Intelligent quartz sand impurity sorting system based on machine vision

Abstract

The present application relates to the technical field of quartz sand sorting, and particularly relates to a quartz sand impurity intelligent sorting system based on machine vision, which comprises: an illumination module for providing switchable illumination modes; an image acquisition module for acquiring images of quartz sand in a static state and a continuous motion state; an image analysis module, a first analysis unit for determining a suspected impurity area, a second analysis unit for adjusting the illumination mode, and a third analysis unit for determining the final impurity type and corresponding position information; and an execution control module for switching the illumination mode and performing directional sorting actions according to the impurity position. The present application overcomes the problem that it is difficult to automatically identify and sort multiple impurities in the continuous production process of quartz sand in the prior art.

Description

Technical Field

[0001] This invention relates to the field of quartz sand sorting technology, and more particularly to an intelligent quartz sand impurity sorting system based on machine vision. Background Technology

[0002] High-purity quartz sand is a key raw material for the photovoltaic, electronics, and high-end glass industries, and its quality depends heavily on the content of impurities. Traditional sorting methods mainly rely on manual visual inspection or machine vision systems based on static single light sources, which have significant limitations: manual sorting is inefficient and inconsistent; while static vision systems are insufficient in distinguishing impurities that are similar in color and texture to the quartz matrix, because under these fixed lighting conditions, the characteristic differences between impurities and quartz are not fully stimulated and captured.

[0003] While existing technologies employ color or multispectral cameras for detection, the illumination conditions are typically fixed. This approach fails to account for the drastically different reflection and absorption characteristics exhibited by various impurities under illumination at different wavelengths, angles, or polarization states due to variations in their material composition, crystal structure, and surface morphology. Therefore, a fixed illumination pattern may only be sensitive to a few types of impurities and cannot comprehensively and accurately cover the complex impurity spectrum that may exist in quartz sand.

[0004] Chinese Patent Publication No. CN121222693A discloses a rapid detection system and method for impurities in quartz sand, comprising: an image processing unit that periodically captures images of the conveyor belt as logistics images and determines the target area corresponding to each logistics image based on overlapping parts; an impurity identification unit that identifies the material contours in the target area, analyzes the impurity types corresponding to the material contours, records the material contours corresponding to the impurities as target contours, and determines their positions; and an air jet control unit that analyzes and determines the air jet timing of each air jet nozzle based on the position coordinates of the target contour when it is identified and marked, combined with the conveyor belt's conveying speed; while outputting the air jet timing corresponding to the target contour, it analyzes the shape characteristics of the target contour to obtain the projected area and thickness measurement values, and adjusts the airflow parameters of the air jet nozzles in conjunction with the corresponding impurity types. This prior art achieves an effective rapid detection and sorting scheme based on contours and shapes, but its core limitation lies in its relatively singular perception dimension, mainly relying on geometric features, which significantly reduces its detection rate, classification accuracy, and overall system adaptability when dealing with small, irregularly shaped, or novel impurities. Summary of the Invention

[0005] To address this issue, the present invention provides a machine vision-based intelligent sorting system for impurities in quartz sand, which overcomes the problem in the prior art that it is difficult to automatically identify and sort various impurities during the continuous production of quartz sand.

[0006] To achieve the above objectives, the present invention provides a machine vision-based intelligent sorting system for impurities in quartz sand, comprising: A lighting module for providing at least two switchable lighting modes to continuously moving quartz sand; An image acquisition module is used to acquire a first image of the quartz sand in a static state, and a second image of the quartz sand in continuous motion under the illumination mode. The image analysis module includes a first analysis unit for determining the color feature distribution of impurities based on the first image to determine an initial illumination pattern corresponding to at least one suspected impurity region; a second analysis unit for adjusting the initial illumination pattern based on the first reflectance brightness distribution features of the second image under the initial illumination pattern; and a third analysis unit for determining the final impurity type and corresponding location information based on the second reflectance brightness distribution features of the second image under the adjusted illumination pattern. The execution control module includes a first execution unit for controlling the switching of the lighting mode, and a second execution unit for performing directional sorting actions at the location of the corresponding impurity type in response to the impurity type.

[0007] Furthermore, the lighting modes of the lighting module include a first lighting mode that provides high-angle ring-shaped white light diffusion, a second lighting mode that provides lateral illumination of monochromatic light with a specific wavelength at an angle, and a third lighting mode that provides illumination with linear polarization characteristics.

[0008] Furthermore, the first analysis unit performs color space conversion on the first image and extracts color features to compare and analyze with a preset impurity color threshold to determine suspected impurity areas and corresponding lighting patterns.

[0009] Furthermore, the second analysis unit acquires the grayscale value of each pixel in the second image to generate the first reflective brightness distribution feature, and compares and analyzes the first reflective brightness distribution feature with the preset reflective brightness distribution feature to determine the brightness, angle, and wavelength of the illumination mode to be adjusted according to the deviation direction.

[0010] Furthermore, the third analysis unit includes a pre-trained image analysis model, which takes the second reflectance brightness distribution features of the second image under the adjusted lighting mode as input and outputs the category information of the impurities in the quartz sand.

[0011] Furthermore, the image analysis module also includes a model update unit, which is used to incrementally learn or periodically update the pre-trained image analysis model based on the impurity types and corresponding location information determined by the third analysis unit.

[0012] Furthermore, the second execution unit includes multiple high-speed air valves and a controller for controlling each of the high-speed air valves. The controller is communicatively connected to the image analysis module to control the corresponding high-speed air valve to open at a preset injection point according to the type and location information of the impurities, so as to blow out the quartz sand particles containing impurities from the main material flow and complete the directional sorting.

[0013] Furthermore, the image analysis module also includes a coordinate mapping unit, which is used to establish a mapping relationship between the pixel coordinates in the second image and the physical space coordinates of the second execution unit performing the sorting action, so as to accurately determine the timing of triggering the high-speed air valve according to the location information corresponding to the impurity type.

[0014] Furthermore, the lighting module includes a switchable multispectral light source array, which is composed of LED beads with different angles and wavelengths. Different combinations of LED beads are selectively lit by the control circuit to achieve rapid switching between the first lighting mode, the second lighting mode and the third lighting mode.

[0015] Furthermore, the illumination module also includes a polarizer rotation mechanism, which is disposed in the optical path of the third illumination mode to adjust the polarization direction of the linearly polarized light.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention utilizes a three-stage process of static color analysis, dynamic reflection feature analysis, and model judgment to comprehensively leverage the multi-dimensional visual features of impurities, such as color, surface texture, and reflection / polarization characteristics. Compared to traditional sorting methods that rely solely on a single grayscale or color threshold, this invention can more effectively distinguish impurities with similar optical characteristics to the background of quartz sand, significantly reducing the false positive and false negative rates.

[0017] Furthermore, the second analysis unit can provide real-time feedback and adjust the brightness, angle, or wavelength of the illumination mode based on the actual imaging effect of the initially identified suspected impurity areas under dynamic conditions. This closed-loop mechanism ensures optimal imaging conditions for the current batch of quartz sand and specific impurities, overcoming the problem of decreased imaging quality caused by material fluctuations, surface moisture, or dust under fixed illumination, thereby guaranteeing the reliability of subsequent analysis data.

[0018] Furthermore, this invention employs a high-speed air valve array as the actuator, achieving sorting through compressed air injection. This is a non-contact physical sorting method that will not cause mechanical damage to qualified quartz sand particles or introduce secondary pollution. Combined with the high-precision pixel coordinate and physical space coordinate mapping relationship established by the coordinate mapping unit, the system can control the air valve action at the corresponding position in real time and accurately, ensuring that the target impurity particles can be accurately blown out even during the high-speed movement of quartz sand, resulting in high sorting accuracy.

[0019] Furthermore, the second execution unit can respond to the judgment results of different impurity types. This means that the system can not only distinguish between impurities and good products, but also identify the specific type of impurities. This allows for the preset of different air valve injection forces or subsequent collection channels for different types of impurities, enabling more refined and customized sorting operations and improving the product grade of high-purity quartz sand.

[0020] Furthermore, this invention integrates a pre-trained image analysis model and a dedicated model update unit, enabling the system to continuously accumulate new impurity sample features through incremental learning. As the amount of material processed increases, the system's ability to identify complex, rare, or novel impurities will continuously improve, effectively addressing the challenges brought about by changes in mineral sources, extending the system's technical lifecycle, and reducing the cost of frequent manual readjustment due to changes in materials.

[0021] Furthermore, the present invention automates the entire process from lighting mode switching, image acquisition, analysis and judgment to sorting; operators only need to perform advanced management such as start / stop and parameter range setting, without having to manually check each frame of image or each suspected point, which greatly saves manpower and improves the continuous operation capability of the production line.

[0022] Furthermore, the rapidly switchable multispectral LED light source array and polarizer mechanism provide a powerful and flexible lighting tool. It can flexibly select and combine lighting modes according to different quartz sand ore properties and main impurity types, enabling a single system to adapt to a wider range of raw material processing needs, thereby improving the equipment's versatility and return on investment.

[0023] Furthermore, the functions of the lighting, acquisition, analysis, and control modules are clearly defined and the coupling is moderate. This design facilitates fault diagnosis, module replacement, and upgrades. For example, when upgrading the image analysis model, there is no need to modify the hardware actuator. The light source array uses LEDs, which have the advantages of long lifespan, low heat generation, and easy replacement.

[0024] Furthermore, high-precision sorting directly improves the chemical purity of the finished quartz sand, meeting the stringent quality requirements of high-value-added industries such as photovoltaics, electronics, and high-end glass for raw materials, thereby significantly enhancing product value and market competitiveness.

[0025] Furthermore, efficient sorting reduces the waste of high-quality quartz sand raw materials and improves the utilization rate of raw materials; at the same time, accurate sorting avoids misjudging a large amount of qualified materials as impurities and requiring further processing or disposal, indirectly reducing overall production energy consumption and material transportation costs.

[0026] Furthermore, compared to traditional methods that rely on flotation, acid washing, and other processes to remove impurities, this physical separation method basically does not involve the use of water, chemical agents, or acid, thus avoiding the generation of pollutants such as waste acid, wastewater, and tailings slurry from the source. It is a clean and environmentally friendly production technology that aligns with the development trend of green manufacturing. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating the workflow of the intelligent quartz sand impurity sorting system based on machine vision, as described in an embodiment of the present invention. Figure 2 This is a logic diagram of the first analysis unit of the intelligent quartz sand impurity sorting system based on machine vision, according to an embodiment of the present invention. Figure 3 This is a logic diagram of the second analysis unit of the intelligent quartz sand impurity sorting system based on machine vision, according to an embodiment of the present invention. Figure 4 This is a flowchart of the third analysis unit of the intelligent quartz sand impurity sorting system based on machine vision, according to an embodiment of the present invention. Detailed Implementation

[0028] 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.

[0029] 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.

[0030] Please see Figure 1 As shown, Figure 1 This is a structural diagram of the intelligent quartz sand impurity sorting system based on machine vision, according to an embodiment of the present invention.

[0031] This invention provides a machine vision-based intelligent sorting system for impurities in quartz sand, comprising: An illumination module is provided to provide at least two switchable illumination modes to continuously moving quartz sand to meet the needs of highlighting the characteristics of different impurities under different optical conditions. An image acquisition module is used to acquire a first image of the quartz sand in a static state, and a second image of the quartz sand in continuous motion under the illumination mode. The first image is a static material sample obtained by briefly stopping the production line conveyor belt or taking individual samples. It is mainly used for fine color space analysis. The second image is a picture of quartz sand taken during the operation of the quartz sand production line when the conveyor belt is moving at a continuous and uniform speed. This is the mainstream working state of the production line. Under this condition, the system completes real-time image acquisition, analysis and sorting.

[0032] The image analysis module includes a first analysis unit for determining the color feature distribution of impurities based on the first image to determine an initial illumination pattern corresponding to at least one suspected impurity region; a second analysis unit for adjusting the initial illumination pattern based on the first reflectance brightness distribution features of the second image under the initial illumination pattern; and a third analysis unit for determining the final impurity type and corresponding location information based on the second reflectance brightness distribution features of the second image under the adjusted illumination pattern. The execution control module includes a first execution unit for controlling the switching of the lighting mode, and a second execution unit for performing directional sorting actions at the location of the corresponding impurity type in response to the impurity type.

[0033] Specifically, the lighting modes of the lighting module include a first lighting mode that provides high-angle ring-shaped white light diffusion, a second lighting mode that provides lateral illumination of monochromatic light with a specific wavelength at an angle, and a third lighting mode that provides illumination with linear polarization characteristics.

[0034] Understandably, the first illumination mode can uniformly illuminate the sample, reduce shadows, and facilitate color analysis; the second illumination mode can enhance the surface texture and contour contrast of specific materials under specific lighting angles; and the third illumination mode can be used to suppress specular reflection of quartz sand matrix while highlighting the unique reflective properties of certain impurities under polarized light.

[0035] Please see Figure 2 As shown, it is the first analysis unit judgment logic diagram of the intelligent sorting system for quartz sand impurities based on machine vision in an embodiment of the present invention. The first analysis unit performs color space conversion on the first image and extracts color features to compare and analyze with a preset impurity color threshold to determine the suspected impurity area and the corresponding lighting mode. It is understood that in this embodiment, the color space is ultimately converted to the HSV color space, and three color features, namely hue, saturation, and brightness, are extracted. Preferably, the preset impurity color threshold can be set according to the type of impurity. For example, saturation greater than or equal to 0.3 and brightness less than 0.5 are set as hematite impurities, and brightness greater than or equal to 0.9 are set as organic impurities.

[0036] Specifically, the second analysis unit acquires the grayscale value of each pixel in the second image to generate the first reflective brightness distribution feature, and compares and analyzes the first reflective brightness distribution feature with the preset reflective brightness distribution feature to determine the brightness, angle, and wavelength of the lighting mode to be adjusted according to the deviation direction. Preferably, the preset reflective brightness distribution feature is 100. The first reflectance brightness distribution feature is obtained by calculating the gray-level variance of the suspected area, and the preset reflectance brightness distribution feature is obtained by calibrating various pure impurity samples under different lighting modes in the early stage. It is understandable that, in order to overcome the problem that the optical characteristics of impurities and quartz sand background are similar, this invention needs to focus on the differences in brightness and texture roughness within the impurity area or between the impurity and the surrounding background. Simply using the average gray value of the area cannot distinguish between a uniform dark spot and an area with light and dark texture. The gray variance is used to quantify the degree to which a set of data deviates from its average value, that is, the contrast or texture richness of the area. A large variance value indicates that the brightness and darkness of the area are drastic, and there may be edges, spots or special textures. A small variance value indicates that the brightness of the area is uniform. The calculation method for the first reflection brightness distribution characteristics is as follows: Where V represents the first reflection brightness distribution feature, N represents the total number of pixels in the suspected region, and g i Let be the grayscale value of the i-th pixel in the suspected region. This represents the average grayscale value of all pixels in the suspected region.

[0037] After comparing the first reflection brightness distribution characteristics with the preset reflection brightness distribution characteristics, three types of deviations are identified, and adjustments are made based on the corresponding deviation types, as follows: If the reflected brightness / contrast is lower than the preset value, the brightness is increased by 20%, and the illumination angle is optimized by adjusting the incident angle from 45° to 30° while keeping the wavelength unchanged. The measured first reflected brightness distribution characteristic value is lower than the preset value, resulting in insufficient contrast between impurities and background and indistinct features. Therefore, the system determines that the feature visibility needs to be enhanced by increasing the light intensity and adjusting the angle to strengthen the reflective properties of the hematite surface. The expected characteristic pattern was not observed, so the wavelength was adjusted to 450nm blue light, and the angle was significantly adjusted, with the side illumination angle reduced from 45° to 15°. Under the expected illumination mode, the specific reflection characteristics that the impurity should have were not captured. The system determined that the current illumination mode was not suitable for exciting the identification characteristics of the impurity, and it was necessary to switch or significantly adjust the illumination parameters to find the impurity characteristics. For mica impurities, the expected strong reflection characteristic peak did not appear at the initial side illumination angle. The system determined to capture the specular reflection of the mica sheet structure at a lower angle and a specific wavelength. The reflection characteristics do not match the conventional mode. The system determines that the third illumination mode, linearly polarized light, needs to be activated. The angle of the polarizer is rotated sequentially to 0°, 45°, 90°, and 135° to obtain the most obvious reflection characteristics. The reflection characteristics do not match those under common unpolarized light, which may involve smooth or anisotropic materials. The system determines that polarized light illumination needs to be enabled to suppress background reflection and highlight the optical anisotropy of impurities. For smooth organic materials, their reflection characteristics are difficult to distinguish under ordinary light. It is necessary to use polarization characteristics to distinguish impurities from the quartz sand matrix.

[0038] Specifically, the third analysis unit includes a pre-trained image analysis model, which takes the standardized reflectance brightness feature map corresponding to the suspected impurity region extracted from the second image acquired after optimized illumination as input, outputs the probability distribution of each suspected region belonging to each impurity category through forward calculation, and determines the category with the highest probability as the final category information of the impurity. In the image analysis model, the second reflectance brightness distribution feature is specifically represented as an image patch cropped from the original second image, containing only a single suspected impurity region. To further highlight the effect of the lighting optimization and unify the data scale, this image patch needs to be converted into a reflectance brightness feature map before being input into the image analysis model. The specific method is as follows: The color image patch is converted from the RGB color space to a grayscale image. The pixel values ​​of this grayscale image directly represent the reflectance distribution of that area under a specific optimized lighting mode. Subsequently, this grayscale image patch is normalized to normalize its pixel value range to the 0~1 interval.

[0039] All cropped grayscale image patches of suspected regions need to be uniformly scaled to a fixed size as the uniform input size for the image analysis model. In this embodiment, the fixed size is set to 224 pixels × 224 pixels. Therefore, each input sample is a tensor of size (224, 224, 1), representing the length, width, and number of channels.

[0040] The image analysis model employs a convolutional neural network architecture based on transfer learning, and the specific construction steps are as follows: The ResNet50 model, pre-trained on the large ImageNet dataset, was selected as the backbone network for feature extraction. This pre-trained model already possesses powerful general image feature extraction capabilities.

[0041] The network structure was modified by removing the global average pooling layer at the top of the original ResNet50 model and the subsequent fully connected layers. A new classification head suitable for this impurity classification task was then constructed after the backbone network, with the following structure: Layer 1 is the Flatten layer, which flattens the feature map output by the backbone network into a one-dimensional vector; Layer 2 is a fully connected layer with 1024 neurons. It uses the ReLU activation function, and its input dimension is the length of the flattened vector, while its output dimension is 1024. Layer 3 is a dropout layer. During training, 50% of the neuron outputs in layer 2 are randomly dropped with a probability of 0.5 to prevent overfitting. Layer 4 is a fully connected layer, which is the final classification output layer. The number of neurons in this layer is equal to the total number of impurity categories.

[0042] In this embodiment, the total number of impurity categories includes at least: hematite, mica, organic matter, feldspar, and background quartz sand. This layer uses the Softmax activation function to convert the output into a probability distribution for each category.

[0043] The error between the predicted probability distribution and the true label is calculated using the standard cross-entropy loss function used in classification tasks. The formula is as follows: Where L is the error between the predicted probability distribution and the true label, N is the number of samples in a batch, M is the total number of categories, and y i,c It is an indicator function, p = 1 when the true class of sample i is c, and 0 otherwise. i,c It is the probability that sample i belongs to category c as predicted by the model.

[0044] The Adam optimizer is used to minimize the loss function. The key training hyperparameter settings are as follows: The initial learning rate is set to e. -4 The batch size was set to 32, the number of training rounds was set to 50, and the optimization parameters used the default Adam values.

[0045] The prepared dataset is divided into training, validation, and test sets in a ratio of 70%, 15%, and 15%, respectively. On the training set, forward propagation is performed to calculate the loss and backpropagation to update the parameters. On the validation set, the loss and accuracy are monitored to adjust hyperparameters and determine the convergence of the model. Finally, the generalization performance of the image analysis model is evaluated on the test set.

[0046] After training, the optimal image analysis model weight file is saved. In the actual quartz sand sorting system, the integrated inference process is as follows: Receive the second image acquired by the second analysis unit after optimizing the lighting parameters; Locate the suspected impurity region in the image; For each suspected area, following the steps of preparing and standardizing the model input data described above, a (224, 224, 1) reflectance brightness feature map is generated; The feature map is then input into the loaded model for forward inference; The model outputs a probability vector of length equal to the total number of impurity categories.

[0047] The system selects the category with the highest probability value as the final classification category of the impurity, and sends this category information along with its location information to the second execution unit of the execution control module to trigger the corresponding directional sorting action.

[0048] The above supplements fully disclose the architecture, input / output format, key algorithm formulas, and training parameters of the image analysis model. Those skilled in the art can use the publicly available deep learning framework to collect or generate corresponding data based on this description, complete the construction, training, and integration of the image analysis model, and thus realize the technical solution claimed in this invention.

[0049] Understandably, this invention addresses complex classification problems that traditional algorithms struggle to handle by introducing an artificial intelligence model as the core of the final judgment. An image patch containing only a single suspected impurity, captured after optimized lighting, is input into a pre-trained model. The model automatically learns and identifies the complex visual patterns, including subtle textures, structures, and reflective properties, exhibited by different impurities such as hematite, mica, and organic matter under optimized lighting, and outputs probabilistic classification results. Compared to fixed rules, this method has stronger generalization ability and greater potential for identifying novel / complex impurities.

[0050] Specifically, the image analysis module further includes a model update unit, which is used to incrementally learn or periodically update the pre-trained image analysis model based on the impurity type and corresponding location information determined by the third analysis unit, so that the system can adapt to new raw materials or newly emerging impurity types and continuously improve the recognition accuracy.

[0051] The model update unit is triggered by the identification of a suspected impurity region in the third analysis unit. It selects the output sample of the image analysis model, packages the image patch of the sample, the optimized lighting mode parameters, and the category label determined by the model, and stores it in the incremental learning sample library. The deployed model is then retrained and optimized in an incremental learning manner, so that the system has the ability to continuously learn from actual production data and evolve itself.

[0052] The cropped image patch of the impurity region or its reflectance brightness feature map is used as the model input. After the image is processed by the network, a predicted category is obtained. By comparing the predicted result with the true result, the error is calculated using the cross-entropy loss function. Using an optimization algorithm, the loss error is backpropagated back to the network to update the network's weight parameters. The core objective of this process is to minimize the loss function, making the model's predictions increasingly accurate.

[0053] Understandably, the model update unit, through incremental learning, saves samples identified by the third analysis unit as new training data during actual operation. Periodically or after accumulating a certain amount, it retrains the deployed pre-trained model using this new data. This allows the model to continuously absorb new knowledge, optimize its decision boundaries, and gradually improve the accuracy of identifying rare impurities, newly emerging impurities, or impurities from specific mineral sources. This enables the system to adapt to changes in production conditions over a long period, extending its technological lifespan.

[0054] Specifically, the second execution unit includes multiple high-speed air valves and a controller for controlling each of the high-speed air valves. The controller is communicatively connected to the image analysis module to control the corresponding high-speed air valve to open at a preset injection point according to the type and location information of the impurities, so as to blow out the quartz sand particles containing impurities from the main material flow and complete the directional sorting. Preferably, the high-speed air valve is a solenoid valve with a response time of less than 5 milliseconds, and different air valve opening durations and air pressures are preset according to the type of impurities. For example, a higher air pressure of 0.4 MPa is used for heavy mineral impurities, and a lower air pressure of 0.25 MPa is used for light organic matter.

[0055] Understandably, using a high-speed solenoid valve array as the actuator offers advantages such as rapid action, high controllability, and no mechanical wear. After receiving impurity type and location commands from the image analysis module, the controller calls preset injection parameters based on the impurity type and triggers the corresponding air valve at a precisely calculated moment. Compressed air forms a short, directional airflow that blows the target impurity particles from the falling material stream into the foreign matter channel, while qualified materials fall naturally due to gravity, thus achieving non-destructive and efficient physical sorting and avoiding chemical contamination or mechanical damage.

[0056] Specifically, the image analysis module further includes a coordinate mapping unit, which is used to establish a mapping relationship between the pixel coordinates in the second image and the physical space coordinates of the second execution unit performing the sorting action, so as to accurately determine the timing of triggering the high-speed air valve based on the location information corresponding to the impurity type.

[0057] In this embodiment of the invention, the specific method for establishing the mapping relationship between pixel coordinates in the second image and the physical space coordinates of the second execution unit performing the sorting action is as follows: A standard checkerboard or dot array calibration plate with known accuracy is laid flat on the conveyor belt of the quartz sand transport, ensuring that it is flat and covers the main area of ​​the image acquisition module's field of view. In the first illumination mode, the image acquisition module acquires multiple images of the calibration board and obtains more data points by adjusting the position and angle of the calibration board; The Zhang Zhengyou calibration method is used to calculate the camera's intrinsic and extrinsic parameter matrices based on the world coordinates of feature points on the calibration board and their corresponding image pixel coordinates. The intrinsic parameter matrix includes camera parameters such as focal length and principal point; the extrinsic parameter matrix describes the positional relationship between the camera coordinate system and the conveyor belt's world coordinate system. Using the parameters mentioned above, a perspective transformation matrix can be determined from the two-dimensional pixel coordinates of the image to the two-dimensional physical coordinates of the conveyor belt plane. This matrix is ​​the core mapping relationship and is stored in the software of the coordinate mapping unit.

[0058] After the third analysis unit identifies the impurity and provides its image pixel coordinates, the coordinate mapping unit uses the stored homography matrix to obtain the real-time physical coordinates of the impurity on the conveyor belt plane through the perspective transformation matrix.

[0059] Understandably, by pre-calibrating the image acquisition system using a calibration board, the camera's intrinsic and extrinsic parameters can be accurately calculated, thereby establishing a mathematical model from image pixel coordinates to the physical coordinates of the conveyor belt plane. Once the system identifies the pixel position of an impurity in the image, the coordinate mapping unit can use this model to calculate the impurity's actual physical position on the conveyor belt in real time, ensuring that the high-speed air valve is triggered at the correct position and time, achieving millimeter-level sorting accuracy.

[0060] Specifically, the lighting module includes a switchable multispectral light source array composed of LED beads with different angles and wavelengths. A control circuit selectively illuminates different combinations of LED beads to achieve rapid switching between the first, second, and third lighting modes. The first lighting mode illuminates all the LED beads in the ring-shaped white LED array; the second lighting mode illuminates a combination of lateral LED beads with specific wavelengths and angles; and the third lighting mode illuminates the base light source used for polarization mode while a polarizer rotation mechanism simultaneously rotates the polarizer to a preset or optimized angle. Preferably, the lighting module includes LED beads emitting white light, 450nm blue light, 630nm red light, and 850nm infrared light.

[0061] Understandably, the light source array is composed of LED beads with different wavelengths and different installation angles. By selectively illuminating different combinations of LED beads through the control circuit, different lighting modes can be switched to meet the needs of complex impurity detection.

[0062] Specifically, the illumination module further includes a polarizer rotation mechanism, which is disposed in the optical path of the third illumination mode to adjust the polarization direction of the linearly polarized light; When the second analysis unit determines that the third illumination mode needs to be activated for a certain type of impurity, the motor rotates to the corresponding preset angle to adjust the polarization direction of the linearly polarized light. The preset angles include 0°, 45°, 90°, and 135°.

[0063] Understandably, when the third illumination mode is activated, the light emitted by the light source is first converted into linearly polarized light by a fixed polarizer; the polarizer rotation mechanism set in the optical path can precisely rotate the angle of the polarizer, and by changing the polarization direction, the reflection difference between impurities and background materials under different polarization states can be determined, thereby finding the best polarization angle that can suppress background glare while highlighting the characteristics of impurities.

[0064] Example 1:

[0065] In this embodiment, the quartz sand raw material contains hematite impurities, which are typically dark red or brownish-black in color and have a high color contrast with the quartz sand matrix, making it easy to perform preliminary screening by color. However, due to the presence of uneven oxide layers or deposits on its surface, its reflectivity under fixed lighting conditions is not stable, which may lead to misjudgments by traditional sorting systems based on a single image.

[0066] After system startup, the image acquisition module acquires the first image of a stationary quartz sand sample under the default first illumination mode. The first analysis unit performs HSV color space conversion on this image and sets thresholds for comparison based on the typical color characteristics of hematite, with saturation greater than 0.3 and brightness less than 0.5. After analysis, the system successfully locates several suspected impurity areas whose color characteristics match those of hematite. Based on this color analysis result, the system decides to use the second illumination mode for further dynamic feature analysis to stimulate its surface reflective properties.

[0067] The image acquisition module switches to the second illumination mode to acquire a second image of the continuously moving quartz sand flow. The second analysis unit analyzes this image, extracts the pixel grayscale values ​​of suspected impurity areas, and generates a first reflection brightness distribution feature. The calculated feature value is 400, lower than the preset reflection brightness threshold for hematite. The system determines that the current illumination fails to achieve optimal imaging contrast for the impurities, and therefore generates optimization instructions: adjust the monochromatic light wavelength to 630nm, closer to the deep red absorption and reflection characteristics of hematite, reduce the lateral incident angle to 30°, and increase the overall illumination brightness by 20%. The first execution unit of the illumination module receives the instructions and completes the real-time adjustment of the illumination parameters.

[0068] Under the optimized lighting parameters, the image acquisition module re-acquires the second image of the moving material. The third analysis unit takes the second reflection brightness distribution characteristics of the target area in the image as input, performs the final analysis, and outputs the judgment result as iron impurity, and gives the precise image coordinates x=120px, y=85px.

[0069] The coordinate mapping unit converts the image coordinates into physical space coordinates (x=6.0mm, y=4.25mm) based on the pre-calibrated parameters 1px=0.05mm. The second execution unit, according to these coordinates and the type of impurity, controls the high-speed air valve at the corresponding position, activating it precisely at the moment the target particle reaches the injection point. Considering the high density of ferrous impurities, the system employs a higher injection air pressure of 0.4MPa and an appropriate injection duration of 20ms to ensure effective removal of the impurities from the main material flow.

[0070] When applied to a production line with a processing capacity of 200 tons / hour, the system achieved a sorting efficiency of 99.2% for hematite impurities, representing an improvement of approximately 14.2% in sorting accuracy compared to traditional mechanical screening or magnetic separation methods. The system operates fully automatically, and no performance degradation was observed after 72 hours of continuous operation.

[0071] Example 2:

[0072] In this embodiment, the quartz sand raw material contains organic impurities such as plant fibers and resin particles. These impurities typically have smooth surfaces and high reflectivity under broadband white light illumination, and their color may be very similar to that of the quartz sand itself. Under uniform diffuse illumination conditions, their grayscale and texture differences from qualified quartz sand particles are very slight, making them easily missed by traditional vision systems.

[0073] After the system is started, the image acquisition module acquires the first image of the static sample in the first illumination mode. When the first analysis unit performs HSV color space analysis, it captures the brightness values ​​of some areas with a value greater than 0.9 by analyzing the highlight areas in the image. This is a typical feature of smooth organic surfaces. These highlight areas are marked as suspected organic impurity areas, and the third illumination mode is started for in-depth identification.

[0074] The image acquisition module first acquires images of the moving material in the initial mode. The second analysis unit analyzes the image and finds that, under normal lighting, although the first reflection brightness distribution characteristics of the suspected area are relatively bright, they lack a clearly identifiable feature pattern, and the feature values ​​do not match the preset organic polarization reflection model. The system switches to the third lighting mode and adjusts the polarization direction of the polarizer to 45°.

[0075] Under polarized light, the smooth surface of non-metallic organic materials changes the polarization state of the reflected light, resulting in a significant difference in reflection characteristics compared to the quartz sand matrix. The system then re-acquires a second image. The third analysis unit analyzes the image and successfully determines the target impurity type as an organic impurity based on the unique characteristics under polarized light imaging, outputting its image coordinates x=205px, y=110px.

[0076] The coordinate mapping unit converts the coordinates into physical positions x=6.15mm and y=3.3mm based on the mapping parameter 1px=0.03mm. The second execution unit receives the type and coordinate information of the organic impurities. Since the organic impurities are light, the system uses a relatively low air pressure of 0.25MPa and short pulses for injection, which can accurately remove them from the material flow, effectively avoiding the situation where qualified quartz sand is accidentally blown out due to excessive impact force.

[0077] The system was applied to a high-purity quartz sand production line with a processing capacity of 150 tons / hour, achieving a detection rate of over 98.5% for sorting organic impurities such as plant fibers and resins. Compared to traditional optical sorting methods that rely on color or a single grayscale threshold, this system, by introducing a polarized light mode, improves the identification accuracy of organic impurities with smooth surfaces and similar colors by more than 20%. Simultaneously, because it can accurately distinguish between impurities and good products, the raw material utilization rate is increased by approximately 2.5%, effectively reducing the waste of qualified materials. The system operates stably, achieving efficient and automated sorting of low-contrast impurities.

[0078] Example 3:

[0079] In this embodiment, the quartz sand raw material contains mica-like flaky mineral impurities, whose color is similar to that of the quartz sand matrix. However, under specific angle lighting, they will produce strong specular reflections, and under uniform diffused light, they are difficult to distinguish from the background. Traditional sorting methods are prone to missing detection.

[0080] After the system is started, the image acquisition module acquires the first image of the stationary quartz sand in the initial first illumination mode; the first analysis unit performs HSV color space conversion and analysis on the first image and finds that the color difference between mica and quartz is slight. Based on the color threshold alone, the saturation is less than 0.1 and the brightness is 0.7. Several suspected impurity areas with questionable reflective properties are preliminarily identified, and it is determined that the second illumination mode, which can highlight the surface texture and angular reflection characteristics, needs to be activated for further identification.

[0081] The image acquisition module acquires a second image of the quartz sand in continuous motion under the initial second illumination mode. The second analysis unit analyzes this image, obtaining the grayscale values ​​of each pixel to generate the first reflection brightness distribution characteristics. Analysis reveals that, at the preset incident angle, the suspected area does not exhibit the expected strong reflection characteristic peak, and its distribution characteristics deviate from the preset mica reflection model. Therefore, the second analysis unit sends out adjustment instructions: switching the illumination mode to the optimized parameters of the second illumination mode, specifically adjusting the monochromatic light wavelength to 450nm blue light, which is more easily absorbed by mica and produces characteristic reflections, and adjusting the side illumination angle from 45° to 15°; the first execution unit of the illumination module completes the rapid switching of the illumination mode according to the instructions.

[0082] Under the adjusted and optimized lighting mode, the system re-acquired the second image. The second reflection brightness distribution feature generated by the second analysis unit showed that a bright reflection band consistent with mica characteristics appeared in the previously identified suspected area. The third analysis unit used this second reflection brightness distribution feature as input for analysis and judgment, and finally output that the impurity category was mica, and gave its precise location coordinates in the image: x=150px, y=200px.

[0083] The coordinate mapping unit in the image analysis module maps the pixel coordinates to physical space coordinates (x=6.0mm, y=8.0mm) based on preset calibration parameters (1px=0.04mm). The second execution unit of the execution control module, based on this coordinate information and the type of impurity, controls the opening of the high-speed air valve at the corresponding physical position the instant the quartz sand particle reaches that position. Because mica is lightweight and flaky, the system presets a relatively low injection air pressure of 0.3MPa and a short pulse of 15ms to precisely blow it from the main material stream to the foreign matter collection channel, completing the sorting process.

[0084] In actual production testing, the system achieved a sorting efficiency of up to 99.0% for mica impurities with a particle size greater than 75 micrometers. By employing optimizable low-angle side lighting, the system successfully captured the unique specular reflection characteristics of mica, reducing the false negative rate of this type of impurity by more than 15% compared to a fixed-lighting system. When continuously processing materials from different mineral sources with fluctuating reflectivity, the system maintained stable sorting accuracy with a fluctuation range of less than ±0.5% thanks to its adaptive lighting parameter adjustment mechanism, significantly improving the quality consistency of the final quartz sand product in photovoltaic and high-end glass applications.

[0085] 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 intelligent sorting system for impurities in quartz sand, characterized in that, include: A lighting module for providing at least two switchable lighting modes to continuously moving quartz sand; An image acquisition module is used to acquire a first image of the quartz sand in a static state, and a second image of the quartz sand in continuous motion under the illumination mode. The image analysis module includes a first analysis unit for determining the color feature distribution of impurities based on the first image to determine an initial illumination pattern corresponding to at least one suspected impurity region; a second analysis unit for adjusting the initial illumination pattern based on the first reflectance brightness distribution features of the second image under the initial illumination pattern; and a third analysis unit for determining the final impurity type and corresponding location information based on the second reflectance brightness distribution features of the second image under the adjusted illumination pattern. The execution control module includes a first execution unit for controlling the switching of the lighting mode, and a second execution unit for performing directional sorting actions at the location of the corresponding impurity type in response to the impurity type.

2. The intelligent quartz sand impurity sorting system based on machine vision according to claim 1, characterized in that, The lighting modules include a first lighting mode that provides high-angle ring-shaped white light diffusion, a second lighting mode that provides lateral illumination of monochromatic light of a specific wavelength, and a third lighting mode that provides illumination with linear polarization characteristics.

3. The intelligent quartz sand impurity sorting system based on machine vision according to claim 2, characterized in that, The first analysis unit performs color space conversion on the first image and extracts color features to compare and analyze with a preset impurity color threshold to determine suspected impurity areas and corresponding lighting patterns.

4. The intelligent quartz sand impurity sorting system based on machine vision according to claim 3, characterized in that, The second analysis unit acquires the grayscale value of each pixel in the second image to generate the first reflective brightness distribution feature, and compares and analyzes the first reflective brightness distribution feature with the preset reflective brightness distribution feature to determine the brightness, angle, and wavelength of the illumination mode to be adjusted according to the deviation direction.

5. The intelligent quartz sand impurity sorting system based on machine vision according to claim 4, characterized in that, The third analysis unit includes a pre-trained image analysis model, which takes the second reflectance brightness distribution features of the second image under the adjusted lighting mode as input and outputs the category information of the impurities in the quartz sand.

6. The intelligent quartz sand impurity sorting system based on machine vision according to claim 5, characterized in that, The image analysis module further includes a model update unit, which is used to incrementally learn or periodically update the pre-trained image analysis model based on the impurity types and corresponding location information determined by the third analysis unit.

7. The intelligent quartz sand impurity sorting system based on machine vision according to claim 6, characterized in that, The second execution unit includes multiple high-speed air valves and a controller for controlling each of the high-speed air valves. The controller is communicatively connected to the image analysis module to control the corresponding high-speed air valve to open at a preset injection point according to the type and location information of the impurities, so as to blow out the quartz sand particles containing impurities from the main material flow and complete the directional sorting.

8. The intelligent quartz sand impurity sorting system based on machine vision according to claim 7, characterized in that, The image analysis module also includes a coordinate mapping unit, which is used to establish a mapping relationship between the pixel coordinates in the second image and the physical space coordinates of the second execution unit when performing the sorting action, so as to accurately determine the timing of triggering the high-speed air valve according to the location information corresponding to the impurity type.

9. The intelligent quartz sand impurity sorting system based on machine vision according to claim 8, characterized in that, The lighting module includes a switchable multispectral light source array, which is composed of LED beads with different angles and wavelengths. Different combinations of LED beads are selectively lit by the control circuit to achieve rapid switching between the first lighting mode, the second lighting mode and the third lighting mode.

10. The intelligent quartz sand impurity sorting system based on machine vision according to claim 9, characterized in that, The illumination module also includes a polarizer rotation mechanism, which is disposed in the optical path of the third illumination mode to adjust the polarization direction of linearly polarized light.

Images