Glass cleaning and sorting system based on multi-mode perception

Abstract

The invention discloses a glass cleaning and sorting system based on multi-mode perception, and relates to the related technical field of waste glass recovery and reutilization, an acquisition module captures the color and transparency characteristics of waste glass through a spectrum camera, an infrared thermal imager measures the temperature of the surface of a mixture, an image processor collects and processes data, and the image processor obtains the color and transparency characteristics of the waste glass. The classification module uses a convolutional neural network to classify waste glass, the composite separation module performs preliminary screening through a magnetic field and then performs refining separation through a vibrating screen, and the impurity cleaning and drying module uses airflow generated by a high-pressure air nozzle in a closed gas injection chamber to blow away dust and small chippings. According to the invention, more comprehensive waste glass characteristic description is provided through the image processor by combining data of the spectrum camera and the thermal infrared imager, the misjudgment rate is reduced, the convolutional neural network is used for training and classifying the data, the method is suitable for various complex waste glass samples, and the detection accuracy is improved. And the accuracy and the stability of image recognition in the waste glass sorting process are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field related to waste glass recovery and reuse, and in particular to a glass cleaning and sorting system based on multimodal sensing. Background Art

[0002] With the acceleration of urbanization and the improvement of consumption level, the amount of waste glass generated has increased dramatically. As a recyclable resource, the effective recycling and treatment of waste glass is of great significance for saving resources and reducing environmental pollution.

[0003] However, the existing waste glass sorting system has obvious defects in the following aspects: 1. Image recognition technology and algorithms are not advanced enough: Traditional waste glass sorting systems mostly rely on basic image recognition technology, which is often unable to accurately distinguish different types of waste glass, especially in terms of subtle features such as color and transparency. The recognition accuracy is not high, which leads to low sorting efficiency and high error rate, affecting the subsequent reuse effect; 2. Single impurity removal means: In the existing system, the removal of impurities in waste glass mainly relies on physical screening and simple mechanical separation methods. These methods have a good effect on the removal of heavy metal impurities, but for non-metallic impurities, such as plastics, paper and other light impurities, the removal effect is not good, which not only affects the quality of waste glass, but also increases the difficulty and cost of subsequent processing; 3. Traditional wet cleaning process has high energy consumption and serious pollution: Although the traditional wet cleaning process can remove dirt and impurities on the surface of waste glass to a certain extent, it has high energy consumption, large water resource consumption, and the wastewater generated needs to be further treated, otherwise it will cause secondary pollution to the environment.

[0004] Therefore, it is necessary to propose a glass cleaning and sorting system based on multimodal perception to solve the above problems. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a glass cleaning and sorting system based on multimodal sensing, which solves the problems of low image recognition technology and algorithm recognition accuracy of the existing waste glass sorting system; single impurity removal means, especially poor removal effect of non-metallic impurities; high energy consumption and serious pollution of the wet cleaning process.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A glass cleaning and sorting system based on multimodal sensing, wherein the waste glass sorting comprises: Data acquisition module: including spectral camera and infrared thermal imager. The spectral camera is used to capture the color and transparency characteristics of waste glass, and the infrared thermal imager is used to measure the temperature distribution information on the surface of the waste glass and impurity mixture. The combination of these two data sources can provide a more comprehensive description of the characteristics of waste glass and transmit them to the image processor in real time for analysis and processing; Image processor: responsible for real-time data collection and processing, data preprocessing, feature extraction and data fusion of the color and transparency characteristics and temperature distribution information collected by the spectral camera and infrared thermal imager; Classification module: Using convolutional neural network (CNN) as the main framework, waste glass is classified according to the pre-trained data set to determine different types of waste glass (such as transparent glass and colored glass), and the rotary diverter is controlled to sort. The rotary diverter is controlled by the control system to rotate around the central axis based on the identified glass color information, thereby achieving accurate color classification; Composite separation module: used to transport the glass to the processing area, where it is guided by a magnetic field to complete the initial screening, and then further refined and separated by a vibrating screen to ensure that metal fragments and other large particles are effectively removed; Impurity cleaning and drying module: The conveyor belt delivers the waste glass after the classification and separation processes into a closed gas injection chamber. The chamber is equipped with several high-pressure gas nozzles. The airflow generated by the gas nozzles blows away the dust and small debris attached to it, and can also quickly evaporate the residual moisture, realizing a truly dry cleaning operation.

[0007] Optionally, the data collection method of the spectral camera is as follows: assuming that the camera resolution is R×C, where R represents the number of rows and C represents the number of columns, and each pixel records the light intensity value I(λi,x,y) within a specific wavelength range, where λi is the i-th wavelength band and (x,y) is the pixel coordinate; The data collection method of the infrared thermal imager is as follows: the output is a two-dimensional matrix T (x, y), which represents the temperature value corresponding to each pixel.

[0008] Optionally, the data preprocessing method in the image processor is: using an affine transformation model to perform image registration, aligning the images acquired by two different sensors to ensure that they correspond in space, applying a Gaussian filter to remove noise to reduce random noise in the image and improve data quality; Feature extraction: For spectral images, the Sobel operator is used to extract edge features, color and transparency features are extracted from spectral images, and temperature distribution features are extracted from infrared images; Data fusion: Apply the weighted average fusion rule to combine the features of the two data. Assume that the fused feature vector is F=[I,T], where I is the spectral data obtained from the spectral camera and T is the temperature data obtained from the infrared thermal imager. Post-processing and analysis: Further analysis is performed on the fused data, such as threshold segmentation or cluster analysis, to identify specific targets or areas.

[0009] Optionally, the convolutional neural network (CNN) consists of an input layer, a convolution layer, a pooling layer and a fully connected layer; Input layer: The input layer receives the fused feature vector F from the spectral camera and the infrared thermal imager, and the size is set to N×M; Convolutional layer: extract features from the image. The first convolutional layer has K filters, each of which is F×F in size. Pooling layer: reduces the size of the feature map and reduces the computational complexity. The pooling methods are maximum pooling and average pooling. Fully connected layer: The output of the previous layer is converted into the final classification result. The output of the fully connected layer is passed through the softmax function to obtain the probability distribution of each category.

[0010] Optionally, the classification module also includes a training unit and an optimization unit; The training unit is the loss function: the cross entropy loss function is used to evaluate the performance of the model; The optimization unit is an optimization algorithm: gradient descent is used to minimize the loss function.

[0011] Optionally, the composite separation module includes a magnetic field generator, a vibrating screen and a conveyor belt; Under the action of the magnetic field, the magnetic field generator attracts the metal impurities to the vicinity of the magnetic field source due to their magnetic properties, thereby achieving preliminary screening; The vibrating screen is driven by a motor to generate vibration, so that the material forms a jumping motion on the screen, so that particles of different sizes and densities are separated due to different forces. Large particles of debris are not easy to pass through the screen holes due to their large inertia, thereby achieving separation from small glass fragments. Assuming the vibration frequency of the vibrating screen is f, the amplitude is A, and the screen aperture is d, the movement of the material on the screen can be regarded as a simple harmonic vibration. According to Newton's second law, the resultant force F=ma (m is the mass of the material, a is the acceleration) on the material, combined with the parameters of the vibrating screen, a mathematical model of material movement can be established, and then the separation effect of the material can be analyzed.

[0012] Optionally, the composite separation module further includes a separation efficiency evaluation unit and an optimization suggestion unit; The separation efficiency evaluation unit is used to calculate the separation efficiency. The separation efficiency η is evaluated by comparing the mass concentration changes of the target substance in the feed port and the discharge port. The mathematical expression is: η=(C in -C out ) / C in ×100%, where C in and C outare the mass concentrations of the target substances in the feed inlet and the discharge outlet, respectively. By experimentally measuring these parameters, the performance of the composite separation device can be evaluated; The optimization suggestion unit optimizes the separation effect by adjusting the vibration frequency, amplitude and mesh aperture parameters of the vibrating screen according to the mathematical model and experimental results. For example, increasing the amplitude can improve the separation efficiency of large particles, but may reduce the passing rate of small glass fragments. Therefore, it is necessary to weigh various parameters to achieve the best separation effect.

[0013] Optionally, the impurity cleaning and drying module includes a closed gas injection chamber, a high-pressure gas nozzle, a high-pressure gas source system, an airflow control system, a programmable logic controller, a waste glass transmission system, and a dust and small debris collection system; The closed gas injection chamber is made of stainless steel and is equipped with a dust screen and a temperature sensor to ensure uniform distribution of airflow. High-pressure gas source system: The high-pressure air nozzle is made of aluminum alloy, the diameter of the air nozzle is ≤1mm, and the nozzle angle is adjusted by a motor; The high-pressure gas source system includes a compressor, a gas storage tank and a pressure regulating valve to ensure the supply of stable and sufficient high-pressure gas. The pressure regulating valve is used to adjust the pressure of the gas to meet different cleaning requirements; The airflow control system controls the opening and closing of the airflow through a solenoid valve.

[0014] The programmable logic controller (PLC) is used to control the operation of the entire system, including the intensity, timing and sequence of airflow; The waste glass transport system uses a conveyor belt to transport waste glass into and out of the gas injection chamber; The dust and small debris collection system comprises a filter, which is arranged below the gas injection chamber and is used to collect dust and small debris blown away by the airflow; the collection device is cleaned or replaced regularly to maintain the cleaning efficiency of the system.

[0015] Optionally, the impurity cleaning and drying module further includes a water evaporation auxiliary system, which is a microwave heater arranged in the gas injection chamber, and the microwave heater is used to evaporate the water in the waste glass.

[0016] The present invention provides a glass cleaning and sorting system based on multimodal sensing, which has the following beneficial effects: 1. The image processor of the present invention provides a more comprehensive description of waste glass characteristics by combining data from a spectral camera and an infrared thermal imager. This multimodal perception method significantly improves the accuracy of classification and reduces the misjudgment rate. A convolutional neural network is used to train and classify data. The deep learning model can learn complex feature representations from a large number of samples and adapt to various complex waste glass samples, significantly improving the accuracy and stability of image recognition in the waste glass sorting process.

[0017] 2. The composite separation device of the present invention significantly improves the sorting efficiency of waste glass through the two steps of preliminary screening and detailed separation. The preliminary screening under the action of the magnetic field can quickly remove metal impurities, and the vibrating screen further ensures the complete removal of other large particles of debris. This efficient separation process reduces manual intervention and improves the overall processing speed.

[0018] 3. The novel dry cleaning process based on the pneumatic principle of the present invention uses a strong airflow to blow away the dust and small debris on the surface of the waste glass, and uses a water evaporation auxiliary system to quickly evaporate the residual water to achieve efficient dry cleaning operations. This process not only improves the cleaning efficiency, but also ensures the cleanliness of the waste glass, avoids the waste of water resources and chemical pollution problems in traditional wet cleaning, and realizes a new dry cleaning process with low energy consumption and no pollution. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic diagram of the modules of the waste glass sorting system of the present invention; Figure 2 The figure is a schematic diagram of the working process of the waste glass sorting system of the present invention. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] Embodiment 1:

[0022] Please refer to Figure 1, a glass cleaning and sorting system based on multimodal sensing, waste glass sorting includes: Data acquisition module: including spectral camera and infrared thermal imager. The spectral camera is used to capture the color and transparency characteristics of waste glass, and the infrared thermal imager is used to measure the temperature distribution information on the surface of waste glass and impurity mixture. The combination of these two data sources can provide a more comprehensive description of the characteristics of waste glass and transmit them to the image processor in real time for analysis and processing; the data collection method of the spectral camera is: assuming that the camera resolution is R×C, where R represents the number of rows and C represents the number of columns, and each pixel point records the light intensity value I(λi,x,y) within a specific wavelength range, where λi is the i-th wavelength band and (x,y) is the pixel coordinate; the data collection method of the infrared thermal imager is: the output is a two-dimensional matrix T(x,y), which represents the temperature value corresponding to each pixel point; Image processor: responsible for real-time data collection and processing, data preprocessing, feature extraction and data fusion of the color and transparency features and temperature distribution information collected by the spectral camera and infrared thermal imager; the data preprocessing method in the image processor is: use the affine transformation model for image registration, align the images obtained by two different sensors to ensure that they correspond in space, apply Gaussian filter to remove noise and reduce random noise in the image to improve data quality; feature extraction: for spectral images, use the Sobel operator to extract edge features, extract color and transparency features from spectral images, and extract temperature distribution features from infrared images; data fusion: apply the weighted average fusion rule, combine the features of the two data, and assume that the fused feature vector is F=[I,T], where I is the spectral data obtained from the spectral camera and T is the temperature data obtained from the infrared thermal imager; post-processing and analysis: further analyze the fused data, perform threshold segmentation or cluster analysis to identify specific targets or areas; Classification module: Using convolutional neural network (CNN) as the main framework, waste glass is classified according to the pre-trained data set to determine different types of waste glass (such as transparent glass and colored glass), and the rotary diverter is controlled to sort. The rotary diverter is controlled by the control system to rotate around the central axis based on the identified glass color information, thereby achieving accurate color classification; Composite separation module: used to transport the glass to the processing area, where it is guided by a magnetic field to complete the initial screening, and then further refined and separated by a vibrating screen to ensure that metal fragments and other large particles are effectively removed; Impurity cleaning and drying module: The conveyor belt delivers the waste glass after the classification and separation process into the closed gas injection chamber. There are several high-pressure gas nozzles in the chamber. The airflow generated by the gas nozzles blows away the dust and small debris attached to it, and can also quickly evaporate the residual moisture, realizing a true dry cleaning operation; Convolutional neural network (CNN) consists of input layer, convolution layer, pooling layer and fully connected layer; Input layer: The input layer receives the fused feature vector F from the spectral camera and the infrared thermal imager, and the size is set to N×M; Convolutional layer: extract features from the image. The first convolutional layer has K filters, each filter size is F×F, and the convolution operation can be expressed as: , in, is the activation value at position i,j in layer l, is the weight parameter, is the bias term; Pooling layer: reduces the size of the feature map and reduces the computational complexity. The pooling methods are maximum pooling and average pooling. Fully connected layer: converts the output of the previous layer into the final classification result. The output of the fully connected layer is passed through the softmax function to obtain the probability distribution of each category. The classification module also includes a training unit and an optimization unit; The training unit is the loss function: The cross entropy loss function is used to evaluate the performance of the model. For multi-classification problems, the cross entropy loss function is defined as: , Where N is the number of samples, M is the number of categories, and y ic is the true label, p ic is the predicted probability; The optimization unit is an optimization algorithm: gradient descent is used to minimize the loss function.

[0023] In this embodiment, the image processor provides a more comprehensive description of waste glass characteristics by combining the data of the spectral camera and the infrared thermal imager. This multimodal perception method significantly improves the accuracy of classification, reduces the misjudgment rate, and significantly improves the accuracy and stability of image recognition in the waste glass sorting process. Convolutional neural networks (CNN) are used to train and classify data. The deep learning model can learn complex feature representations from a large number of samples and adapt to various complex waste glass samples, thereby improving the intelligence level of the system. The highly automated image processor reduces dependence on manual operations, reduces labor costs and the possibility of human errors, and improves the stability and reliability of the system.

[0024] Embodiment 2:

[0025] This embodiment is a further optimization based on the embodiment 1. Specifically, the composite separation module includes a magnetic field generator, a vibrating screen and a conveyor belt; Under the action of the magnetic field, the magnetic field generator attracts metal impurities to the vicinity of the magnetic field source due to their magnetic properties, thereby achieving preliminary screening. This process is based on the Lorentz force law, that is, the force of the magnetic field on the moving charge, causing the metal particles to be displaced in the magnetic field under the action of the magnetic force. Based on the Lorentz force law, when a conductor is placed in a changing magnetic field, it will be subject to a force perpendicular to the direction of the current and the magnetic field. Assuming the magnetic field strength is B, the conductor length is L, and the current is I, the force magnitude is: F=B×L×I; The vibrating screen is driven by a motor to generate vibration, so that the material forms a jumping motion on the screen, so that particles of different sizes and densities are separated due to different forces. Large particles of debris are difficult to pass through the screen holes due to their large inertia, thus achieving separation from small glass fragments. Assuming the vibration frequency of the vibrating screen is f, the amplitude is A, the time is t, and the screen aperture is d, the movement of the material on the screen can be regarded as simple harmonic vibration. According to Newton's second law, the resultant force F=ma (m is the mass of the material, a is the acceleration) on the material, combined with the parameters of the vibrating screen, a mathematical model of material movement can be established, and then the separation effect of the material can be analyzed. The mathematical model is expressed as: , Where y(t) represents the position offset at time t; The composite separation module also includes a separation efficiency evaluation unit and an optimization suggestion unit; The separation efficiency evaluation unit is used to calculate the separation efficiency. The separation efficiency η is evaluated by comparing the mass concentration changes of the target substance (such as metal impurities or large particles) in the feed port and the discharge port. The mathematical expression is: η=(C in -C out ) / C in ×100%, where C in and C out are the mass concentrations of the target substances in the feed inlet and the discharge outlet, respectively. By experimentally measuring these parameters, the performance of the composite separation device can be evaluated; Optimization suggestion unit: According to the mathematical model and experimental results, the separation effect is optimized by adjusting the vibration frequency, amplitude and mesh aperture parameters of the vibrating screen. For example, increasing the amplitude can improve the separation efficiency of large particles, but may reduce the passing rate of fine glass fragments. Therefore, it is necessary to weigh various parameters to achieve the best separation effect.

[0026] In this embodiment, the composite separation module significantly improves the sorting efficiency of waste glass through the two steps of preliminary screening and detailed separation. The preliminary screening under the action of the magnetic field can quickly remove metal impurities, and the vibrating screen further ensures the thorough removal of other large particles of debris. This efficient separation process reduces manual intervention and improves the overall processing speed. The dry cleaning process is used instead of the traditional wet cleaning process to avoid the large consumption of water resources and the generation of secondary sewage. This not only reduces the pollution to the environment, but also complies with the principle of sustainable development and helps to achieve the goal of green production. The highly automated separation system reduces dependence on manual operation and reduces labor costs. At the same time, due to the relatively simple equipment maintenance, the long-term operating costs are also effectively controlled. The design of the composite separation module takes into account the advantages of multiple physical separation technologies to form a stable and reliable multi-stage sorting system. This system can not only adapt to different types of waste glass, but also maintain stable operation in complex environments, ensuring the continuity and reliability of the production process.

[0027] Embodiment 3:

[0028] This embodiment is based on Example 1 or Example 2 and is optimized as follows. Specifically, the impurity cleaning and drying module includes a closed gas injection chamber, a high-pressure gas nozzle, a high-pressure gas source system, an air flow control system, a programmable logic controller, a waste glass transmission system, and a dust and small debris collection system; The closed gas injection chamber is made of stainless steel and is equipped with a dust screen and temperature sensor: The high-pressure air nozzle is made of aluminum alloy, the diameter of the air nozzle is ≤1mm, and the nozzle angle is adjusted by a motor; The high-pressure gas source system includes a compressor, a gas storage tank and a pressure regulating valve to ensure the supply of stable and sufficient high-pressure gas. The pressure regulating valve is used to adjust the gas pressure to meet different cleaning requirements; The airflow control system controls the opening and closing of the airflow through the solenoid valve; A programmable logic controller (PLC) is used to control the operation of the entire system, including the intensity, timing, and sequence of airflow; The waste glass transport system uses a conveyor belt to transport waste glass into and out of the gas injection chamber; The dust and small debris collection system includes a filter, which is arranged below the gas injection chamber to collect dust and small debris blown away by the air flow; the collection device is cleaned or replaced regularly to maintain the cleaning efficiency of the system; The impurity cleaning and drying module also includes a moisture evaporation auxiliary system, which is a microwave heater arranged in the gas injection chamber. The microwave heater is used to evaporate the moisture in the waste glass. Microwave heating technology is used instead of traditional electric heating methods to accelerate the evaporation rate of moisture in the waste glass, thereby shortening the overall processing time.

[0029] In this embodiment, the new dry cleaning process based on the pneumatic principle uses a strong airflow to blow away dust and small debris on the surface of the waste glass, and uses a water evaporation auxiliary system to quickly evaporate residual water, thereby achieving efficient dry cleaning operations. This process not only improves the cleaning efficiency, but also ensures the cleanliness of the waste glass. The new dry cleaning process uses a closed gas injection chamber and a high-pressure gas nozzle to avoid the waste of water resources and chemical pollution problems in traditional wet cleaning. In addition, the process uses the pneumatic principle to reduce energy consumption, meets the requirements of modern industry for environmental protection and energy saving, and realizes a new dry cleaning process with low energy consumption and no pollution. The process equipment is simple in design and easy to operate. The entire cleaning process can be completed by simply sending the waste glass into a closed gas injection chamber. It has a high degree of automation, which reduces the need for manual intervention, reduces operating difficulty and maintenance costs. The dry cleaning process based on pneumatic principles can quickly complete the cleaning of waste glass, shorten processing time and improve overall production efficiency. Since the process uses non-contact physical purification methods, it avoids potential safety hazards in traditional wet cleaning, such as chemical leakage and power failure. The application of pneumatic principles makes the entire process safer and more reliable, ensuring the safety of operators.

[0030] Please refer to FIG. 2 , the working process of the waste glass sorting system of the present invention is as follows: S1: Turn on the power of the spectral camera and infrared thermal imager, set parameters and calibrate the lens focal length. The spectral camera captures the color and transparency characteristics of the waste glass and transmits them to the image processor in real time for analysis and processing. Spectral cameras of different specifications can be replaced to adapt to waste glass samples of different sizes. The infrared thermal imager measures the temperature distribution information on the surface of the waste glass and impurity mixture; S2: Start the image processor, connect the communication line, collect and process data in real time, and perform data preprocessing, feature extraction and data fusion on the glass color and transparency characteristics and the temperature distribution information of the impurity mixture; S3: Load the deep learning model and use the convolutional neural network to classify waste glass based on the pre-trained dataset; S4: Control the rotary diverter for sorting. The rotary diverter is controlled by the control system to rotate around the central axis based on the identified glass color, transparency and temperature information, so as to achieve accurate classification; S5: Adjust the position and strength of the magnetic field generator to ensure that the entire processing area is covered. The magnetic field generator attracts metal impurities to the vicinity of the magnetic field source, thereby achieving preliminary screening. The layout and strength of the magnetic field generator are adjusted according to the cleaning situation of the metal impurities to improve the removal rate of metal impurities; S6: Set the working frequency and amplitude of the vibrating screen and check whether the conveyor belt is operating normally. The vibrating screen generates vibration through the motor drive, which makes the material form a jumping motion on the screen, so that particles of different sizes and densities are separated due to different forces. The design number of layers and aperture size of the vibrating screen can be changed to enhance the sorting effect of waste with specific particle size; S7: Open the closed gas injection chamber door, and the conveyor belt transfers the waste glass to the closed gas injection chamber; S8: Close the chamber door, turn on the compressor in the high-pressure gas source system, and make the gas nozzle produce strong airflow. Use the programmable logic controller to adjust the intensity of the airflow according to the cleaning situation of impurities. The high-pressure gas nozzles of different materials and shapes can be replaced to improve the airflow distribution and cleaning effect; S9: When finished, turn off all equipment and take out the clean waste glass.

[0031] In summary: The design of the glass cleaning and sorting system based on multimodal perception aims to solve the problems of low image recognition accuracy, insufficient impurity removal efficiency and serious environmental pollution in traditional sorting systems. The system improves the accuracy and robustness of waste glass sorting by constructing an image processor and classification module that combines a spectral camera with an AI algorithm; in addition, the automatic classification mechanism under the action of an electromagnetic field is integrated with other non-contact physical purification methods to form a composite separation device to ensure that metal fragments and other large particles of debris are effectively removed; finally, a new dry cleaning process based on pneumatic principles is developed to replace the traditional wet cleaning process, which reduces energy consumption and pollutant emissions; through the comprehensive upgrade and improvement of the waste glass sorting system, the goal of comprehensively improving the resource utilization rate of waste glass is finally achieved.

[0032] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A glass cleaning and sorting system based on multimodal sensing, characterized in that: The sorting system includes: Data acquisition module: including a spectral camera and an infrared thermal imager. The spectral camera is used to capture the color and transparency characteristics of waste glass, and the infrared thermal imager is used to measure the temperature distribution information on the surface of the mixture of waste glass and impurities, and transmit it to the image processor in real time; Image processor: responsible for real-time data collection and processing, data preprocessing, feature extraction and data fusion of the color and transparency characteristics and temperature distribution information collected by the spectral camera and infrared thermal imager; Classification module: Uses convolutional neural network as the main framework to classify waste glass according to the pre-trained data set to determine different types of waste glass and control the rotary diverter for sorting; Composite separation module: used to transport the glass to the processing area, where it is guided by a magnetic field to complete the initial screening, and then further refined and separated by a vibrating screen; Impurity cleaning and drying module: The conveyor belt delivers the waste glass after the classification and separation process into a closed gas injection chamber. The chamber is equipped with several high-pressure gas nozzles, and the airflow generated by the gas nozzles blows away the dust and small debris attached to it.

2. The glass cleaning and sorting system based on multimodal sensing according to claim 1, characterized in that: The data collection method of the spectral camera is as follows: assuming that the camera resolution is R×C, where R represents the number of rows and C represents the number of columns, each pixel records the light intensity value I(λi,x,y) within a specific wavelength range, where λi is the i-th wavelength band and (x,y) is the pixel coordinate; The data collection method of the infrared thermal imager is as follows: the output is a two-dimensional matrix T (x, y), which represents the temperature value corresponding to each pixel.

3. The glass cleaning and sorting system based on multimodal sensing according to claim 2, characterized in that: The data preprocessing method in the image processor is: using an affine transformation model to perform image registration, aligning the images acquired by two different sensors to ensure spatial correspondence, and applying a Gaussian filter to remove noise and reduce random noise in the image; Feature extraction: For spectral images, the Sobel operator is used to extract edge features, color and transparency features are extracted from spectral images, and temperature distribution features are extracted from infrared images; Data fusion: Apply the weighted average fusion rule to combine the features of the two data. Assume that the fused feature vector is F=[I,T], where I is the spectral data obtained from the spectral camera and T is the temperature data obtained from the infrared thermal imager. Post-processing and analysis: Further analysis is performed on the fused data, such as threshold segmentation or cluster analysis, to identify specific targets or areas.

4. The glass cleaning and sorting system based on multimodal sensing according to claim 3, characterized in that: The convolutional neural network consists of an input layer, a convolutional layer, a pooling layer and a fully connected layer; Input layer: The input layer receives the fused feature vector F from the spectral camera and the infrared thermal imager, and the size is set to N×M; Convolutional layer: extract features from the image. The first convolutional layer has K filters, each of which is F×F in size. Pooling layer: reduces the size of the feature map and reduces the computational complexity. The pooling methods are maximum pooling and average pooling. Fully connected layer: The output of the previous layer is converted into the final classification result. The output of the fully connected layer is passed through the softmax function to obtain the probability distribution of each category.

5. The glass cleaning and sorting system based on multimodal sensing according to claim 1, characterized in that: The classification module also includes a training unit and an optimization unit; The training unit is the loss function: the cross entropy loss function is used to evaluate the performance of the model; The optimization unit is an optimization algorithm: gradient descent is used to minimize the loss function.

6. The glass cleaning and sorting system based on multimodal sensing according to claim 1, characterized in that: The composite separation module includes a magnetic field generator, a vibrating screen and a conveyor belt; Under the action of the magnetic field, the magnetic field generator attracts the metal impurities to the vicinity of the magnetic field source due to their magnetic properties, thereby achieving preliminary screening; The vibrating screen is driven by a motor. Assuming the vibration frequency of the vibrating screen is f, the amplitude is A, and the mesh aperture is d, the movement of the material on the mesh is regarded as simple harmonic vibration. Combined with the parameters of the vibrating screen, a mathematical model of material movement is established to analyze the separation effect of the material.

7. The glass cleaning and sorting system based on multimodal sensing according to claim 6, characterized in that: The composite separation module also includes a separation efficiency evaluation unit and an optimization suggestion unit; The separation efficiency evaluation unit is used to calculate the separation efficiency. The separation efficiency η is evaluated by comparing the mass concentration changes of the target substance in the feed port and the discharge port. The mathematical expression is: η=(C in -C out ) / C in ×100%, where C in and C out are the mass concentrations of the target substances in the feed inlet and the discharge outlet, respectively; The optimization suggestion unit optimizes the separation effect by adjusting the vibration frequency, amplitude and mesh aperture parameters of the vibrating screen according to the mathematical model and experimental results.

8. The glass cleaning and sorting system based on multimodal sensing according to claim 1, characterized in that: The impurity cleaning and drying module includes a closed gas injection chamber, a high-pressure gas nozzle, a high-pressure gas source system, an airflow control system, a programmable logic controller, a waste glass transmission system, and a dust and small debris collection system; The closed gas injection chamber is made of stainless steel and is provided with a dustproof net and a temperature sensor inside; The high-pressure air nozzle is made of aluminum alloy, the diameter of the air nozzle is ≤1mm, and the nozzle angle is adjusted by a motor; The high-pressure gas source system includes a compressor, a gas storage tank and a pressure regulating valve, and the pressure regulating valve is used to adjust the pressure of the gas; The airflow control system controls the opening and closing of the airflow through a solenoid valve; The programmable logic controller is used to control the operation of the entire system, including the intensity, timing and sequence of airflow; The waste glass transport system uses a conveyor belt to transport waste glass into and out of the gas injection chamber; The dust and small debris collection system comprises a filter, which is arranged below the gas injection chamber and is used to collect dust and small debris blown away by the air flow.

9. The glass cleaning and sorting system based on multimodal sensing according to claim 8, characterized in that: The impurity cleaning and drying module also includes a water evaporation auxiliary system, which is a microwave heater arranged in the gas injection chamber, and the microwave heater is used to evaporate the water in the waste glass.

Images