A dynamic grading of different color particle rice reverse selection control system and method

Abstract

The application relates to the technical field of rice color selection control, in particular to a dynamic grading abnormal-color-rice reverse selection control system and method, which comprises the following steps: making a single rice grain fall by using a vibration motor and a guide rail and collecting an image by using an industrial camera; performing local contrast adjustment and dynamic range adjustment on the image based on a color difference enhancement model of a generative adversarial network; establishing a rice standard chroma library, identifying abnormal-color-rice by using a dynamic reference chromatogram bidirectional recognition algorithm and a relative color difference mapping strategy, and constructing an abnormal-color-rice feature reverse index mechanism for bidirectional recognition; according to the recognition result, controlling an air jet nozzle to blow the abnormal-color-rice into an abnormal-color-rice container; after separating the current batch, the rice in the abnormal-color-rice container is re-identified and separated to separate normal rice from the abnormal-color-rice. The method can realize adaptive bidirectional identification and accurate separation of rice color selection.

Description

Technical Field

[0001] This application relates to the field of rice color sorting control technology, and specifically discloses a dynamic grading system and method for reverse sorting of discolored rice grains. Background Technology

[0002] In rice processing, it is necessary to separate and remove discolored grains to improve the quality and grade of the rice. Currently, traditional rice color sorting methods mainly include photoelectric colorimetric sorting and image recognition sorting, but these methods have many problems and shortcomings in practical applications.

[0003] Traditional rice color sorting equipment typically uses a one-way discrimination method, meaning it can only identify rice grains with different colors, but cannot effectively distinguish normal rice mixed in with these grains. This results in a large amount of normally sorted rice being mistakenly separated and thus wasted. Furthermore, existing color sorting technologies lack the ability to adaptively adjust to fluctuations in the quality of different batches of rice, making it difficult to dynamically adapt to changes in raw material quality.

[0004] Furthermore, in actual rice color sorting processes, the high speed and density of the falling rice make precise identification and separation difficult. Commonly used photoelectric colorimetry is easily affected by lighting conditions and rice posture, and it cannot identify and separate rice grains with the same light transmittance but different colors. Image recognition methods, on the other hand, require high-quality images, and their recognition speed and accuracy need improvement.

[0005] Therefore, there is an urgent need for a rice color sorting method that can achieve adaptive separation control and reverse selection of discolored grains in order to improve the rice yield and quality, reduce losses and waste, and adapt to the quality fluctuations of different batches of raw materials.

[0006] In view of this, this application proposes a dynamic grading system and method for reverse selection of discolored rice grains. Summary of the Invention

[0007] To achieve the above objectives, this application provides a dynamic grading system and method for reverse selection of discolored rice grains, the specific technical solution of which is as follows:

[0008] A method for controlling the reverse selection of discolored rice grains through dynamic grading includes:

[0009] The rice on the conveyor belt is vibrated by a vibrating motor, and a guide rail is set at the end of the conveyor belt to make individual grains of rice detach from the conveyor belt and fall.

[0010] The trajectory of falling rice is fitted to a trajectory surface, and the collection area, recognition area and separation area are divided according to the trajectory surface. An industrial camera is used to continuously collect images of falling rice in the collection area.

[0011] A color difference enhancement model based on generative adversarial networks is used to adjust the local contrast of the collected images of falling rice, and to adjust the dynamic range of the collected images of falling rice in both RGB and HSI domains.

[0012] A standard color library for rice was established. A dynamic reference chromatographic bidirectional identification algorithm and a relative color difference mapping strategy were used to identify rice grains with different colors and to perform calibration within the identification area. At the same time, a reverse indexing mechanism for the characteristics of rice grains with different colors was constructed to perform bidirectional identification of rice grains with different colors and normal rice.

[0013] The trajectory of the discolored rice grains in the separation zone is calculated based on the calibration results, and the timing of the opening of the airflow nozzles is controlled to blow the discolored rice grains into the discolored grain container.

[0014] After separating the discolored rice grains from the current batch, the rice in the discolored grain container is returned to the conveyor belt inlet. The vibration parameters applied to the conveyor belt by the vibration motor are readjusted, and the conveyor belt speed is adjusted. Then, the normal rice is identified and separated from the discolored rice grains using the discolored grain feature reverse indexing mechanism.

[0015] Preferably, the rice to be sorted is conveyed onto a vibrating conveyor belt, and vibration is applied to the conveyor belt to widen the spacing between the rice grains on the conveyor belt.

[0016] A servo vibration motor is installed below the conveyor belt, and the servo vibration motor applies controllable vibration to the conveyor belt. The controllable vibration includes adjustable vibration frequency and vibration amplitude. When the rice moves to the end of the conveyor belt, the rice grains are separated into individual grains through a guide rail device.

[0017] The surface roughness, effective working width, and length of the conveyor belt are configured according to production requirements.

[0018] Preferably, the trajectory of all rice grains falling from the conveyor belt is fitted in three-dimensional space to form a trajectory surface, and based on the fitted trajectory surface, the trajectory surface is divided into multiple functional areas along the falling direction;

[0019] The functional area includes a collection area, an identification area, and a separation area; the collection area is located in the initial falling section after the rice grains leave the conveyor belt rail.

[0020] Preferably, a high-speed linear CCD camera is used to capture images of falling rice grains during the falling process within the acquisition area. The high-speed linear CCD camera is installed at the normal direction of the tangent surface of the trajectory surface in the acquisition area, and the optical axis of the camera maintains a fixed distance from the trajectory surface of the rice grains.

[0021] The high-speed linear CCD camera uses a high-frequency flash LED array for illumination during shooting, which is synchronized with the camera's exposure. The LED array is distributed in a ring around the lens of the high-speed linear CCD camera. The illumination intensity of the LED array is dynamically adjusted by pulse width modulation, and the flash frequency is synchronized with the camera's line frequency.

[0022] Preferably, a color difference enhancement model is constructed, wherein the color difference enhancement model is constructed using an improved conditional generative adversarial network architecture, wherein the generative adversarial network architecture includes a generator network and a discriminator network;

[0023] The generator network includes an encoder and a decoder. The encoder adopts a progressive convolutional layer structure, and the decoder reconstructs the enhanced image of falling rice through deconvolution operations and introduces skip connections in each layer. The discriminator network adopts a PatchGAN structure, which divides the input image of falling rice into multiple local regions for real and fake discrimination.

[0024] Preferably, the local contrast of the falling rice image processed by the color difference enhancement model is magnified. The local contrast magnification adopts adaptive histogram equalization technology. A dynamic block strategy is designed according to the color distribution characteristics of the falling rice image to divide each falling rice image into sub-blocks. The area of ​​the sub-blocks is adaptively determined according to the size of the rice grains.

[0025] The image of falling rice is magnified by local contrast and then processed using dual-domain color shift stretching technology. The image of falling rice is processed in both RGB and HSI color spaces. In the RGB domain, each color channel is dynamically stretched independently, and in the HSI domain, the hue and saturation components are enhanced.

[0026] Preferably, a standard colorimetric library for rice is constructed as a benchmark for rice identification. The standard colorimetric library obtains colorimetric parameters by collecting normal rice samples, extracts colorimetric feature vectors for each sample in the CIELab color space, and divides the colorimetric parameters into multiple standard colorimetric intervals through cluster analysis.

[0027] A dynamic reference colorimetric bidirectional recognition algorithm is constructed. The algorithm extracts chromaticity feature vectors from the enhanced image of falling rice, calculates the color difference distance between the extracted chromaticity feature vectors and each reference point in the standard colorimetric library, and constructs a dynamic threshold function based on the color difference distribution. The dynamic threshold function automatically adjusts the discrimination criteria according to the overall chromaticity characteristics of different batches of rice.

[0028] A relative color difference mapping strategy is constructed to identify normal rice and rice with discolored grains by constructing a two-dimensional color difference distribution map.

[0029] Preferably, the different colored rice grains are calibrated in real time within the identification area. After calibration, the spatial coordinates of each different colored rice grain within the identification area are obtained by laser positioning, and the trajectory of the different colored rice grain within the identification area is calculated. Based on the trajectory of the different colored rice grain within the identification area, the movement trajectory of the different colored rice grain in the separation area is calculated.

[0030] Preferably, a reverse indexing mechanism for heterochromatic grain features is established. For each rice grain identified as heterochromatic, the chromaticity features of the heterochromatic rice grain are recorded, and a comprehensive feature vector of heterochromatic rice grains including texture, shape, and local color distribution is constructed.

[0031] The normal rice in the collected non-colored rice grains is separated by using a reverse indexing mechanism based on the non-colored grain characteristics.

[0032] Preferably, in the separation zone, the discolored rice grains in the falling rice are blown out by airflow, and the airflow is blown out from the airflow nozzle corresponding to the separation zone;

[0033] The airflow nozzles are arranged in an array, with multiple airflow nozzles evenly distributed laterally in the separation zone. Each airflow nozzle is controlled to open and close by a high-speed solenoid valve. The effective range of the airflow nozzles is determined by fluid dynamics simulation.

[0034] The airflow nozzle control strategy adopts a three-level control mechanism of prediction-trigger-feedback. In the triggering stage, the actual arrival position of the rice grain is detected by the laser sensor array. When the deviation between the detection signal and the prediction timing is less than the set threshold, the corresponding nozzle is activated to blow the discolored rice grain into the discolored rice container.

[0035] Preferably, the rice collected in the container of discolored rice grains is returned to the inlet of the vibrating conveyor belt, and the vibration parameters and running speed of the conveyor belt are optimized and adjusted to increase the spacing between the rice grains on the conveyor belt.

[0036] When a grain of rice passes through the identification area again, the reverse indexing mechanism for the discolored grain feature is activated to identify normal rice. For each grain of rice, its comprehensive feature vector is extracted and matched and identified in the reverse index table.

[0037] During the reverse selection process, normal rice is blown out through airflow nozzles in the separation zone of the falling rice surface, causing the normal rice to deflect during its fall and enter the collection container.

[0038] A dynamic grading system for reverse selection of discolored rice grains, used in the aforementioned dynamic grading method for reverse selection of discolored rice grains, includes: a vibration transmission module, a falling partition module, an image enhancement module, a rice grain recognition module, a pneumatic separation module, and a reverse selection module.

[0039] The vibration conveying module uses a vibration motor to vibrate the rice on the conveyor belt, and a guide rail is set at the end of the conveyor belt to allow individual grains of rice to detach from the conveyor belt and fall.

[0040] The falling partitioning module fits the falling trajectory of rice into a trajectory surface and divides the trajectory surface into a collection area, a recognition area, and a separation area. An industrial camera is used to continuously collect images of the falling rice in the collection area.

[0041] The image enhancement module performs local contrast adjustment on the acquired falling rice image based on the color difference enhancement model of generative adversarial network, and performs dynamic range adjustment on the acquired falling rice image in both RGB and HSI domains.

[0042] The rice grain identification module establishes a standard color library for rice, uses a dynamic reference chromatogram bidirectional identification algorithm and a relative color difference mapping strategy to identify rice grains with different colors, and performs calibration within the identification area. At the same time, it constructs a reverse indexing mechanism for the characteristics of rice grains with different colors to perform bidirectional identification of rice grains with different colors and normal rice.

[0043] The pneumatic separation module calculates the movement trajectory of the discolored rice grains in the separation zone based on the calibration results, and controls the opening timing of the airflow nozzles to blow the discolored rice grains into the discolored grain container.

[0044] The reverse selection module, after separating the discolored rice grains of the current batch, returns the rice from the discolored grain container to the conveyor belt inlet, readjusts the vibration parameters applied to the conveyor belt by the vibration motor and adjusts the conveyor belt speed, and then uses the discolored grain feature reverse indexing mechanism to identify and separate the normal rice from the discolored grain rice.

[0045] The beneficial effects of this application are as follows: By setting up a vibration motor and guide rail, this application realizes the single-grain falling of rice, which provides good conditions for subsequent image acquisition and recognition. The mechanical preprocessing method is simple, reliable and easy to control, and has good stability and reliability.

[0046] This application precisely divides the collection area, recognition area, and separation area by fitting the trajectory of falling rice, and then uses an industrial camera to continuously acquire images, providing a data foundation for subsequent image processing and discrimination, and ensuring the real-time performance and accuracy of the acquired data.

[0047] This application employs a color difference enhancement model based on generative adversarial networks and dynamic range adjustment of RGB and HSI dual domains, which effectively improves the quality of the acquired images and enhances color contrast and local details. This is beneficial for improving the accuracy of subsequent identification of discolored rice grains, especially for discolored grains with small color differences, which can significantly improve the identification effect.

[0048] This application constructs a standard colorimetric library for rice and adopts a dynamic reference chromatogram bidirectional recognition algorithm and a relative color difference mapping strategy, which can accurately identify rice grains with different colors. The bidirectional recognition mechanism can not only identify rice grains with different colors, but also identify normal rice grains among the different colors through reverse indexing.

[0049] Based on the identification results, this application precisely controls the opening timing of the airflow nozzles, which can efficiently and accurately separate discolored rice grains; the airflow separation response speed is fast, which helps to improve the sorting efficiency of rice.

[0050] This application uses a reverse indexing mechanism to perform secondary screening by re-injecting discolored grains and readjusting vibration and transmission parameters, thereby minimizing losses during the rice screening process while ensuring the removal of discolored grains. Attached Figure Description

[0051] Figure 1 A flowchart of a dynamic grading method for controlling the reverse selection of discolored rice grains provided in this application;

[0052] Figure 2 This application provides a flowchart of rice transport and single-grain shedding;

[0053] Figure 3 This application provides a flowchart of the rice falling area division and image acquisition process;

[0054] Figure 4 This application provides a flowchart for color difference enhancement and dynamic range stretching of rice images;

[0055] Figure 5 This application provides a flowchart for the identification and calibration of discolored rice grains;

[0056] Figure 6 This application provides a schematic diagram of the calibration results for discolored rice grains;

[0057] Figure 7 This application provides a flow chart for the airflow separation of discolored rice grains;

[0058] Figure 8 Provide a flowchart of the normal rice reverse selection process for this application;

[0059] Figure 9 This application provides a structural diagram of a dynamic grading and discolored rice reverse selection control device. Detailed Implementation

[0060] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the specific embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0061] Many specific details are set forth in the following description in order to provide a full understanding of this application. However, this application may also be implemented in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0062] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of this application. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0063] Example 1

[0064] Reference Figures 1 to 8 This is the first embodiment of the present application, such as Figure 1 The diagram shows a flowchart of a dynamic grading method for reverse selection control of discolored rice grains.

[0065] Step 1: Vibration motors apply vibration to the rice on the conveyor belt, and guide rails are installed at the end of the conveyor belt to allow individual grains of rice to detach from the conveyor belt and fall; see [link / reference] Figure 2 This is a flowchart of the rice transfer and individual grain shedding process in this step.

[0066] The rice to be sorted is spread on a vibrating conveyor belt made of food-grade material with a fine mesh texture on the surface. The texture depth and spacing are optimized according to the characteristics of the rice variety. The surface of the conveyor belt must maintain a certain roughness to maintain friction with the rice and prevent the rice from slipping or rolling during the conveying process, ensuring that the rice can move synchronously with the conveyor belt. The surface parameters, effective working width, and length of the conveyor belt are configured and changed according to production capacity requirements to meet the capacity requirements for sorting discolored rice.

[0067] A vibration motor is installed below the conveyor belt; the vibration motor is a servo motor, which applies controllable vibration to the conveyor belt. The controllable vibration includes adjustable vibration frequency and vibration amplitude. By applying vibration, the rice grains on the conveyor belt are evenly dispersed.

[0068] Applied vibration frequency Achieving uniform dispersion of rice is crucial and requires comprehensive consideration of factors such as rice variety, particle size distribution, and required sorting capacity. Vibration frequency... The calculation formula is: ;in, It is the equivalent stiffness coefficient of the conveyor belt, which is related to parameters such as the material, size, and tension of the conveyor belt; It is the acceleration due to gravity; The amplitude of the conveyor belt; The mass per unit length of the conveyor belt; The mass per unit length of the rice sample on the conveyor belt.

[0069] For example, for rice varieties with typical particle size, when the equivalent stiffness coefficient of the conveyor belt is 750 N / mm, the amplitude can be set to 6 mm, the unit length mass of the conveyor belt is 2.5 kg / m, and the unit length mass of the rice is 0.75 kg / m, the calculated vibration frequency is approximately 25 Hz. By monitoring the rice feeding process online and combining it with visual feedback, the vibration parameters are dynamically optimized and adjusted to achieve the ideal dispersion effect of the rice.

[0070] After being thoroughly vibrated and dispersed, the rice is driven by a conveyor belt at a constant speed. For stable operation, the conveyor belt speed should be specifically set according to the grain size, shape characteristics, and vibration state of different rice varieties. This ensures the rice is sufficiently dispersed while preventing spillage due to excessive speed. The calculation formula is: ;in, This refers to the distance the rice travels on the conveyor belt, i.e., the effective length of the conveyor belt. The time required for the rice to travel along the conveyor belt; The frequency of the vibrating motor; This represents the number of vibrations the rice experiences on the conveyor belt.

[0071] For example, when the effective length of the conveyor belt is 2000mm and the vibration frequency is 25Hz, and it is expected that the rice will receive 100 vibrations during the conveying process, the calculated conveyor belt speed should be set to 500mm / s. At this time, the time for the rice to pass through the conveyor belt is 4 seconds, which can achieve a sufficient vibration dispersion effect.

[0072] As the rice moves to the end of the conveyor belt, a guide rail device separates each grain individually and directs its fall. The guide rail is made of engineering plastic with excellent self-lubricating properties, ensuring minimal sliding resistance for the rice within it. The guide rail's trapezoidal cross-section, wider at the top and narrower at the bottom, helps guide the rice towards the center. A gradually increasing array of wedge-shaped grooves is machined at the guide rail inlet, with the groove width gradually increasing from the inlet to the outlet. This effectively diverts and guides the rice, preventing multiple grains from entering the same channel simultaneously and causing blockages.

[0073] This step, through the synergistic effect of conveyor belt vibration dispersion and guide rail directional separation, achieves an efficient transformation of rice from a disordered accumulation state to an ordered, single-grain falling state. The operating parameters can be flexibly adjusted according to the physical characteristics of different rice varieties, exhibiting good adaptability and versatility. The vibration dispersion mechanism in this step effectively disrupts the agglomeration of rice grains, while the guide rail separation device ensures the single-row arrangement and trajectory controllability of the rice grains, creating ideal material flow conditions for subsequent high-speed image acquisition and intelligent recognition.

[0074] Step 2: Fit the trajectory of the falling rice into a trajectory surface and divide the area into a collection area, a recognition area, and a separation area based on the trajectory surface. Use an industrial camera to continuously collect images of the falling rice in the collection area; see [link / reference] Figure 3 This is a flowchart of the rice falling area division and image acquisition process for this step.

[0075] A mathematical model is used to depict the trajectory of rice grains after they leave the conveyor belt. The rice grains undergo parabolic motion under gravity. Considering the minimal difference between the spatial distribution of the conveyor belt's exit track and the initial velocities of the rice grains, the trajectories of all grains are represented as a regular parabola in three-dimensional space. A spatial coordinate system is established with the center of the exit track as the origin, the horizontal direction of motion as the x-axis, the vertical downward direction as the z-axis, and the horizontal direction as the y-axis. The equation for the trajectory of a single rice grain is as follows: , ,in, , These are the coordinates of the rice grain in the horizontal and vertical directions at time t, respectively. The initial velocity of the rice grain when it leaves the guide rail is the same as the conveyor belt speed. Let be the angle between the initial velocity and the horizontal plane. It is the acceleration due to gravity. For time. Through statistical analysis of multiple trajectories, the trajectory envelope is fitted as a quadratic surface. ,in to The fitting coefficients can be obtained by the least squares method. x, y, and z are the spatial coordinates of the rice as it falls, with the starting point of the fall as the origin.

[0076] Based on the fitted trajectory surface, the space is divided into three functional regions along the falling direction: a collection region, a recognition region, and a separation region. The collection region is located in the initial falling segment after the rice grains leave the guide rail. Within this region, the spacing between rice grains is moderate and their movement is stable, which is conducive to obtaining clear individual images. The starting position height of the collection region is... and end position height Based on the camera's depth of field and the speed of the rice grains, the following relationship is satisfied. ,in, For the effective depth of field of industrial cameras, This refers to the optical magnification. The recognition area is located below the acquisition area, allowing sufficient time for image processing and feature recognition. The length of the recognition area... ,in This represents the average falling speed of the rice grains in that area. This represents the execution time of the image processing and recognition algorithms. The separation zone is located below the recognition zone and at the end of the descent trajectory. It is the area where the physical separation action is performed, and its position and length are determined based on the response time and effective range of the airflow nozzles.

[0077] The acquisition area uses a high-speed linear CCD camera to capture images of falling rice grains. The high-speed linear CCD camera is installed at the normal direction of the curved surface of the acquisition area; the camera's optical axis maintains a fixed distance from the curved surface of the rice grain's trajectory. This distance is determined through depth-of-field calculations. ,in, For camera lens focal length, This refers to the aperture number. The diameter of the circle of confusion. This is for hyperfocal distance. The camera uses continuous scan mode with a line frequency of... Matching the falling speed of the rice grains to meet the requirements ,in, Let be the vertical velocity of the rice grain falling in the collection area. The desired spatial resolution. For example, when the rice grain falls at a speed of 2.5 m / s and a spatial resolution of 0.1 mm is required, the camera's line frequency needs to be set to 25 kHz.

[0078] To ensure the integrity and continuity of image acquisition, a synchronous triggering mechanism is configured in the acquisition area. A laser sensor detects whether rice grains have entered the acquisition area; when a rice grain is detected, a trigger signal is generated, initiating the camera's image acquisition sequence. The sensor array is arranged in a matrix, with row spacing of... and column spacing satisfy , ,in, and These are the set minimum length and width of the rice grains. This dense arrangement ensures that every grain of rice can be reliably detected, preventing any missed detections.

[0079] The high-speed linear CCD camera employs a high-frequency flash LED array for illumination, synchronized with the camera's exposure. The LED array is arranged in a ring around the camera lens, providing uniform diffused illumination to eliminate specular reflections and shadows on the rice grain surface. The illumination intensity is dynamically adjusted via PWM to maintain consistent lighting conditions during the separation of dissimilarly colored rice from the same batch, ensuring stable brightness and contrast in the acquired rice images. The flash frequency is strictly synchronized with the camera's line frequency, and the pulse width... To avoid motion blur.

[0080] This step transforms the continuous falling rice grain process into discrete functional regions through precise trajectory modeling and spatial partitioning. The parameters of each region are optimized for its specific function. The reasonable setting of the acquisition area ensures image quality, the recognition area allows sufficient processing time, and the precise positioning of the separation area improves sorting accuracy. The combined use of a high-speed camera and synchronized illumination achieves clear imaging of fast-moving rice grains, providing high-quality raw data for subsequent image processing and feature recognition. This step effectively solves the problem of accurate acquisition of dynamic targets, laying a solid foundation for achieving high-precision identification and classification of dissimilar colored grains.

[0081] Step 3: A color difference enhancement model based on a generative adversarial network is used to locally adjust the contrast of the acquired falling rice images, and the dynamic range of the acquired falling rice images is adjusted in both the RGB and HSI domains; see [link / reference]. Figure 4 This is a flowchart of the color difference enhancement and dynamic range stretching process for the rice image in this step.

[0082] In this step, a specialized Generative Adversarial Network (GAN) color difference enhancement model is constructed to address the extremely subtle color difference features between slightly yellow rice, light yellow rice, and normal rice. The color difference enhancement model employs an improved conditional generative adversarial network architecture, with the generator network consisting of an encoder and a decoder. The encoder of the generator network uses a progressive convolutional layer structure, extracting multi-scale features of the rice image layer by layer. Each convolutional layer is followed by a batch normalization layer and a ReLU activation function. The decoder reconstructs the enhanced image through deconvolution operations and introduces skip connections at each layer to preserve the detailed information of the original image. The discriminator network uses a PatchGAN structure, dividing the input image into multiple local regions for real / fake detection, improving sensitivity to local color difference features.

[0083] Loss function of generator network Designed for ,in, To counteract losses, it is used to deceive the discriminator; To compensate for color enhancement loss, it is specifically optimized for subtle color differences; To preserve edge quality and prevent grain contour distortion during enhancement; , , These are weighting coefficients, dynamically adjusted based on the training process; color enhancement loss. The definition is based on the change in color difference between the images before and after enhancement in the Lab color space. ,in, Image location Weighting factor at the location, Image location The CIE color difference value at the location; this design allows the model to focus on enhancing areas with smaller color differences while maintaining the naturalness of areas with significant color differences.

[0084] The rice image processed by the color difference enhancement model undergoes local contrast amplification using adaptive histogram equalization. Unlike traditional methods, this application designs a dynamic block-segmentation strategy based on the color distribution characteristics of the rice grain images, dividing each rice grain image into... Each sub-block has a size that is adaptively determined based on the size of a rice grain, satisfying the following requirements: ,in, For the area of ​​the sub-block, The projected area of ​​a grain of rice. These are the block coefficients. The local contrast enhancement function is calculated for each sub-block. ,in, These are the original pixel values. The average value of the sub-blocks. and For linear transformation parameters, It is a non-linear enhancement index.

[0085] A dual-domain color stretching technique is employed to process the acquired rice images simultaneously in both the RGB and HSI color spaces. Within the RGB domain, each color channel is independently subjected to dynamic range stretching, with the stretching function being... ,in, The original red channel value. This is the minimum value of the channel. The elongation factor is 1. The target is the minimum value. The green and blue channels in the RGB domain are treated similarly, but the stretching factor is set differently based on the contribution of each channel to the color difference.

[0086] In the HSI domain, the focus is on enhancing the hue (H) and saturation (S) components, while keeping the luminance (I) component essentially unchanged. Hue enhancement employs a piecewise linear mapping. ,in, For the enhanced hue, As a hue shift function, when the different colored rice is set to slightly yellow rice, the hue enhancement has a larger gain in the slightly yellow hue range, while other hue ranges maintain a smaller shift to avoid color distortion.

[0087] This step leverages the deep learning capabilities of generative adversarial networks to automatically learn and identify the unique color patterns of discolored rice grains, achieving targeted enhancement. Local contrast amplification technology then enhances the discernibility of subtle color differences in the image, making previously imperceptible color variations clearly visible. Dual-domain color shift stretching fully utilizes the advantages of different color spaces, enhancing color differences while maintaining the naturalness and realism of the image. Throughout the color difference enhancement process, the color difference between slightly yellow rice and normal rice is amplified several times, significantly improving the detection capability for discolored rice grains with minute color differences.

[0088] Step 4: Establish a standard colorimetric library for rice, use a dynamic reference chromatogram bidirectional identification algorithm and a relative color difference mapping strategy to identify discolored rice grains, and calibrate them within the identification area. Simultaneously, construct a reverse indexing mechanism for discolored grain features to bidirectionally identify discolored rice grains and normal rice; see [link / reference]. Figure 5 This is a flowchart of the identification and calibration of discolored rice grains in this step;

[0089] A standard rice colorimetric library was constructed as the identification benchmark. Normal rice samples were collected, and accurate colorimetric parameters were obtained using a spectrophotometer and a standard light source box. For each sample, a colorimetric feature vector was extracted in the CIELab color space. ,in, This refers to the brightness value. and For chromaticity coordinates, The chroma value is... The colorimetric data is divided into several standard colorimetric intervals through cluster analysis, with the center value of each interval serving as the standard colorimetric benchmark for that category. The colorimetric library can adopt a hierarchical storage structure, establishing multi-dimensional indexes based on factors such as rice variety, origin, and season, facilitating rapid retrieval and dynamic updates.

[0090] A dynamic reference chromatogram bidirectional recognition algorithm is constructed. The core of this algorithm lies in establishing an adaptive colorimetric comparison mechanism. The dynamic reference chromatogram bidirectional recognition algorithm first extracts colorimetric features from the enhanced rice grain image and calculates the color difference distance between it and each reference point in the standard colorimetric library. ,in, , and These represent the lightness value and chromaticity coordinates of the rice grain to be measured, respectively. , and This represents the lightness and chromaticity coordinates of normal rice. A dynamic threshold function is constructed based on the color difference distribution. ,in, For dynamic thresholds, Based on the threshold, This represents the standard deviation of color difference for the current batch. The average color difference. and The adaptive coefficient; the dynamic threshold mechanism can automatically adjust the discrimination criteria according to the overall color characteristics of different batches of rice, which can improve the adaptability of the algorithm.

[0091] A relative color difference mapping technique is constructed to achieve accurate discrimination by building a two-dimensional color difference distribution map; the chromaticity coordinates of each grain of rice are mapped to a relative color difference distribution map. - Calculate the offset vector of the rice cluster relative to the normal rice cluster center on the plane of coordinate axes. ,in, This indicates the offset along the a* axis (representing the "red-green" direction in the Lab color space). This represents the offset along the b* axis (representing the "yellow-blue" direction in the Lab color space). A relative color difference index is defined. ,in, The magnitude of the offset vector. The normal rice cluster radius, The angle between the offset vector and the direction of the slightly yellow feature is considered relative to the color difference index. If the value exceeds the set threshold, the rice grain is determined to be an off-color rice grain.

[0092] Discolored rice grains are identified in real time within the identification area by marking them with wireframes. (See attached image.) Figure 6 This is a schematic diagram of the calibration results for the discolored rice grains. After calibrating the discolored rice grains, the spatial coordinates of each discolored rice grain in the identification area are obtained through laser positioning, and the trajectory of the discolored rice grains in the identification area is calculated. Then, based on the trajectory of the discolored rice grains in the identification area, the movement trajectory of the discolored rice grains in the separation area is calculated, and the discolored rice grains are separated in the separation area.

[0093] To achieve bidirectional discrimination between discolored rice grains and normal rice, this application establishes a bidirectional discrimination mechanism for identifying rice color. This mechanism identifies rice from both forward and reverse perspectives. Forward discrimination, based on comparison with a standard colorimetric library, determines whether rice grains deviate from the normal range, thus separating discolored rice grains from normal rice. Reverse discrimination utilizes a reverse indexing mechanism based on discolored grain features to verify whether rice grains identified as discolored truly possess typical discolored characteristics, thereby enabling the identification and separation of normal rice from discolored rice grains. This bidirectional discrimination mechanism allows for the reverse selection of normal rice from discolored rice grains, reducing losses during the rice screening process.

[0094] A reverse indexing mechanism for heterochromatic grain features is established; for each rice grain identified as heterochromatic, not only is its chromaticity feature vector recorded, but a comprehensive feature vector including texture, shape, and local color distribution is also constructed. These characteristics are used to generate unique index keys through a hash function. The data is stored in an inverted index table. When performing reverse selection of rice, historical records are retrieved based on feature matching. Assess the similarity between the current rice grain and historical records to identify whether the current rice is of discolored origin. and These are the current feature and the historical feature vector, respectively.

[0095] This step establishes a comprehensive standard colorimetric library, obtaining a reliable identification benchmark that can adapt to different varieties and batches of rice. The adaptive nature of the dynamic benchmark chromatography algorithm ensures stable identification performance even under changing environmental conditions. Relative color difference mapping technology transforms abstract color differences into intuitive geometric relationships, improving the accuracy and interpretability of the discrimination. The inverse indexing mechanism for discolored grain features enables reverse selection of rice. The entire process, through multi-dimensional and multi-level identification strategies, achieves accurate discrimination of subtle color differences, such as slightly yellow rice, providing an accurate and reliable decision-making basis for subsequent physical separation.

[0096] Step 5: Calculate the coordinates of the discolored rice grains in the separation zone based on the calibration results, and control the timing of the airflow nozzle opening to blow the discolored rice grains into the discolored grain container; see [link / reference]. Figure 7 This is a flowchart of the airflow separation process for discolored rice grains in this step.

[0097] Establish a trajectory prediction equation for heterochromatic rice grains , ,in, and The predicted coordinates of the rice grains in the separation zone. and The coordinates of the identification area, and The velocity component of the rice grain in the recognition area. To identify the time, It is the acceleration due to gravity. and The drag coefficient is measured in real time by a high-speed linear CCD camera to measure the motion parameters of rice grains in the recognition area. The drag coefficient is then dynamically corrected based on the mass and shape characteristics of the rice grains, achieving millimeter-level position prediction accuracy.

[0098] The airflow nozzles are arranged in an array, with multiple independently controllable micro-nozzles evenly distributed laterally in the separation zone. Each nozzle is equipped with a high-speed solenoid valve with a response time of less than 5 milliseconds, enabling it to precisely open and close the nozzle the instant a rice grain passes through.

[0099] The effective range of the airflow nozzle was determined through fluid dynamics simulation, and the airflow field distribution function was established. ,in For spatial points air pressure at that location This refers to the nozzle outlet pressure. Here are the nozzle position coordinates. , , These are the airflow diffusion parameters. Based on the predicted trajectory of the rice grains and the airflow field distribution, the optimal nozzle combination is automatically selected to ensure that sufficient lateral thrust is applied to the target rice grains while minimizing impact on nearby normal rice.

[0100] The airflow nozzle control strategy employs a three-stage control mechanism of prediction-triggering-feedback; in the prediction stage, the optimal triggering timing is calculated based on the grain trajectory. ,in, The nozzle's effective height is defined as follows: During the triggering phase, a laser sensor array detects the actual arrival position of the rice grains. When the deviation between the detected signal and the predicted timing is less than a set threshold, the corresponding nozzle is immediately activated. During the feedback phase, downstream sensors verify the separation effect, dynamically adjusting the injection pressure and duration. For example, when discolored rice grains enter the separation zone at a speed of 2.8 m / s, it is predicted that they will reach the nozzle's effective position after 15 milliseconds. At the moment the rice grain arrives, the corresponding nozzle is activated 2 milliseconds before it arrives, and a lateral airflow of 0.25 MPa is applied at the instant the rice grain arrives, lasting for 8 milliseconds, generating an impulse of about 15 N·ms, which deflects the rice grain to the container for the different colored grains.

[0101] This step achieves accurate physical separation of discolored rice grains through the synergistic effect of precise trajectory prediction and high-speed airflow control. The predictive model considering air resistance significantly improves position prediction accuracy, laying the foundation for precise separation. The rapid response and precise control capabilities of the array nozzle system ensure that force is applied only to the target discolored rice grains, avoiding impact on normal rice. The adaptive airflow parameter optimization strategy can handle rice grains with various physical characteristics, improving the stability and reliability of the separation. This separation process is efficient and precise, reliably removing discolored rice grains while minimizing the impact on normal rice, significantly improving the automation level and sorting accuracy of rice quality control.

[0102] Step 6: After separating the discolored rice grains from the current batch, return the rice from the discolored grain container to the conveyor belt inlet, readjust the vibration parameters applied to the conveyor belt by the vibration motor, and adjust the conveyor belt speed. Then, use the discolored grain feature reverse indexing mechanism to identify and separate the normal rice from the discolored grains, thus separating the normal rice from the discolored grains; see [link / reference]. Figure 8 This is a flowchart of the normal rice reverse selection process for this step.

[0103] When separating normal rice from discolored rice grains, the number of separated discolored rice grains (containing mixed normal rice) is significantly reduced compared to the number of rice grains before separation. Therefore, normal rice can be more accurately separated by increasing the falling time interval between discolored rice grains, thus avoiding the re-mixing of normal rice with discolored rice grains.

[0104] The rice collected in the container with the discolored grains is returned to the inlet of the vibrating conveyor belt. To better separate the normal rice mixed with the discolored grains, the vibration parameters of the conveyor belt need to be optimized. First, the vibration frequency of the conveyor belt is reduced, and the vibration amplitude is increased to allow the rice grains to be more fully dispersed and jump around on the conveyor belt, thus increasing the spacing between the rice grains on the conveyor belt. The optimized vibration frequency... The calculation formula is: ;in, The optimized equivalent stiffness coefficient of the conveyor belt. To increase the vibration amplitude of the conveyor belt It is the acceleration due to gravity. The mass per unit length of the conveyor belt, To reduce the mass per unit length of the discolored particles on the conveyor belt; by lowering Increasing the vibration amplitude can effectively reduce the vibration frequency and prolong the movement time of rice grains on the conveyor belt.

[0105] By reducing the vibration frequency and increasing the vibration amplitude of the conveyor belt, and simultaneously reducing the conveyor belt speed, the residence time of rice grains on the conveyor belt is further extended, allowing for more thorough dispersion and jumping. The optimized conveyor belt speed... The calculation formula is: ;in, The effective length of the conveyor belt. This is the time required for a grain of rice to pass through the conveyor belt. For the optimized vibration frequency, This refers to the number of vibrations experienced by the rice grains on the conveyor belt. By appropriately reducing... This allows the rice grains to achieve a more thorough vibration and dispersion effect on the conveyor belt.

[0106] After optimization and adjustment, the discolored rice grains that were returned to the conveyor belt were dispersed and jumped at a greater distance, creating favorable conditions for the subsequent reverse selection process to accurately identify and separate normal rice.

[0107] When a grain of rice passes through the recognition area again, the inverse indexing mechanism for discolored grains is activated for reverse selection. For each grain of rice, its comprehensive feature vector is extracted. Through hash function Generate a unique index key. It is the unique index key of the current rice grain, and it is used for fast matching in the reverse index table to identify whether the rice grain is normal rice. Unlike the forward discrimination during the initial sorting, the goal of reverse sorting is to identify normal rice.

[0108] Define feature similarity function ,in, This is the feature vector of the current rice grain. This is the feature vector of normal rice stored in the inverted index table. When similarity... Exceeding the preset similarity threshold At that time, that is If so, the current rice grain will be marked as normal rice and used as the target for current airflow separation.

[0109] During the reverse rice selection process, within the separation zone of the rice falling surface, the airflow nozzle applies airflow force to the target marked as normal rice, causing it to deflect and enter the normal rice collection container. The control strategy for airflow parameters is the same as in step 5, and can be optimized in real time according to the physical characteristics and motion state of the rice grains to ensure accurate separation.

[0110] This step, through optimized adjustment of the vibrating conveyor belt parameters, expands the falling distance of discolored rice grains, providing more favorable conditions for accurate identification of normal rice in reverse selection. The application of hash indexing and feature similarity matching algorithms enables rapid screening of normal rice mixed with discolored grains, significantly improving identification efficiency and accuracy. The perfect combination with airflow separation technology ensures that normal rice can be reliably separated from discolored grains, minimizing rice waste. This step's technical solution, through the optimized combination of vibration dispersion, reverse indexing identification, and airflow separation, forms a highly efficient, accurate, and comprehensive screening scheme for both positive and negative selection of discolored rice grains.

[0111] This technical solution also makes full use of the big data resources accumulated from the initial color sorting, combined with innovative algorithm design and precise process control, to improve rice yield while ensuring finished product quality, and realize closed-loop optimization and intelligent upgrading of the color sorting process; providing new ideas and methods for the high-quality development of the rice processing industry.

[0112] Example 2

[0113] Reference Figure 9 This is the second embodiment of the present application, which provides a dynamic grading system for the reverse selection of discolored rice grains.

[0114] The system includes: a vibration transmission module, a falling partition module, an image enhancement module, a rice grain recognition module, a pneumatic separation module, and an inverse selection module;

[0115] The vibration conveying module uses a vibration motor to vibrate the rice on the conveyor belt, and a guide rail is set at the end of the conveyor belt to allow individual grains of rice to detach from the conveyor belt and fall.

[0116] The falling partitioning module fits the falling trajectory of rice into a trajectory surface and divides the trajectory surface into a collection area, a recognition area, and a separation area. An industrial camera is used to continuously collect images of the falling rice in the collection area.

[0117] The image enhancement module performs local contrast adjustment on the acquired falling rice image based on the color difference enhancement model of generative adversarial network, and performs dynamic range adjustment on the acquired falling rice image in both RGB and HSI domains.

[0118] The rice grain identification module establishes a standard color library for rice, uses a dynamic reference chromatogram bidirectional identification algorithm and a relative color difference mapping strategy to identify rice grains with different colors, and performs calibration within the identification area. At the same time, it constructs a reverse indexing mechanism for the characteristics of rice grains with different colors to perform bidirectional identification of rice grains with different colors and normal rice.

[0119] The pneumatic separation module calculates the coordinates of the discolored rice grains in the separation zone based on the calibration results, and controls the opening timing of the airflow nozzles to blow the discolored rice grains into the discolored grain container.

[0120] The reverse selection module, after separating the discolored rice grains of the current batch, returns the rice from the discolored grain container to the conveyor belt inlet, readjusts the vibration parameters applied to the conveyor belt by the vibration motor and adjusts the conveyor belt speed, and then uses the discolored grain feature reverse indexing mechanism to identify and separate the normal rice from the discolored grain rice.

[0121] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings or direct couplings or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0122] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of this application without departing from the spirit and scope of protection of the claims. All of these variations are within the protection scope of this application.

Claims

1. A method for controlling the reverse selection of discolored rice grains through dynamic grading, characterized in that, include: The rice on the conveyor belt is vibrated by a vibrating motor, and a guide rail is set at the end of the conveyor belt to make individual grains of rice detach from the conveyor belt and fall. The trajectory of falling rice is fitted to a trajectory surface, and the collection area, recognition area and separation area are divided according to the trajectory surface. An industrial camera is used to continuously collect images of falling rice in the collection area. A color difference enhancement model based on generative adversarial networks is used to adjust the local contrast of the collected images of falling rice, and to adjust the dynamic range of the collected images of falling rice in both RGB and HSI domains. A standard color library for rice was established. A dynamic reference chromatographic bidirectional identification algorithm and a relative color difference mapping strategy were used to identify rice grains with different colors and to perform calibration within the identification area. At the same time, a reverse indexing mechanism for the characteristics of rice grains with different colors was constructed to perform bidirectional identification of rice grains with different colors and normal rice. The trajectory of the discolored rice grains in the separation zone is calculated based on the calibration results, and the timing of the opening of the airflow nozzles is controlled to blow the discolored rice grains into the discolored grain container. After separating the discolored rice grains from the current batch, the rice in the discolored grain container is returned to the conveyor belt inlet. The vibration parameters applied to the conveyor belt by the vibration motor are readjusted, and the conveyor belt speed is adjusted. Then, the normal rice is identified and separated from the discolored rice grains using the discolored grain feature reverse indexing mechanism.

2. The method for dynamic grading and reverse selection control of discolored rice grains according to claim 1, characterized in that, The rice to be sorted is fed onto a vibrating conveyor belt, and vibration is applied to the conveyor belt to widen the spacing between the rice grains on the conveyor belt. A servo vibration motor is installed below the conveyor belt, and the servo vibration motor applies controllable vibration to the conveyor belt. The controllable vibration includes adjustable vibration frequency and vibration amplitude. When the rice moves to the end of the conveyor belt, the rice grains are separated into individual grains through a guide rail device. The surface roughness, effective working width, and length of the conveyor belt are configured according to production requirements.

3. The method for dynamic grading and reverse selection control of discolored rice grains according to claim 2, characterized in that, The trajectory of all the rice grains falling from the conveyor belt is fitted in three-dimensional space to form a trajectory surface. Based on the fitted trajectory surface, the trajectory surface is divided into multiple functional areas along the falling direction. The functional area includes a collection area, an identification area, and a separation area; the collection area is located in the initial falling section after the rice grains leave the conveyor belt rail.

4. The method for dynamic grading and reverse selection control of discolored rice grains according to claim 3, characterized in that, Within the acquisition area, a high-speed linear CCD camera is used to capture images of falling rice grains during the falling process. The high-speed linear CCD camera is installed at the normal direction of the tangent surface of the trajectory surface in the acquisition area, and the optical axis of the camera maintains a fixed distance from the trajectory surface of the rice grains. The high-speed linear CCD camera uses a high-frequency flash LED array for illumination during shooting, which is synchronized with the camera's exposure. The LED array is distributed in a ring around the lens of the high-speed linear CCD camera. The illumination intensity of the LED array is dynamically adjusted by pulse width modulation, and the flash frequency is synchronized with the camera's line frequency.

5. The method for dynamic grading and reverse selection control of discolored rice grains according to claim 4, characterized in that, A color difference enhancement model is constructed using an improved conditional generative adversarial network architecture, which includes a generator network and a discriminator network. The generator network includes an encoder and a decoder. The encoder adopts a progressive convolutional layer structure, and the decoder reconstructs the enhanced image of falling rice through deconvolution operations and introduces skip connections in each layer. The discriminator network adopts the PatchGAN structure, which divides the input image of falling rice into multiple local regions for authenticity discrimination.

6. The method for dynamic grading and reverse selection control of discolored rice grains according to claim 5, characterized in that, The local contrast of the falling rice image processed by the color difference enhancement model is magnified. The local contrast magnification adopts the adaptive histogram equalization technology. A dynamic block strategy is designed according to the color distribution characteristics of the falling rice image to divide each falling rice image into sub-blocks. The area of ​​the sub-blocks is adaptively determined according to the size of the rice grains. The image of falling rice is magnified by local contrast and then processed using dual-domain color shift stretching technology. The image of falling rice is processed in both RGB and HSI color spaces. In the RGB domain, each color channel is dynamically stretched independently, and in the HSI domain, the hue and saturation components are enhanced.

7. The method for dynamic grading and reverse selection control of discolored rice grains according to claim 6, characterized in that, A standard colorimetric library for rice is constructed as a benchmark for rice identification. The standard colorimetric library obtains colorimetric parameters by collecting normal rice samples, extracts colorimetric feature vectors for each sample in the CIELab color space, and divides the colorimetric parameters into multiple standard colorimetric intervals through cluster analysis. A dynamic reference colorimetric bidirectional recognition algorithm is constructed. The algorithm extracts chromaticity feature vectors from the enhanced image of falling rice, calculates the color difference distance between the extracted chromaticity feature vectors and each reference point in the standard colorimetric library, and constructs a dynamic threshold function based on the color difference distribution. The dynamic threshold function automatically adjusts the discrimination criteria according to the overall chromaticity characteristics of different batches of rice. A relative color difference mapping strategy is constructed to identify normal rice and rice with discolored grains by constructing a two-dimensional color difference distribution map.

8. The method for dynamic grading and reverse selection control of discolored rice grains according to claim 7, characterized in that, The different colored rice grains are calibrated in real time within the identification area. After calibration, the spatial coordinates of each different colored rice grain within the identification area are obtained through laser positioning, and the trajectory of the different colored rice grains within the identification area is calculated. Based on the trajectory of the different colored rice grains within the identification area, the movement trajectory of the different colored rice grains in the separation area is calculated.

9. The method for dynamic grading and reverse selection control of discolored rice grains according to claim 8, characterized in that, Establish a reverse indexing mechanism for heterochromatic grain features. For each rice grain identified as heterochromatic, record the chromaticity features of the heterochromatic rice and construct a comprehensive feature vector of heterochromatic rice that includes texture, shape, and local color distribution. The normal rice in the collected non-colored rice grains is separated by using a reverse indexing mechanism based on the non-colored grain characteristics.

10. The method for dynamic grading and reverse selection control of discolored rice grains according to claim 9, characterized in that, In the separation zone, discolored rice grains are blown out of the falling rice by airflow, which comes from the airflow nozzles corresponding to the separation zone. The airflow nozzles are arranged in an array, with multiple airflow nozzles evenly distributed laterally in the separation zone. Each airflow nozzle is controlled to open and close by a high-speed solenoid valve. The effective range of the airflow nozzles is determined by fluid dynamics simulation. The airflow nozzle control strategy adopts a three-level control mechanism of prediction-trigger-feedback. In the triggering stage, the actual arrival position of the rice grain is detected by the laser sensor array. When the deviation between the detection signal and the prediction timing is less than the set threshold, the corresponding nozzle is activated to blow the discolored rice grain into the discolored rice container.

11. The method for dynamic grading and reverse selection control of discolored rice grains according to claim 10, characterized in that, The rice collected in the container of discolored rice grains is returned to the entrance of the vibrating conveyor belt, and the vibration parameters and running speed of the conveyor belt are optimized and adjusted to increase the spacing between the rice grains on the conveyor belt. When a grain of rice passes through the identification area again, the reverse indexing mechanism for the discolored grain feature is activated to identify normal rice. For each grain of rice, its comprehensive feature vector is extracted and matched and identified in the reverse index table. During the reverse selection process, normal rice is blown out through airflow nozzles in the separation zone of the falling rice surface, causing the normal rice to deflect during its fall and enter the collection container.

12. A dynamic grading system for reverse selection of discolored rice grains, used to implement the dynamic grading system for reverse selection of discolored rice grains as described in any one of claims 1 to 11, characterized in that, include: Vibration transmission module, falling partition module, image enhancement module, rice grain recognition module, pneumatic separation module, and reverse selection module; The vibration conveying module uses a vibration motor to vibrate the rice on the conveyor belt, and a guide rail is set at the end of the conveyor belt to allow individual grains of rice to detach from the conveyor belt and fall. The falling partitioning module fits the falling trajectory of rice into a trajectory surface and divides the trajectory surface into a collection area, a recognition area, and a separation area. An industrial camera is used to continuously collect images of the falling rice in the collection area. The image enhancement module performs local contrast adjustment on the acquired falling rice image based on the color difference enhancement model of generative adversarial network, and performs dynamic range adjustment on the acquired falling rice image in both RGB and HSI domains. The rice grain identification module establishes a standard color library for rice, uses a dynamic reference chromatogram bidirectional identification algorithm and a relative color difference mapping strategy to identify rice grains with different colors, and performs calibration within the identification area. At the same time, it constructs a reverse indexing mechanism for the characteristics of rice grains with different colors to perform bidirectional identification of rice grains with different colors and normal rice. The pneumatic separation module calculates the movement trajectory of the discolored rice grains in the separation zone based on the calibration results, and controls the opening timing of the airflow nozzles to blow the discolored rice grains into the discolored grain container. The reverse selection module, after separating the discolored rice grains of the current batch, returns the rice from the discolored grain container to the conveyor belt inlet, readjusts the vibration parameters applied to the conveyor belt by the vibration motor and adjusts the conveyor belt speed, and then uses the discolored grain feature reverse indexing mechanism to identify and separate the normal rice from the discolored grain rice.

Images