silk seedlings

Abstract

The application relates to a rice grain detection technical field, in particular to a silk sprout rice integrity online detection device based on machine vision. The device comprises a transmission assembly, a rice storage assembly and a detection assembly. A control execution unit comprises an overlap identification module which calculates an area overlap probability index through morphological characteristics, automatically distinguishes overlapping and non-overlapping areas; a rapid judgment module directly extracts morphological characteristics of non-overlapping rice grains, judges integrity through a complete morphological index; an intelligent reconstruction module fuses three-dimensional image data, calculates a depth layering coefficient to identify upper and lower layer relationships, and generates a virtual contour of a hidden part based on visible end feature parameters; an overlap judgment module calculates a virtual complete morphological index to evaluate the integrity of single rice grains in the overlapping area; and an optimization calibration module comprehensively calculates a detection performance drift coefficient to realize self-optimization calibration of the system. The application solves the problem of integrity detection of overlapping silk sprout rice.

Description

Technical Field

[0001] This invention relates to the field of rice grain detection technology, and in particular to an online detection device for the integrity of silk rice based on machine vision. Background Technology

[0002] Rice integrity testing is a core aspect of rice processing quality control. Currently, mainstream testing methods include manual visual inspection and automated inspection based on two-dimensional machine vision. Manual inspection is inefficient, labor-intensive, and its standards are highly subjective, failing to meet the efficiency and consistency requirements of modern production lines. While two-dimensional machine vision-based inspection technology has achieved automation to some extent, it has inherent limitations in practical applications: when rice grains overlap or stick together on the conveyor belt, the two-dimensional image cannot provide spatial hierarchy information, causing the system to misclassify multiple grains as a single connected region, resulting in serious missed and false detections of broken rice. This problem is particularly pronounced in long, thin rice varieties.

[0003] Therefore, there is an urgent need in this field for an online detection device that can efficiently and accurately solve the problem of overlapping detection of high-quality rice, while also taking into account cost and efficiency.

[0004] Chinese Patent Publication No. CN105139405A discloses a visual separation and detection method for overlapping broken rice grains and whole rice grains, comprising two processes: overlapping rice grain image separation and rice grain appearance structure quality detection and segmentation. The overlapping rice grain image separation process uses the angle formed by the circular area of ​​the rice grain image and the overlapping rice grain image combined with a concave point detection method to separate non-overlapping and overlapping rice grain images. The rice grain appearance structure quality detection and segmentation process uses structural parameters of rice grain length, width, and area to distinguish between whole rice grains and incomplete broken rice grains, and uses the concave points to be matched to obtain matching segmentation lines to segment the overlapping rice grain images. Based on this invention, the visual separation and detection method for overlapping broken rice grains can separate overlapping rice grains and identify whether the rice grains are whole. It has the advantages of high automation and intelligence, effective detection and segmentation methods, high detection rate, and good real-time performance, laying an important foundation for subsequent interpretation of rice grain image information.

[0005] Therefore, the aforementioned visual separation and detection method for overlapping broken rice and whole rice has the following problems:

[0006] 1. During the detection process, concave point detection and straight line segmentation methods are used for overlapping rice grains. When there is obvious overlapping of upper and lower layers of rice grains, this method will lead to the occurrence of missed detection of lower layer rice grains and a significant decrease in detection accuracy.

[0007] 2. During the detection process, overlapping rice grains are separated by segmentation, and the separated rice grains are then inspected. This can lead to distortion of the shape of the segmented rice grains, making it easy to misjudge whole rice as broken rice in the subsequent integrity assessment. Summary of the Invention

[0008] To address this, the present invention provides an online detection device for the integrity of high-quality rice based on machine vision, which overcomes the problems in the prior art that it cannot detect rice grains with obvious overlapping layers, and that the detection accuracy is reduced and false phenomena occur due to the distortion of rice grain shape.

[0009] To achieve the above objectives, this invention provides an online inspection device for the integrity of high-quality rice based on machine vision. It includes:

[0010] A transmission assembly used to transport the rice grains to be tested for integrity detection;

[0011] The detection assembly includes a two-dimensional color camera and a three-dimensional depth camera disposed on the upper part of the conveyor belt to acquire image data of the rice to be detected, and a parallel light source disposed in the middle of the conveyor belt corresponding to the two-dimensional color camera and the three-dimensional depth camera to illuminate the rice to be detected.

[0012] The overlap identification module is used to obtain the region overlap probability index based on the morphological features of several connected regions to determine the overlap type of the current region;

[0013] The rapid determination module is used to directly extract morphological features from non-overlapping areas of the silky rice and obtain the integrity morphology index to determine the integrity of the silky rice in non-overlapping areas.

[0014] The intelligent reconstruction module is used to combine three-dimensional image data to obtain the depth layering coefficient to determine the upper and lower layer relationship of the overlapping area of ​​the glutinous rice, and to identify and analyze the end feature parameters of the visible part of each glutinous rice to generate the virtual outline of the occluded part.

[0015] The overlap determination module is used to calculate the virtual integrity morphology index of a single virtual rice grain under reasonable reconstruction conditions to determine the integrity of a single virtual rice grain in the overlapping area.

[0016] The optimization calibration module is used to calculate the detection performance drift coefficient based on the complete morphology index and the virtual complete morphology index to determine whether the current detection state meets the standard. If it does not meet the standard, it determines to extend the exposure time of the two-dimensional color camera, increase the driving current of the parallel light source, and move the lower and upper limits of the typical aspect ratio fluctuation range closer to the actual mode of the current statistical distribution based on the relative deviation of the complete morphology index and the relative deviation of the virtual complete morphology index.

[0017] Furthermore, it also includes an image acquisition module, which is used to acquire two-dimensional and three-dimensional image data of the rice to be detected;

[0018] The image processing module is used to perform grayscale conversion, filtering and noise reduction, and background segmentation on the acquired image data;

[0019] The real-time control module receives the processed image and calculates the overall image quality coefficient to determine the parameters for adjusting the camera and light source;

[0020] The results output module is used to receive, integrate, and output the final detection results.

[0021] Furthermore, in response to the overall image quality coefficient being less than a preset overall image quality coefficient, the real-time control module determines to increase the driving current of the parallel light source and extend the exposure time of the two-dimensional color camera.

[0022] Furthermore, in response to the real-time control module determining that the image processed by the image processing module meets the detection requirements, the overlap recognition module determines whether the current region is an overlapping region or a non-overlapping region based on the comparison result between the region overlap probability index and the preset region overlap probability index.

[0023] Based on the fact that the region overlap probability index is less than or equal to the preset region overlap probability index, the current region is determined to be a non-overlapping region.

[0024] Based on the fact that the region overlap probability index is greater than the preset region overlap probability index, the current region is determined to be an overlapping region.

[0025] Furthermore, the rapid determination module responds to the overlap recognition module in determining that the current region is a non-overlapping region, and determines the integrity of the rice grains in the non-overlapping region based on the comparison result between the integrity morphology index and the preset integrity morphology index.

[0026] Based on the fact that the complete morphology index is less than the preset complete morphology index, it is determined that the rice in the non-overlapping area is incomplete, and defective rice is output.

[0027] Furthermore, in response to the overlap recognition module determining that the current region is an overlapping region, the intelligent reconstruction module determines the upper and lower layer relationship of the overlapping region's silk rice based on the comparison result between the depth layering coefficient and the preset depth layering coefficient, wherein,

[0028] Based on the fact that the depth layering coefficient is greater than the preset depth layering coefficient, it is determined that the overlapping area has a clear upper and lower layering relationship;

[0029] Based on the fact that the depth layering coefficient is less than or equal to the preset depth layering coefficient, the overlapping areas are determined to be adhesions or slight intersections within the same plane.

[0030] Furthermore, the intelligent reconstruction module divides the pixels into upper and lower layers based on the clear upper and lower layer relationship of the overlapping area of ​​the rice, identifies the set of pixels with larger depth values ​​as the visible part of the lower layer of rice and locates the unobstructed end position, and generates a virtual outline of the obstructed part.

[0031] The intelligent reconstruction module, based on the overlapping area being the sticky or slightly intersecting silk rice in the same plane, directly uses an endpoint detection algorithm to locate the unobstructed end position of the visible part of each silk rice to be processed in several connected areas, and generates a virtual outline of the obstructed part.

[0032] Furthermore, in response to the intelligent reconstruction module, the virtual contour of the reconstructed occluded portion is verified as reasonable, and the integrity of a single virtual rice grain in the overlapping area is determined based on the comparison result between the virtual completeness index and the preset virtual completeness index.

[0033] Based on the fact that the virtual completeness index is less than the preset virtual completeness index, it is determined that a single virtual silk rice in the overlapping area is incomplete, and defective rice is output.

[0034] Based on the virtual completeness index being greater than or equal to the preset virtual completeness index, the integrity of a single virtual silk rice in the overlapping area is determined.

[0035] Furthermore, the optimization calibration module determines that the current detection state is substandard based on the detection performance drift coefficient being greater than the preset detection performance drift coefficient, and determines to extend the exposure time of the two-dimensional color camera and increase the driving current of the parallel light source based on the relative deviation of the complete morphology index being greater than the relative deviation of the virtual complete morphology index. The adjustment range is positively correlated with the detection performance drift coefficient.

[0036] Furthermore, the optimization calibration module determines that the current detection state is substandard based on the detection performance drift coefficient being greater than the preset detection performance drift coefficient, and determines that the lower and upper limits of the typical aspect ratio fluctuation range should be moved closer to the actual mode of the current statistical distribution based on the fact that the relative deviation of the complete morphology index is less than the relative deviation of the virtual complete morphology index. The adjustment range is positively correlated with the detection performance drift coefficient.

[0037] Compared with existing technologies, the advantages of this invention lie in its ability to intelligently analyze the spatial structure of overlapping areas by fusing two-dimensional and three-dimensional visual information and proposing a virtual contour reconstruction algorithm. For lower-layer or occluded rice grains, the system automatically generates a virtual contour of the occluded portion based on the end-point vector and curvature features, combined with a pre-stored model of the morphological characteristics of rice grains. This achieves accurate restoration of the complete rice grain shape, fundamentally overcoming misjudgments caused by overlapping rice grains and significantly improving the detection accuracy of overlapping areas.

[0038] Furthermore, the virtual contour reconstruction algorithm of this invention deeply integrates the morphological prior knowledge unique to Silky Rice. The key parameters in its morphological feature model, such as the typical length range, aspect ratio range, and width variation law, are all derived from a large amount of statistical analysis of Silky Rice. This makes the virtual contour reconstruction and integrity judgment more targeted, and significantly improves the recognition accuracy and adaptability of this variety compared to general rice detection models.

[0039] Furthermore, this invention calculates a comprehensive image quality coefficient through a real-time control module to evaluate the acquired image and compares it with a preset comprehensive image quality coefficient. When the quality is substandard, the driving current of the parallel light source is automatically increased by 10%, and the exposure time of the two-dimensional color camera is increased by 5ms to ensure that the image data input to subsequent algorithms is always clear and stable, thus guaranteeing the reliability of the overall detection process from the source.

[0040] Furthermore, this invention employs an overlap recognition module to intelligently divide the glutinous rice region. By calculating the region overlap probability index and comparing it with a preset region overlap probability index, the detection target is quickly divided into non-overlapping and overlapping regions. For simple regions without overlap, the morphological features are directly extracted by the rapid judgment module for judgment. Only complex overlapping regions are activated with a computationally intensive intelligent reconstruction process. The hierarchical processing mechanism greatly optimizes the allocation of computing resources and fully meets the high timeliness requirements of online detection.

[0041] Furthermore, this invention employs an intelligent reconstruction module that integrates three-dimensional depth information with a variety-specific morphological model, solving the problem of accurate detection of overlapping rice grains. The intelligent reconstruction module first obtains a depth layering coefficient by calculating the standard deviation of the depth values ​​set of the overlapping area, intelligently identifying the overlapping type as vertical stacking, coplanar adhesion, or slight intersection. Then, for rice grains with incomplete outlines, it extracts the direction vector and curvature features of their visible ends by fitting a straight line using the least squares method. Based on a pre-stored morphological feature model of silk rice containing typical length ranges, aspect ratio ranges, and width variation patterns, it uses spline curve fitting to generate a virtual outline that conforms to the real shape. Finally, the overall aspect ratio is verified to ensure the reliability of the reconstruction, achieving rapid restoration of the complete shape of obscured rice grains. This solves the problem of insufficient detection accuracy of traditional methods for overlapping areas, significantly improving detection accuracy and effectively reducing manual detection costs.

[0042] Furthermore, this invention achieves closed-loop monitoring of system performance by periodically introducing a standard sample set and calculating the detection performance drift coefficient through an optimized calibration module. When the detection performance drift coefficient exceeds the preset drift coefficient, the module can automatically diagnose the source of performance degradation: if the relative deviation of the complete morphology index is larger, the exposure and current are adjusted according to the drift coefficient ratio to optimize image acquisition parameters; if the relative deviation of the virtual complete morphology index is larger, the parameters of the intelligent reconstruction model are adjusted within the typical aspect ratio fluctuation range of 1% to 3%, ensuring that the system can maintain stable and reliable detection accuracy over a long period of time, reducing manual maintenance costs, and improving online detection efficiency. Attached Figure Description

[0043] Figure 1 This is a block diagram showing the module functional connection of the online detection device for the integrity of silk rice based on machine vision, according to an embodiment of the present invention.

[0044] Figure 2 This is a schematic diagram of the online integrity detection device for silk rice based on machine vision according to an embodiment of the present invention;

[0045] Figure 3 This is a logic block diagram of an embodiment of the present invention for determining whether the current region is an overlapping region or a non-overlapping region based on the region overlap probability index;

[0046] Figure 4 This is a logic block diagram illustrating how the integrity of non-overlapping areas of the silk rice is determined based on the integrity morphology index in an embodiment of the present invention.

[0047] Figure 5 This is a logic block diagram illustrating how the upper and lower layer relationships of overlapping region glutinous rice are determined based on a depth layering coefficient according to an embodiment of the present invention.

[0048] Figure 6This is a logic block diagram illustrating how the integrity of a single virtual silk rice grain in an overlapping region is determined based on a virtual completeness index, according to an embodiment of the present invention.

[0049] Figure 7 This is a logic diagram of an embodiment of the present invention for determining specific optimization targets and performing optimization based on the relative deviation of the complete form index and the relative deviation of the virtual complete form index;

[0050] In the diagram, 1 - rice storage bin, 2 - discharge port, 3 - two-dimensional color camera, 4 - three-dimensional depth camera, 5 - parallel light source, 6 - conveyor belt, and 7 - drive roller. Detailed Implementation

[0051] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0052] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0053] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0054] Please see Figures 1 to 2 As shown, Figure 1 This is a block diagram showing the module functional connection of the online detection device for the integrity of silk rice based on machine vision, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the online integrity detection device for high-quality rice based on machine vision, according to an embodiment of the present invention.

[0055] This invention relates to a machine vision-based online detection device for the integrity of high-quality rice, comprising:

[0056] The transmission mechanism includes a conveyor belt 6 for conveying the rice to be tested, and a drive roller 7 disposed at the end of the conveyor belt 6 for driving the conveyor belt 6 to convey.

[0057] The rice storage assembly includes a rice storage bin 1 disposed at the upper part of one end of the conveyor belt 6 for storing the rice to be tested, and a discharge port 2 disposed at the lower part of the rice storage bin 1 and fixedly connected to the rice storage bin 1 for discharging the rice to be tested.

[0058] The detection assembly includes a two-dimensional color camera 3 and a three-dimensional depth camera 4 disposed on the upper part of the conveyor belt 6 to acquire image data of the rice to be detected, and a parallel light source 5 disposed in the middle of the conveyor belt 6 to illuminate the rice to be detected.

[0059] A control execution unit, connected to the detection component, includes,

[0060] The image acquisition module is used to acquire two-dimensional and three-dimensional image data of the rice to be detected.

[0061] An image processing module, which is connected to the image acquisition module, is used to perform grayscale conversion, filtering and noise reduction, and background segmentation on the acquired image data;

[0062] A real-time control module, which is connected to the image processing module and the detection component respectively, is used to receive the processed image and calculate the comprehensive image quality coefficient; when the comprehensive image quality coefficient does not meet the preset value, it determines to adjust the parameters of the camera or light source.

[0063] An overlap recognition module, which is connected to the image processing module, is used to perform connected region analysis on the processed image data and obtain the region overlap probability index based on the morphological features of several connected regions to determine the overlap type of the current region.

[0064] A rapid determination module, which is connected to the overlap recognition module, is used to directly extract morphological features of the non-overlapping area of ​​the silk rice under the condition that the current area is determined to be a non-overlapping area, and obtain the integrity morphology index to determine the integrity of the silk rice in the non-overlapping area.

[0065] The intelligent reconstruction module, which is connected to the overlap recognition module, is used to combine three-dimensional image data to obtain the depth layering coefficient to determine the upper and lower layer relationship of the overlapping area of ​​the glutinous rice, and directly identify and analyze the end feature parameters of the visible part of each glutinous rice to generate the virtual outline of the occluded part.

[0066] An overlap determination module, which is connected to the intelligent reconstruction module, is used to calculate the virtual integrity morphology index of a single virtual rice grain to determine the integrity of a single virtual rice grain in the overlapping area, under the condition that the single reconstruction is reliable.

[0067] An optimization calibration module, which is connected to the rapid determination module and the overlap determination module, is used to calculate the detection performance drift coefficient based on the complete morphology index and the virtual complete morphology index to determine whether the detection component and the intelligent reconstruction module need to be optimized, and to determine the corresponding optimization strategy to be executed based on the comparison results of the relative deviation of the complete morphology index and the relative deviation of the virtual complete morphology index when optimization is required.

[0068] The result output module is connected to the fast determination module and the overlap determination module respectively, and is used to receive, integrate and output the final detection result.

[0069] Specifically, this invention integrates two-dimensional and three-dimensional visual information and proposes a virtual contour reconstruction algorithm to intelligently analyze the spatial structure of overlapping areas. For lower or occluded rice grains, the system automatically generates a virtual contour of the occluded portion based on the end-point vector and curvature features, combined with a pre-stored model of the morphological characteristics of long rice grains. This achieves accurate restoration of the complete rice grain shape, fundamentally overcoming misjudgments caused by overlapping long rice grains and significantly improving the detection accuracy of overlapping areas.

[0070] In this embodiment of the invention, under the condition that the conveyor belt speed is stable at 0.5 m / s and the trigger sensor detects the rice grain stream entering the field of view, the image acquisition module simultaneously triggers the two-dimensional color camera and the three-dimensional depth camera to capture images, thereby acquiring two-dimensional RGB images and three-dimensional point cloud data of the rice to be detected. Subsequently, the acquired two-dimensional image and the three-dimensional image data packet with attached depth information are passed to the image processing module for preprocessing. First, the two-dimensional RGB image is converted into a grayscale image; then, a Gaussian filtering algorithm is used to filter and denoise the grayscale image; finally, an Otsu adaptive threshold segmentation algorithm is used to binarize the filtered image and remove small noise points, thereby completing background segmentation and obtaining a clean binary image containing only rice grains in the foreground. After processing, the clean binary image is passed to the control module.

[0071] Specifically, the real-time control module receives the preprocessed image and calculates the comprehensive image quality coefficient. Based on the comparison between the comprehensive image quality coefficient and a preset comprehensive image quality coefficient, it determines and adjusts the operating parameters of the detection component.

[0072] If the overall image quality coefficient is less than the preset overall image quality coefficient, then the operating parameters of the detection component are adjusted.

[0073] If the overall image quality coefficient is greater than or equal to the preset overall image quality coefficient, then the image processed by the image processing module is determined to meet the detection requirements.

[0074] In this embodiment of the invention, the comprehensive image quality coefficient is obtained by weighted sum of the Laplacian variance of the grayscale image and the foreground-background contrast of the clean binary image. In the actual calculation process, the Laplacian variance is first obtained by convolving the grayscale image with the Laplacian operator. The foreground-background contrast is obtained by calculating the absolute difference between the average grayscale value of the foreground region and the average grayscale value of the background region in the clean binary image. Then, a first weight coefficient is assigned to the Laplacian variance and a second weight coefficient is assigned to the foreground-background contrast. The weighted sum is then calculated to obtain the comprehensive image quality coefficient.

[0075] In this embodiment of the invention, the preset comprehensive image quality coefficient ranges from 50 to 70, and the preferred value is 60. The preferred range and preferred value can be determined according to the actual situation, and are not specifically limited here.

[0076] In this embodiment of the invention, adjusting the operating parameters of the detection component includes increasing the driving current of the parallel light source in the detection component by 10% based on the basic driving current and increasing the exposure time of the two-dimensional color camera in the detection component by 5ms based on the basic exposure time. After adjustment, image data is reacquired and a new comprehensive image quality coefficient is recalculated.

[0077] The baseline driving current and the baseline exposure time are determined by performing an initialization calibration process when the system is first enabled. Specifically, the parameters are scanned within the allowable exposure time range of the two-dimensional color camera and the allowable driving current range of the parallel light source to find the exposure time and driving current that make the comprehensive image quality coefficient reach its maximum value, and recorded as the baseline exposure time and the baseline driving current, respectively.

[0078] Specifically, this invention uses a real-time control module to calculate a comprehensive image quality coefficient to evaluate the acquired images and compare them with a preset comprehensive image quality coefficient. When the quality is substandard, the driving current of the parallel light source is automatically increased by 10%, and the exposure time of the two-dimensional color camera is increased by 5ms to ensure that the image data input to subsequent algorithms is always clear and stable, thus guaranteeing the reliability of the overall detection process from the source.

[0079] Please see Figure 3 As shown, it is a logic block diagram of an embodiment of the present invention for determining whether the current region is an overlapping region or a non-overlapping region based on the region overlap probability index.

[0080] Specifically, the real-time control module, after determining that the image processed by the image processing module meets the detection requirements, performs connected component analysis on the image data processed by the overlap recognition module and calculates the region overlap probability index. Based on the comparison result between the region overlap probability index and a preset region overlap probability index, it determines whether the current region is an overlapping region or a non-overlapping region.

[0081] If the region overlap probability index is less than or equal to the preset region overlap probability index, then the current region is determined to be a non-overlapping region.

[0082] If the region overlap probability index is greater than the preset region overlap probability index, then the current region is determined to be an overlapping region.

[0083] In this embodiment of the invention, the connected region analysis involves traversing the pure binary image using a two-pass 8-neighborhood method, marking interconnected white foreground pixels as the same region, and assigning a unique label to each independent region.

[0084] In this embodiment of the invention, the region overlap probability index is obtained by calculating the ratio of the pixel area of ​​the connected region to the pixel area of ​​the minimum bounding rectangle of the connected region. The specific calculation method is as follows: First, the actual pixel area of ​​the connected region is obtained by counting the total number of pixels under the label of the connected region; then, the minimum bounding rectangle of the connected region is calculated, and the pixel area of ​​the minimum bounding rectangle is obtained; finally, the region overlap probability index is obtained by calculating the ratio of the actual pixel area to the pixel area of ​​the minimum bounding rectangle.

[0085] In this embodiment of the invention, the value range of the preset region overlap probability index is 0.6-0.8, and the preferred value is 0.7. The preferred value range and the preferred value can be determined according to the actual situation, and no specific limitation is made here.

[0086] Specifically, this invention employs an overlap recognition module to intelligently divide the jasmine rice region. By calculating the region overlap probability index and comparing it with a preset region overlap probability index, the detection target is quickly divided into non-overlapping and overlapping regions. For simple regions without overlap, the rapid judgment module directly extracts morphological features for judgment. Only complex overlapping regions are activated with a computationally intensive intelligent reconstruction process. The hierarchical processing mechanism greatly optimizes the allocation of computing resources and fully meets the high timeliness requirements of online detection.

[0087] Please see Figure 4 As shown, it is a logic block diagram of determining the integrity of non-overlapping areas of glutinous rice according to the integrity morphology index in an embodiment of the invention.

[0088] Specifically, when the overlap recognition module determines that the current region is a non-overlapping region, the rapid determination module directly extracts morphological features from the non-overlapping region's silky rice and calculates the integrity index. Based on the comparison between the integrity index and a preset integrity index, the integrity of the silky rice in the non-overlapping region is determined.

[0089] If the complete form index is less than the preset complete form index, then the rice in the non-overlapping area is determined to be incomplete, and defective rice is output.

[0090] If the complete morphology index is greater than or equal to the preset complete morphology index, then the rice in the non-overlapping region is determined to be complete.

[0091] In this embodiment of the invention, the morphological features include the pixel length and pixel area of ​​the connected region. The pixel length is obtained by calculating the length of the longer side of the smallest bounding rectangle of the connected region; the pixel area is obtained by counting the total number of pixels under the label of the connected region.

[0092] In this embodiment of the invention, the complete morphology index is obtained by taking the smaller of the ratio of the pixel length of the connected region to the pixel length of a standard complete rice grain, and the ratio of the pixel area of ​​the connected region to the pixel area of ​​a standard complete rice grain. Specifically, the calculation method is as follows: first, the pixel length of the connected region is divided by the pre-stored pixel length of a standard complete rice grain to obtain the length ratio; second, the pixel area of ​​the connected region is divided by the pre-stored pixel area of ​​a standard complete rice grain to obtain the area ratio; finally, the smaller value between the prime area ratio and the length ratio is taken as the complete morphology index.

[0093] In this embodiment of the invention, the preset complete morphology index is the average value of the complete morphology index of a specific batch of known complete rice samples during the system initialization phase. The value range is 0.7-0.8, and the preferred value in this invention is 0.75. The preferred value range and preferred value can be determined according to the actual situation, and are not specifically limited here.

[0094] In this embodiment of the invention, the standard complete rice grain pixel length and the standard complete rice grain pixel area are obtained by acquiring and measuring images of a large number of known complete Silky Rice samples and taking their statistical average values.

[0095] Please see Figure 5 As shown, it is a logic block diagram of determining the upper and lower layer relationships of overlapping area silk rice according to the depth layering coefficient in an embodiment of the present invention.

[0096] Specifically, when the overlap recognition module determines that the current region is an overlapping region, the intelligent reconstruction module calculates a depth layering coefficient by combining the three-dimensional image data of the overlapping region. Based on the comparison between the depth layering coefficient and a preset depth layering coefficient, the upper and lower layer relationships of the overlapping region's silk rice are determined.

[0097] If the depth layering coefficient is greater than the preset depth layering coefficient, it is determined that the overlapping area has a clear upper and lower layering relationship;

[0098] If the depth layering coefficient is less than or equal to the preset depth layering coefficient, then the overlapping areas are determined to be adhesion or slight intersection within the same plane.

[0099] In this embodiment of the invention, the calculation process of the depth layering coefficient is as follows: based on the calibration parameters, the two-dimensional pixel coordinates of the overlapping region to be reconstructed output by the overlap recognition module are accurately registered with the three-dimensional point cloud data; the depth values ​​of all three-dimensional spatial points located in the two-dimensional region are extracted from the registered data to form a depth value set [Z1, Z2, ..., Zn], and finally the standard deviation of the depth value set is calculated to obtain the depth layering coefficient.

[0100] It is understood that the calibration parameters are camera parameters obtained through the calibration process during the system initialization phase. These parameters are collectively referred to as calibration parameters. The calibration parameters include, but are not limited to, the internal parameters of the two-dimensional color camera and the three-dimensional depth camera, such as focal length, principal point, and the external spatial position transformation relationship between the two-dimensional color camera and the three-dimensional depth camera.

[0101] In this embodiment of the invention, the precise registration process is as follows: A unified coordinate system is constructed using the external spatial position transformation relationship between the two-dimensional color camera and the three-dimensional depth camera. For each two-dimensional pixel point within the overlapping area to be reconstructed, the camera imaging model is used to map it to the spatial coordinate system where the three-dimensional point cloud data resides, thereby establishing a one-to-one mapping relationship between the two-dimensional pixel points and the three-dimensional spatial points, thus completing the registration.

[0102] In this embodiment of the invention, the preset depth layering coefficient ranges from 0.05mm to 0.15mm, and the preferred value is 0.1mm. The preferred range and preferred value can be determined according to the actual situation, and are not specifically limited here.

[0103] In this embodiment of the invention, for the overlapping areas of *Syzygium stenoptera* with a clear upper and lower layer relationship, based on the set of depth values, a clustering algorithm, such as K-means clustering, is used to divide the pixels into upper and lower layers. The set of pixels with larger depth values ​​is identified as the visible part of the lower layer of *Syzygium stenoptera*. For the visible part of the lower layer of *Syzygium stenoptera*, the unobstructed end position is located using an endpoint detection algorithm. Then, the direction vector of the contour at the end position is extracted. That is, several points on the contour are selected near the end position, and a straight line is fitted using the least squares method. The direction vector of this straight line is defined as the direction vector, and the radius of curvature of the contour line at the end point is calculated to obtain the curvature feature. The smaller the radius of curvature, the sharper the end; the larger the radius of curvature, the more rounded the end. Contour fitting is performed based on a pre-stored *Syzygium stenoptera* morphological feature model.

[0104] In this embodiment of the invention, for the overlapping areas of rice grains that are adhered or slightly intersecting in the same plane, the visible portion of each rice grain to be processed within the connected area is directly located using an endpoint detection algorithm to pinpoint its unobstructed end position. Then, the direction vector of the contour at the end position is extracted. That is, several points on the contour are selected near the end position, and a straight line is fitted using the least squares method. The direction vector of this straight line is defined as the direction vector, and the radius of curvature of the contour line at the end point is calculated to obtain the curvature feature. The smaller the radius of curvature, the sharper the end; the larger the radius of curvature, the more rounded the end. Contour fitting is performed based on a pre-stored rice grain morphology feature model.

[0105] In this embodiment of the invention, the morphological characteristic model of the silk rice is a digital representation model established by systematically measuring and statistically analyzing a large number of standard samples. The core characteristic parameters mainly include the typical length fluctuation range of the silk rice, that is, the length of most complete silk rice grains is distributed within the typical length fluctuation range; secondly, the typical length-to-width ratio fluctuation range, which characterizes the slenderness of the overall shape of the silk rice; and finally, the characteristic law of the change in the width of the rice grain outline, which characterizes the symmetrical change trend of the width of the silk rice grain as it transitions from both ends to the middle, first gradually increasing to the maximum value and then gradually decreasing, and the symmetrical change trend conforms to the characteristics of a smooth curve.

[0106] Specifically, the fitting method for reconstructing the virtual contour of any occluded part of the jasmine rice is as follows: First, take the identified end position of the jasmine rice as the starting reference point, and draw an initial contour trajectory line along the spatial direction indicated by the contour direction of the starting reference point towards the area occluded by other jasmine rice. Then, adjust the sharpness or roundness of the end of the initial trajectory line according to the extracted curvature features to make it consistent with the end shape of the real jasmine rice. Subsequently, the contour optimization stage based on the jasmine rice morphological feature model is entered.

[0107] In this embodiment of the invention, the contour optimization stage includes length constraints, width and shape constraints, contour generation and splicing, and overall verification. The length constraint involves calculating the total length of the current virtual contour in real time during the extension of the initial trajectory line along the direction vector. If the upper limit of the typical length fluctuation range is reached and no other contour is encountered, the extension stops, and the current endpoint is taken as the endpoint of the virtual contour. The width and shape constraint involves assigning a reasonable width contour to the virtual contour based on the characteristic law of the width change of the grain contour. Specifically, several key points are defined on the virtual contour line, and a target width value is assigned to each key point. The target width value is close to the maximum value in the width variation law in the middle of the contour and gradually decreases towards both ends, thereby simulating the real shape of the silk rice. The contour generation and splicing are constrained by the key points and the target width, and a spline curve fitting algorithm, such as cubic B-spline curve fitting, is used to generate a smooth final virtual contour line that conforms to the width variation law. Finally, the actual detected visible contour and the generated final virtual contour line are smoothly connected at the endpoints to form a complete and closed virtual silk rice contour. The overall verification is to calculate the aspect ratio of the virtual silk rice contour and ensure that it is within the typical aspect ratio fluctuation range, which serves as the final verification of the reconstruction rationality. The complete virtual silk rice contour within the typical aspect ratio fluctuation range is passed to the overlap determination module for final integrity detection.

[0108] Specifically, this invention employs an intelligent reconstruction module that integrates three-dimensional depth information with a variety-specific morphological model to solve the problem of accurate detection of overlapping rice grains. The intelligent reconstruction module first obtains a depth layering coefficient by calculating the standard deviation of the depth values ​​set of the overlapping area, intelligently identifying the overlapping type as vertical stacking, coplanar adhesion, or slight intersection. Then, for rice grains with incomplete outlines, it extracts the direction vector and curvature features of their visible ends by fitting a straight line using the least squares method. Based on a pre-stored morphological feature model of silky rice containing typical length ranges, aspect ratio ranges, and width variation patterns, it uses spline curve fitting to generate a virtual outline that conforms to the real shape. Finally, the overall aspect ratio is verified to ensure the reliability of the reconstruction, achieving rapid restoration of the complete shape of obscured rice grains. This solves the problem of insufficient detection accuracy of traditional methods for overlapping areas, significantly improving detection accuracy and effectively reducing manual detection costs.

[0109] Specifically, the virtual contour reconstruction algorithm of this invention deeply integrates the unique morphological prior knowledge of Silky Rice. The key parameters in the morphological feature model of Silky Rice, such as the typical length range, aspect ratio range, and width variation law, are all derived from a large amount of statistical analysis of Silky Rice. This makes the virtual contour reconstruction and integrity judgment more targeted, and significantly improves the recognition accuracy and adaptability of this variety compared with the general rice detection model.

[0110] Please see Figure 6 As shown, it is a logic block diagram of determining the integrity of a single virtual silk rice in the overlapping area according to the virtual integrity morphology index in an embodiment of the present invention.

[0111] Specifically, after the intelligent reconstruction module reconstructs the virtual outline of the occluded part and the overall verification determines that the reconstruction is reasonable, the overlap determination module calculates the virtual integrity morphology index of a single virtual rice grain outline. Based on the comparison between the virtual integrity morphology index and a preset virtual integrity morphology index, the integrity of a single virtual rice grain in the overlapping area is determined.

[0112] If the virtual complete form index is less than the preset virtual complete form index, then it is determined that a single virtual silk rice in the overlapping area is incomplete, and defective rice is output.

[0113] If the virtual complete form index is greater than or equal to the preset virtual complete form index, then the single virtual silk rice in the overlapping area is determined to be complete.

[0114] In this embodiment of the invention, the virtual complete form index is obtained by taking the smaller of the ratio of the pixel length of the single virtual silk rice outline to the pixel length of the standard complete silk rice, and the ratio of the pixel area of ​​the single virtual silk rice outline to the pixel area of ​​the standard complete silk rice. Specifically, the calculation method is as follows: first, the pixel length of the single virtual silk rice outline is divided by the pre-stored pixel length of the standard complete silk rice to obtain the virtual length ratio; second, the pixel area of ​​the single virtual silk rice outline is divided by the pre-stored pixel area of ​​the standard complete silk rice to obtain the virtual area ratio; finally, the smaller value between the virtual area ratio and the virtual length ratio is taken as the virtual complete form index.

[0115] In this embodiment of the invention, the preset virtual complete form index is the average value of the virtual complete form index of the successfully reconstructed virtual rice grains. For example, the average value of the virtual complete form index of 10,000 successfully reconstructed virtual rice grains is taken as the preset virtual complete form index, with a value range of 0.65-0.75. The preferred value in this invention is 0.7. The preferred value range and preferred value can be determined according to the actual situation, and are not specifically limited here.

[0116] Specifically, under the conditions that the rapid determination module and the overlap determination module execute the corresponding working parameters, the optimization calibration module calculates the detection performance drift coefficient based on the integrity morphology index and the virtual filament integrity morphology index, and determines whether the current detection state meets the standard based on the comparison result of the detection performance drift coefficient and the preset detection performance drift coefficient.

[0117] If the detection performance drift coefficient is less than or equal to the preset detection performance drift coefficient, then the current detection status is determined to meet the standard.

[0118] If the detection performance drift coefficient is greater than the preset detection performance drift coefficient, the current detection state is determined to be substandard, and the relative deviation value of the complete morphology index and the relative deviation value of the virtual complete morphology index are calculated.

[0119] In this embodiment of the invention, the detection performance drift coefficient is obtained by the relative deviation of the judgment accuracy of the standard sample set. Specifically, the calculation method is as follows: First, without affecting production or stopping the detection device, the standard sample set is imported into the detection process. The standard sample set consists of a fixed number of complete meters and defective meters that have been precisely sorted manually and marked with their true integrity status. After online detection of the sample set, the number of correctly judged samples is counted by comparing the detection results with the true state. Then, the ratio of the number of correctly judged samples to the total number of the standard sample set is calculated to obtain the judgment accuracy of the current detection device. Next, the first absolute difference between the judgment accuracy of the current device and the benchmark accuracy established and stored during the initial calibration stage is calculated. Finally, the ratio of the absolute difference to the benchmark accuracy is calculated to obtain the detection performance drift coefficient.

[0120] In this embodiment of the invention, the preset detection performance drift coefficient ranges from 0.03 to 0.06, and the preferred value is 0.05. The preferred range and preferred value can be determined according to the actual situation, and are not specifically limited here.

[0121] In this embodiment of the invention, the relative deviation value of the complete morphology index is obtained by first calculating the second absolute difference between the average complete morphology index in the current detection cycle and the benchmark average value established and stored in the initial calibration stage, and then calculating the ratio of the absolute difference to the benchmark average value.

[0122] In this embodiment of the invention, the relative deviation value of the virtual complete form index is obtained by first calculating the third absolute difference between the average value of the virtual complete form index in the current detection period and the benchmark average value established and stored in the initial calibration stage, and then calculating the ratio of the absolute difference to the benchmark average value.

[0123] The detection cycle is 8 hours.

[0124] Please see Figure 7 As shown, it is a logic block diagram of an embodiment of the present invention that determines the specific optimization target and performs optimization based on the relative deviation of the complete form index and the relative deviation of the virtual complete form index.

[0125] Specifically, when optimization is deemed necessary, the optimization calibration module determines the specific optimization target and performs optimization based on the comparison result between the relative deviation of the complete form index and the relative deviation of the virtual complete form index.

[0126] If the relative deviation of the complete form index is greater than the relative deviation of the virtual complete form index, then it is determined that the image acquisition module needs to be optimized.

[0127] If the relative deviation of the complete form index is less than the relative deviation of the virtual complete form index, then it is determined that the intelligent reconstruction module should be optimized.

[0128] In this embodiment of the invention, the optimization of the image acquisition module is to make a proportional adjustment based on the detection performance drift coefficient. That is, for every 0.01 higher than the preset detection performance drift coefficient, the exposure time of the two-dimensional color camera is extended by 1% and the driving current of the parallel light source is increased by 1%. The cumulative extension of the exposure time should not exceed 20% of the basic exposure time, and the cumulative increase of the driving current should not exceed 15% of the basic driving current.

[0129] In this embodiment of the invention, the optimization of the intelligent reconstruction module is based on long-term statistical virtual silk rice outline data, and the lower limit and upper limit of the typical aspect ratio fluctuation range are brought closer to the actual mode of the current statistical distribution. The adjustment range is positively correlated with the detection performance drift coefficient, and the adjustment range is 1% to 3% of the original range, preferably 2%. The total cumulative adjustment range shall not exceed 10% of the initial range.

[0130] Specifically, the result output module receives parallel judgment results from the fast judgment module and the overlapping judgment module in real time and generates a unified format detection record for each processed rice grain. The detection record includes at least a unique target identifier, integrity status, and a timestamp of the result. The integrated detection results are then output to the outside in the form of data packets through a standard industrial communication interface. The specific output format includes real-time identification of defective rice grains using high-speed I / O interfaces with level pulse signals. For example, a high-level pulse represents the identification of a defective rice grain.

[0131] Specifically, this invention achieves closed-loop monitoring of system performance by periodically introducing a standard sample set and calculating the detection performance drift coefficient through an optimized calibration module. When the detection performance drift coefficient exceeds the preset drift coefficient, the module can automatically diagnose the source of performance degradation: if the relative deviation of the complete morphology index is larger, the exposure and current are adjusted according to the drift coefficient ratio to optimize image acquisition parameters; if the relative deviation of the virtual complete morphology index is larger, the parameters of the intelligent reconstruction model are adjusted within the typical aspect ratio fluctuation range of 1% to 3%, ensuring that the system can maintain stable and reliable detection accuracy over a long period of time, reducing manual maintenance costs, and improving online detection efficiency.

[0132] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A machine vision-based online detection device for the integrity of high-quality rice, characterized in that, include, A transmission assembly used to transport the rice grains to be tested for integrity detection; The detection assembly includes a two-dimensional color camera and a three-dimensional depth camera disposed on the upper part of the conveyor belt to acquire image data of the rice to be detected, and a parallel light source disposed in the middle part of the conveyor belt corresponding to the two-dimensional color camera and the three-dimensional depth camera to illuminate the rice to be detected. The overlap identification module is used to obtain the region overlap probability index based on the morphological features of several connected regions to determine the overlap type of the current region; The rapid determination module is used to directly extract morphological features from non-overlapping areas of the silky rice and obtain the integrity morphology index to determine the integrity of the silky rice in non-overlapping areas. The intelligent reconstruction module is used to combine three-dimensional image data to obtain the depth layering coefficient to determine the upper and lower layer relationship of the overlapping area of ​​the glutinous rice, and to identify and analyze the end feature parameters of the visible part of each glutinous rice to generate the virtual outline of the occluded part. The overlap determination module is used to calculate the virtual integrity morphology index of a single virtual rice grain under reasonable reconstruction conditions to determine the integrity of a single virtual rice grain in the overlapping area. The optimization calibration module is used to calculate the detection performance drift coefficient based on the complete morphology index and the virtual complete morphology index to determine whether the current detection state meets the standard. If it does not meet the standard, it determines to extend the exposure time of the two-dimensional color camera, increase the driving current of the parallel light source, and move the lower and upper limits of the typical aspect ratio fluctuation range closer to the actual mode of the current statistical distribution based on the relative deviation of the complete morphology index and the relative deviation of the virtual complete morphology index.

2. The online detection device for the integrity of high-quality rice based on machine vision according to claim 1, characterized in that, It also includes an image acquisition module, which is used to acquire two-dimensional and three-dimensional image data of the rice to be detected; The image processing module is used to perform grayscale conversion, filtering and noise reduction, and background segmentation on the acquired image data; The real-time control module receives the processed image and calculates the overall image quality coefficient to determine the parameters for adjusting the camera and light source; The results output module is used to receive, integrate, and output the final detection results.

3. The online detection device for the integrity of silk rice based on machine vision according to claim 2, characterized in that, The real-time control module, in response to the overall image quality coefficient being less than a preset overall image quality coefficient, determines to increase the driving current of the parallel light source and extend the exposure time of the two-dimensional color camera.

4. The online detection device for the integrity of high-quality rice based on machine vision according to claim 2, characterized in that, The overlap recognition module responds to the real-time control module by determining that the image processed by the image processing module meets the detection requirements, and determines whether the current region is an overlapping region or a non-overlapping region based on the comparison result between the region overlap probability index and the preset region overlap probability index. Based on the fact that the region overlap probability index is less than or equal to the preset region overlap probability index, the current region is determined to be a non-overlapping region. Based on the fact that the region overlap probability index is greater than the preset region overlap probability index, the current region is determined to be an overlapping region.

5. The online detection device for the integrity of high-quality rice based on machine vision according to claim 1, characterized in that, The rapid determination module responds to the overlap recognition module in determining that the current region is a non-overlapping region, and determines the integrity of the silk rice in the non-overlapping region based on the comparison result of the integrity morphology index and the preset integrity morphology index. Based on the fact that the complete morphology index is less than the preset complete morphology index, it is determined that the rice in the non-overlapping area is incomplete, and defective rice is output.

6. The online detection device for the integrity of high-quality rice based on machine vision according to claim 1, characterized in that, The intelligent reconstruction module responds to the overlap recognition module determining that the current region is an overlapping region, and determines the upper and lower layer relationship of the overlapping region's silk rice based on the comparison result of the depth layering coefficient and the preset depth layering coefficient, wherein, Based on the fact that the depth layering coefficient is greater than the preset depth layering coefficient, it is determined that the overlapping area has a clear upper and lower layering relationship; Based on the fact that the depth layering coefficient is less than or equal to the preset depth layering coefficient, the overlapping areas are determined to be adhesions or slight intersections within the same plane.

7. The online detection device for the integrity of high-quality rice based on machine vision according to claim 6, characterized in that, The intelligent reconstruction module divides the pixels into upper and lower layers based on the clear upper and lower layer relationship of the overlapping area of ​​the rice. It identifies the set of pixels with larger depth values ​​as the visible part of the lower layer of rice and locates the un-occluded end position, generating a virtual outline of the occluded part. The intelligent reconstruction module, based on the overlapping area being the sticky or slightly intersecting silk rice in the same plane, directly uses an endpoint detection algorithm to locate the unobstructed end position of the visible part of each silk rice to be processed in several connected areas, and generates a virtual outline of the obstructed part.

8. The online detection device for the integrity of high-quality rice based on machine vision according to claim 1, characterized in that, The overlap determination module responds to the intelligent reconstruction module's reconstruction of the virtual outline of the occluded portion, which is verified as reasonable. Based on this overall verification, the module determines the integrity of individual virtual rice grains within the overlapping area according to the comparison between the virtual completeness index and a preset virtual completeness index. Based on the fact that the virtual completeness index is less than the preset virtual completeness index, it is determined that a single virtual silk rice in the overlapping area is incomplete, and defective rice is output. Based on the virtual completeness index being greater than or equal to the preset virtual completeness index, the integrity of a single virtual silk rice in the overlapping area is determined.

9. The online detection device for the integrity of high-quality rice based on machine vision according to claim 1, characterized in that, The optimization calibration module determines that the current detection state is substandard based on the detection performance drift coefficient being greater than the preset detection performance drift coefficient, and determines to extend the exposure time of the two-dimensional color camera and increase the driving current of the parallel light source based on the relative deviation of the complete morphology index being greater than the relative deviation of the virtual complete morphology index. The adjustment range is positively correlated with the detection performance drift coefficient.

10. The online detection device for the integrity of high-quality rice based on machine vision according to claim 9, characterized in that, The optimization calibration module determines that the current detection state is substandard based on the detection performance drift coefficient being greater than the preset detection performance drift coefficient. It also determines that the lower and upper limits of the typical aspect ratio fluctuation range should be moved closer to the actual mode of the current statistical distribution based on the fact that the relative deviation of the complete morphology index is less than the relative deviation of the virtual complete morphology index. The adjustment range is positively correlated with the detection performance drift coefficient.

Images