Multi-category plastic automatic backflow sorting method and system and storage medium

Abstract

The invention provides a multi-category plastic automatic backflow sorting method and system and a storage medium. The method comprises the steps that point cloud data of bottle bodies are obtained, and a laser triangulation method is used for processing the point cloud data to obtain bottle body morphological characteristics of the bottle bodies; acquiring initial spectrum information of the bottle body and working temperature and humidity of the working environment, judging whether spectrum compensation is needed or not based on the working temperature and humidity and the initial spectrum information, and if yes, substituting the working temperature and humidity and / or the initial spectrum information into a preset compensation model to obtain spectrum compensation information, determining actual spectral information of the bottle body based on the initial spectral information and the spectral compensation information; and the bottle body morphological characteristics and the actual spectral information are substituted into a mixed neural network model to obtain the plastic types of the bottle bodies and the corresponding confidence degrees, the confidence levels corresponding to the confidence degrees are determined, and corresponding sorting instructions are generated based on the confidence levels to achieve bottle body sorting. According to the plastic sorting device, sorting work of multiple types of plastics can be efficiently and accurately achieved.

Description

Technical Field

[0001] The present application relates to the field of sorting technology, and in particular to a method, system and storage medium for automated reflow sorting of multiple categories of plastics. Background Art

[0002] With the widespread use of plastic products in daily life and industrial production, the recycling and reuse of waste plastics has become a key link in solving resource waste and environmental pollution. However, plastics come in many different types and categories, and their physical and chemical properties vary significantly, requiring the sorting and recycling of recyclable waste plastics.

[0003] Currently, due to the low efficiency and high labor costs of manual sorting, it has gradually been replaced by automated sorting equipment. Automated sorting equipment typically uses infrared spectroscopy or near-infrared sorting methods to sort plastics. While this improves sorting efficiency compared to manual sorting, it can be affected by factors such as plastic color and surface contamination, resulting in lower accuracy. Therefore, existing methods are unable to achieve efficient and accurate plastic sorting. Summary of the Invention

[0004] In order to efficiently and accurately implement the sorting of multiple categories of plastics, the embodiments of the present application provide a method, system and storage medium for automated reflow sorting of multiple categories of plastics.

[0005] In a first aspect, this embodiment provides a method for automated reflow sorting of multiple categories of plastics, the method comprising: Acquiring point cloud data of the bottle body, and processing the point cloud data using a laser triangulation method to obtain bottle body morphological features of the bottle body; Obtaining the initial spectrum information of the bottle and the working temperature and humidity of the working environment, and determining whether spectrum compensation is required based on the working temperature and humidity and the initial spectrum information, If necessary, substituting the working temperature and humidity and / or the initial spectrum information into a preset compensation model to obtain spectrum compensation information, and determining the actual spectrum information of the bottle based on the initial spectrum information and the spectrum compensation information; The bottle morphological characteristics and the actual spectral information are substituted into a hybrid neural network model to obtain the plastic type of the bottle and the corresponding confidence level, determine the confidence level corresponding to the confidence level, and generate corresponding sorting instructions based on the confidence level to achieve sorting of the bottles.

[0006] In some embodiments, the method is applied to a sorting device comprising three lasers, wherein the three lasers are evenly distributed in an annular pattern on the sorting device, and obtaining point cloud data of the bottle includes: Point cloud data of the bottle body is synchronously acquired based on the trigger signal sent by the sorting device.

[0007] In some embodiments, the preset compensation model includes a temperature and humidity layer compensation model and a pollution layer compensation model, the spectral compensation information includes temperature and humidity layer spectral compensation information and / or pollution layer spectral compensation information, and substituting the operating temperature and humidity and / or the initial spectral information into the preset compensation model to obtain the spectral compensation information includes: Substituting the working temperature and humidity into the temperature and humidity layer compensation model to obtain temperature and humidity layer spectrum compensation information; And / or, the initial spectrum information is substituted into a pollution layer compensation model to obtain pollution layer spectrum compensation information.

[0008] In some embodiments, the sorting device includes a conveyor belt, and a set of standard reflective plates are installed on both sides of the conveyor belt, each set of standard reflective plates includes a white plate, and the method further includes: Obtaining the actual reflectivity of the whiteboard at every preset time interval, determining whether the actual reflectivity deviation between the actual reflectivity and the set reflectivity of the whiteboard is greater than the trigger compensation reflection deviation, and if not, generating a reflective plate no deviation instruction; If it is greater, spectrum compensation information of the reflector is generated based on the actual reflectivity deviation, and the spectrum compensation information is used to update the actual spectrum information.

[0009] In some embodiments, the hybrid neural network model is a CNN-Transformer hybrid neural network, and substituting the bottle morphological characteristics and the actual spectral information into the hybrid neural network model to obtain the plastic type of the bottle and the corresponding confidence level includes: Using a three-layer CNN to process the bottle morphological features to obtain first intermediate output information; Performing attention weighting on the actual spectral information through a Transformer to obtain second intermediate output information; Performing tensor splicing on the first intermediate output information and the second intermediate output information to obtain information to be processed; The information to be processed is processed using a first loss function to obtain the plastic type of the bottle and the corresponding confidence level.

[0010] In some embodiments, the method further comprises: The information to be processed is processed using a second loss function to obtain the pollution type and pollution degree value of the bottle.

[0011] In some embodiments, generating corresponding sorting instructions based on the confidence level to implement sorting of the bottles includes: If the confidence level is a direct sorting level, determining the working information of the nozzle based on the morphological characteristics of the bottle, and generating a sorting instruction corresponding to the working information to implement sorting of the bottle; If the confidence level is a reflux sorting level, generating a reflux sorting instruction corresponding to a reflux channel to open the reflux channel to implement sorting of the bottles; If the confidence level is an intervention sorting level, an intervention sorting instruction corresponding to the unknown type is generated to implement sorting of the bottles.

[0012] In some embodiments, the sorting device includes a hyperspectral camera, and obtaining initial spectral information of the bottle includes: The hyperspectral camera synchronously acquires initial spectral information of the bottle body based on the trigger signal sent by the sorting device.

[0013] In a second aspect, this embodiment provides a multi-category plastic automated return sorting system, the system comprising: an information acquisition module, an information processing module and a plastic sorting module; wherein, The information acquisition module is used to obtain point cloud data of the bottle body, process the point cloud data using laser triangulation to obtain the bottle body morphological characteristics of the bottle body, and obtain initial spectral information of the bottle body and the working temperature and humidity of the working environment; The information processing module is configured to determine whether spectrum compensation is required based on the operating temperature and humidity and the initial spectrum information; if required, substitute the operating temperature and humidity and / or the initial spectrum information into a preset compensation model to obtain spectrum compensation information; and determine actual spectrum information of the bottle based on the initial spectrum information and the spectrum compensation information; The plastic sorting module is used to substitute the bottle morphological characteristics and the actual spectral information into the hybrid neural network model to obtain the plastic type of the bottle and the corresponding confidence level, determine the confidence level corresponding to the confidence level, and generate corresponding sorting instructions based on the confidence level to achieve sorting of the bottles.

[0014] In a third aspect, this embodiment provides a computer-readable storage medium storing a computer program that can be run on a processor. When the computer program is executed by the processor, it implements a multi-category plastic automated reflow sorting method as described in the first aspect.

[0015] By employing the above-described method, the present application first obtains point cloud data of the bottle, then processes the point cloud data using laser triangulation to obtain the bottle morphological characteristics of the bottle. Initial spectral information of the bottle and the operating temperature and humidity of the working environment are obtained. Based on the operating temperature and humidity and initial spectral information, a determination is made as to whether spectral compensation is required. If necessary, the operating temperature and humidity and / or the initial spectral information are substituted into a preset compensation model to obtain spectral compensation information. The actual spectral information of the bottle is then determined based on the initial spectral information and the spectral compensation information. By compensating the detected information for bottle contamination, actual spectral information that more accurately reflects the bottle material can be obtained, providing accurate information guidance for subsequent bottle sorting operations.

[0016] The bottle morphological characteristics and actual spectral information are then substituted into the hybrid neural network model to obtain the plastic type of the bottle and the corresponding confidence level.

[0017] Finally, the confidence level corresponding to the confidence degree is determined, and based on this confidence level, corresponding sorting instructions are generated to sort the bottles. This approach, by using a hybrid neural network to specifically process the actual spectral information, coupled with the high accuracy of the actual spectral information, enables the hybrid neural network model to output a more accurate prediction of the bottle's plastic type and the corresponding confidence level. This, combined with the automated sorting process performed by the sorting device, enables efficient sorting. Ultimately, this allows for efficient and accurate sorting of multiple plastic categories. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a block diagram of a multi-category plastic automated reflow sorting method provided by this application.

[0019] Figure 2 This is a block diagram of a method provided by the present application for substituting working temperature and humidity and / or initial spectral information into a preset compensation model to obtain spectral compensation information.

[0020] Figure 3 This is a block diagram of a method provided by the present application for substituting bottle morphological characteristics and actual spectral information into a hybrid neural network model to obtain the plastic type of the bottle and the corresponding confidence level.

[0021] Figure 4 This is a block diagram of the method disclosed in this application for generating corresponding sorting instructions based on confidence levels to achieve bottle sorting.

[0022] Figure 5 This is a connection diagram of a multi-category plastic automated reflow sorting system provided by this application. DETAILED DESCRIPTION

[0023] In order to more clearly understand the purpose, technical solutions and advantages of the present application, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments. However, it should be understood by those skilled in the art that the present application can be implemented without these details. It is obvious to those skilled in the art that various changes can be made to the embodiments disclosed in the present application, and the general principles defined in the present application can be applied to other embodiments and application scenarios without departing from the principles and scope of the present application. Therefore, the present application is not limited to the embodiments shown, but conforms to the broadest scope consistent with the scope claimed in the present application.

[0024] The embodiments of the present application are described in further detail below with reference to the accompanying drawings.

[0025] The specific application scenario of this application is a sorting device. The sorting device includes a main conveyor belt, a laser, a hyperspectral camera, a microwave sensor, an air valve array, a spiral elevator, an environmental sensor, an auxiliary conveyor belt and a sorting control terminal. Among them, the main conveyor belt is a conveyor belt with a width of 800mm and an adjustable speed range of 0.5-4m / s; the laser is a 3D linear array camera with a resolution of 2048*2048 pixels and a frame rate of 120fps. There are three lasers, which are installed 1.2m above the main conveyor belt at an inclination angle of 30° and at an angle of 75° to the direction of material movement to ensure that each bottle scans at least 200 spectral lines; the hyperspectral camera is a camera with a band of 900-1700nm, a spectral resolution of 10nm, and a scanning line frequency of 200Hz. The hyperspectral camera is a six-band InGaAs linear array camera, equipped with a narrowband filter. The light sheet wheel realizes rapid switching of six characteristic bands: 905nm, 1070nm, 1215nm, 1300nm, 1450nm and 1550nm; the air valve array consists of five groups with a spacing of 30cm. The nozzle adopts a fan-shaped diffusion design with a nozzle diameter of 8mm, a working air pressure of 0.6MPa and a coverage width of 1.2m to ensure that large-mass bottles can be effectively sorted; the lifting height of the spiral elevator is 3m and the processing capacity is 8 tons / hour; environmental sensors include temperature and humidity sensors; the auxiliary conveyor belt is used to transport objects brought by the spiral elevator; the sorting control end includes a data processing platform, such as the NVIDIA Jetson AGX Xavier edge computing unit, which is equipped with relevant sorting algorithms and control protocols, such as EtherCAT real-time communication with a cycle of 1ms.

[0026] The system uses a spatially coordinated arrangement of lasers and hyperspectral cameras for information collection. Additionally, a set of standard reflectors, including white and black boards, are placed on either side of the main conveyor belt. When a bottle lands on the main conveyor belt, lasers, hyperspectral cameras, environmental sensors, and standard reflectors are used to acquire and compensate for the bottle's location, resulting in more accurate bottle information. This information is then processed using algorithms to determine where the bottle should be sorted, and the corresponding air valves are controlled to ensure efficient and accurate bottle sorting.

[0027] Figure 1 This is a block diagram of a multi-category plastic automated reflow sorting method provided by this application. Figure 1 As shown, a multi-category plastic automated reflow sorting method includes the following steps: Step S100 , obtaining point cloud data of the bottle body, and processing the point cloud data using a laser triangulation method to obtain the bottle body morphological features of the bottle body.

[0028] This application describes the sorting control end of a sorting device. A multi-laser collaborative triangulation solution achieves submillimeter-level 3D reconstruction. Three lasers emit laser light, forming parallel stripes with a spacing of 15 nm on the bottle surface. The deformed laser stripes are then captured at 120 fps to acquire laser data of the bottle. The laser exposure time is automatically adjusted based on the speed of the main conveyor belt, enabling the deformed laser stripes to be captured at 120 fps to eliminate motion blur.

[0029] After the sorting control end obtains the laser data, it can establish the laser plane equation through the standard ball target, use the Steger algorithm to determine the center line of the bottle, and then obtain the point cloud data of the bottle based on the camera optical center, image point and laser plane equation.

[0030] The sorting control terminal synchronously acquires point cloud data of the bottles based on trigger signals sent by the sorting device. Specifically, when a bottle needs to be sorted, the sorting device sends a trigger signal, which activates the laser to acquire laser data about the bottle and sends it to the sorting control terminal. The sorting control terminal then processes the laser data to generate point cloud data, providing guidance for subsequent acquisition of initial spectral information for the bottle, ensuring that both point cloud data and initial spectral information are obtained for the same bottle.

[0031] After the sorting control terminal obtains the point cloud data, it can perform spiral fitting on the point cloud data of the bottle area to obtain the bottle mouth point cloud data in a targeted manner, and use the bottle mouth point cloud data to replace the corresponding point cloud data to obtain more accurate point cloud data of the bottle body. Finally, RANSAC is used to process the newly obtained point cloud data to obtain the bottle body morphological characteristics. In this way, by using three lasers and using the Steger algorithm to process the laser data, and re-processing it at specific locations, more accurate point cloud data can be obtained, thereby ensuring the accuracy of the subsequent bottle body morphological characteristics. On the other hand, the trigger signal sent by the sorting device is also used to synchronously obtain the point cloud data of the bottle body, providing guidance for the subsequent appropriate acquisition of initial spectral information for the same bottle body.

[0032] Step S200 , obtaining initial spectrum information of the bottle and the working temperature and humidity of the working environment, and determining whether spectrum compensation is required based on the working temperature and humidity and the initial spectrum information.

[0033] The hyperspectral camera synchronously acquires the initial spectral information of the bottle based on the trigger signal sent by the sorting device. Specifically, the sorting device simultaneously sends the trigger signal to the hyperspectral camera. Upon receiving the trigger signal, the hyperspectral camera immediately switches between the six characteristic wavelengths of 905nm, 1070nm, 1215nm, 1300nm, 1450nm, and 1550nm in conjunction with the narrowband filter wheel to acquire the initial spectral information of the bottle and send it to the sorting control terminal. This allows the sorting control terminal to obtain the initial spectral information and point cloud data almost synchronously, enabling the acquisition of both point cloud data and initial spectral information for the same bottle, reducing the likelihood that the point cloud data and initial spectral information do not correspond to the same bottle.

[0034] The above-mentioned working temperature and humidity specifically refer to the temperature and humidity of the bottle body. The working temperature and humidity can be obtained by using an environmental sensor and sent to the sorting control terminal so that the sorting control terminal can obtain the working temperature and humidity of the working environment. Among them, the working temperature and humidity can be obtained once each time point cloud data or initial spectral information is obtained. Then, the working temperature and humidity are subtracted from the previous working temperature and humidity to obtain the temperature and humidity difference, and the temperature and humidity difference is compared with the preset temperature and humidity. If the working temperature and humidity exceed the preset temperature and humidity, it indicates that the temperature and humidity level needs to compensate the spectrum; if the working temperature and humidity do not exceed the preset temperature and humidity, it indicates that the temperature and humidity level does not need to compensate the spectrum. Among them, the preset temperature and humidity can be determined based on the thermal expansion coefficient of the material. The thermal expansion coefficient of the material is obtained from historical experiments and stored at the sorting control terminal.

[0035] Of the six characteristic wavelengths at 905nm, 1070nm, 1215nm, 1300nm, 1450nm, and 1550nm, the 1300nm wavelength is specifically used to detect the hydroxyl absorption peak of PET bottles, the 1550nm wavelength identifies the C-H bond stretching vibration characteristic of HDPE, and the 1450nm wavelength detects the C-H bond third harmonic absorption peak of PP. When spectral compensation is not necessary, a certain characteristic wavelength band will have a higher reflectivity. When spectral compensation is required, a certain characteristic wavelength band will have a lower reflectivity. For example, if the bottle is contaminated by oil or labeling, the reflectivity in the characteristic wavelength band corresponding to the material will decrease. Since the bottle will not have reflectivity in characteristic wavelength bands not corresponding to the material, the initial light distribution information can be used to determine whether spectral compensation is necessary at the contamination level.

[0036] Step S300: If necessary, the working temperature and humidity and / or initial spectrum information and / or bottle motion information are substituted into a preset compensation model to obtain spectrum compensation information, and the actual spectrum information of the bottle is determined based on the initial spectrum information and the spectrum compensation information.

[0037] When determining that a spectrum needs to be compensated, whether the spectrum needs to be compensated at the temperature and humidity level or the spectrum needs to be compensated at the pollution level, it is determined that the spectrum needs to be compensated.

[0038] The preset compensation models include a temperature and humidity layer compensation model and a pollution layer compensation model. If it is determined that only the temperature and humidity layer compensation spectrum is required, the spectral compensation information includes the temperature and humidity layer spectrum compensation information. If it is determined that only the pollution layer compensation spectrum is required, the spectral compensation information includes the pollution layer spectrum compensation information. If it is determined that both the temperature and humidity layer compensation spectrum and the pollution layer compensation spectrum are required, the spectral compensation information includes the temperature and humidity layer spectrum compensation information and the pollution layer spectrum compensation information. Figure 2 This is a block diagram of the method provided by the present application for substituting the working temperature and humidity and / or initial spectrum information into a preset compensation model to obtain spectrum compensation information. Figure 2 As shown, substituting the working temperature and humidity and / or initial spectrum information into the preset compensation model to obtain spectrum compensation information includes the following steps: Step S301: Substitute the operating temperature and humidity into the temperature and humidity layer compensation model to obtain temperature and humidity layer spectrum compensation information.

[0039] Step S302, and / or, substituting the initial spectrum information into the pollution layer compensation model to obtain pollution layer spectrum compensation information.

[0040] The aforementioned temperature and humidity layer compensation model refers to the relationship between temperature and humidity and spectral shift. Specifically, a material property database is provided. For temperature compensation, the thermal expansion coefficients of various materials are pre-stored. The temperature and humidity layer spectrum compensation information can be calculated by using the formula: thermal expansion coefficient * (operating temperature and humidity - 25) to obtain the information required for temperature compensation.

[0041] In addition, the sorting control terminal also stores the optical path attenuation curves of different bands under different humidity. For humidity compensation, the information that needs to be compensated due to humidity can be obtained by using the corresponding optical path attenuation curves.

[0042] When it is determined that the temperature and humidity layer compensation spectrum is needed, the temperature and humidity are substituted into the temperature and humidity layer compensation model to obtain the temperature and humidity layer spectrum compensation information.

[0043] The above-mentioned pollution layer compensation model is specifically an anti-interference spectral database. The establishment of this anti-interference spectral database includes three parts: laboratory simulation, data collection, and model establishment. Among them, laboratory simulation is to simulate actual pollution scenarios through accelerated aging experiments, including 12 types of pollution states such as gradient pollution and different material labels. Data collection is to use a hyperspectral camera to record the spectral attenuation curve under various pollution states. Model establishment is to use Gaussian process regression to establish a pollution type-waveform attenuation compensation model, and dynamically adjust parameters through Bayesian optimization to ensure model accuracy. That is, after obtaining the initial spectral information, the Markov decision process is used to process the initial spectral information to determine the pollution type, and then substitute it into the pollution type-waveform attenuation compensation model to obtain the pollution layer spectral compensation information.

[0044] Next, the temperature and humidity layer spectral compensation information and the pollution layer spectral compensation information are superimposed for different bands to obtain spectral compensation information. Finally, the initial spectral information and spectral compensation information are superimposed for different waveforms to obtain the actual spectral information of the bottle after compensation.

[0045] Preferably, the sorting control end also obtains the actual reflectivity of the whiteboard at every preset time interval, and determines whether the actual reflectivity deviation between the actual reflectivity and the set reflectivity of the whiteboard is greater than the trigger compensation reflection deviation. If not, a reflective plate no deviation instruction is generated; if greater, reflective plate spectral compensation information is generated based on the actual reflectivity deviation, and the spectral compensation information is used to update the actual spectral information.

[0046] A set of standard reflectors is installed on each side of the main conveyor belt. Each set includes a whiteboard and a blackboard, with the whiteboard set to have a reflectivity of 98%. A hyperspectral camera is used to obtain the actual reflectivity of the whiteboard at preset intervals. The actual reflectivity is then subtracted from the set reflectivity to determine the actual reflectivity deviation. The actual reflectivity deviation is then compared with the set trigger compensation reflectivity deviation. If the driver's reflectivity deviation is no greater than the trigger compensation reflectivity deviation, the standard reflectors have minimal impact on the initial spectral information and are negligible, eliminating the need for further compensation. At this point, the next step is to generate a zero-deviation instruction for the reflector to further determine the plastic type of the bottle.

[0047] If the actual reflectivity deviation is greater than the trigger compensation reflectivity deviation, it indicates that the standard reflector will affect the value of the initial spectral information. At this time, a trigger compensation mechanism is needed, that is, the actual reflectivity deviation is substituted into the preset trigger compensation algorithm to obtain the reflector spectrum compensation information, and then the reflector spectrum compensation information is superimposed with the above-mentioned actual spectral information for different waveforms to update the above-mentioned actual spectral information to obtain new actual spectral information, thereby completing the work of updating the actual spectral information. Among them, the preset trigger compensation algorithm can be obtained through a large number of experiments offline, and the preset time can be determined according to the actual situation. The present application preferably sets it to thirty minutes. In this way, by further considering the state of the standard reflector to further correct the actual spectral information of the bottle, the actual spectral information obtained can be made more accurate, thereby providing accurate bottle information for subsequent plastic sorting, and then indirectly improving the accuracy of subsequent sorting work while improving the sorting efficiency when using the sorting device for sorting.

[0048] In step S400, the bottle morphological characteristics and actual spectral information are substituted into the hybrid neural network model to obtain the plastic type of the bottle and the corresponding confidence level, determine the confidence level corresponding to the confidence level, and generate corresponding sorting instructions based on the confidence level to implement bottle sorting.

[0049] The above hybrid neural network model is a CNN-Transformer hybrid neural network. Figure 3 This is a block diagram of the method provided by the present application for substituting the bottle morphological characteristics and actual spectral information into the hybrid neural network model to obtain the plastic type of the bottle and the corresponding confidence level. Figure 3 As shown, substituting the bottle morphological characteristics and actual spectral information into the hybrid neural network model to obtain the plastic type of the bottle and the corresponding confidence level includes the following steps: Step S401: Use a three-layer CNN to process the bottle morphological features to obtain first intermediate output information.

[0050] Step S402: Perform attention weighting on the actual spectral information through Transformer to obtain second intermediate output information.

[0051] Step S403 : performing tensor concatenation on the first intermediate output information and the second intermediate output information to obtain information to be processed.

[0052] Step S404: Use the first loss function to process the information to be processed to obtain the plastic type of the bottle and the corresponding confidence level.

[0053] The hybrid neural network consists of an input and an output. The input data is a 64*64*64 voxelized 3D matrix, including morphological features such as the bottle thread and the bumps on the bottom. The input data is a 256*256 reflectance matrix composed of six-band spectral information. The output is the material classification probability, or confidence level, which represents the plastic type and the corresponding confidence level.

[0054] The CNN branch is a 3D model that processes the bottle's morphological features using three layers of 3x3x3 convolutional kernels. Each layer uses ReLU activation and BatchNorm, and then gradually reduces the feature map to an 8x8x8 feature map through max pooling (stride = 2), thereby generating the first intermediate output information. Deformable convolution is introduced in the third layer to adaptively capture the bottle's deformation characteristics, enabling fast and accurate generation of the first intermediate output information.

[0055] The Transformer branch converts the six-band spectral information into a 256-dimensional embedding vector and employs a four-head attention mechanism to focus on learning the correlations between key bands. Environmental parameters such as temperature and humidity are added to the positional encoding to generate the second intermediate output information.

[0056] The first and second intermediate outputs are then concatenated using bilinear interpolation, and a cross-attention mechanism is employed to achieve spatiotemporal alignment, yielding the information to be processed. Finally, the first loss function, FocalLoss, is used to process the information to be processed, outputting the bottle's plastic type and corresponding confidence level through the output of the hybrid neural network. By using a hybrid neural network to specifically process the actual spectral information, and by combining it with the high accuracy of the actual spectral information, the hybrid neural network model can more accurately output the bottle's plastic type and corresponding confidence level.

[0057] Preferably, a second loss function may be used to process the information to be processed to obtain the pollution type and pollution degree value of the bottle body.

[0058] Specifically, the hybrid neural network model also employs a second loss function, Smooth L1 Loss, to process the information to be processed, resulting in a more accurate contamination type and contamination severity value for the bottle. This further verifies the accuracy of the contamination type previously derived solely from the initial spectral information, thereby obtaining a more accurate contamination type and corresponding contamination severity value.

[0059] The sorting control terminal pre-stores a confidence level and confidence level relationship table. This confidence level and confidence level relationship table is obtained through a large number of offline experiments. The corresponding confidence level can be obtained by substituting the confidence level obtained above into the confidence level and confidence level relationship table. Among them, the confidence level and confidence level relationship table contains three levels: direct sorting level, reflux sorting level, and intervention sorting level. For example, when the confidence level is greater than 90%, the confidence level corresponding to the confidence level is the direct sorting level; when the confidence level is not less than 70% and not greater than 90%, the confidence level corresponding to the confidence level is the reflux sorting level; when the confidence level is less than 70%, the confidence level corresponding to the confidence level is the intervention sorting level.

[0060] After the confidence level of the bottle is determined, it is determined to which position in the sorting device the bottle needs to be sorted, and then the sorting action needs to be performed. Figure 4 This is a block diagram of the method disclosed in this application for generating corresponding sorting instructions based on confidence levels to achieve bottle sorting. Figure 4 As shown, generating corresponding sorting instructions based on the confidence level to implement bottle sorting includes the following steps: Step S405: If the confidence level is the direct sorting level, the working information of the nozzle is determined based on the morphological characteristics of the bottle, and a sorting instruction corresponding to the working information is generated to implement bottle sorting.

[0061] Step S406: If the confidence level is the reflux sorting level, a reflux sorting instruction corresponding to the reflux channel is generated to open the reflux channel to implement bottle sorting.

[0062] Step S407: If the confidence level is the intervention sorting level, an intervention sorting instruction corresponding to the unknown type is generated to implement the sorting of the bottles.

[0063] When the confidence level is the direct sorting level, it indicates that the nozzle needs to be used to provide driving force to the bottle body to move the bottle body to the sorting position corresponding to its material. That is, the bottle body shape characteristics obtained in step S100 can be used to obtain the direction of the platform on the main conveyor belt, thereby further determining the working information required for the nozzle to work. This working information includes the working angle, working nozzle and working air valve amount to ensure that the bottle body can be accurately moved to the corresponding position.

[0064] Specifically, to determine the working valve volume, the calculation formula for the working valve volume is working valve volume = nozzle coefficient * (bottle material density * main conveyor belt speed + bottle deformation compensation item) (1 / 2) . Among them, the nozzle coefficient is obtained through offline test calibration, which reflects the influence of the nozzle shape on the airflow efficiency. The calibration method is to measure the airflow coverage of different nozzles under standard working conditions, and fit the sealing coefficient. This application prefers to select the nozzle coefficient as 1.05. The density of the bottle material determines the impact momentum requirement of the airflow. After the material of the bottle is determined, the density of the bottle material can be obtained by looking up the table. The main conveyor belt speed affects the bottle displacement compensation amount, and the main conveyor belt speed can be obtained by reading the motor running speed. The bottle deformation compensation item is obtained by converting the volume compression rate, that is, the bottle deformation compensation item = compensation coefficient * volume compression rate. The compensation coefficient is determined through impact mechanics experiments, and the volume compression rate can be obtained at the same time as the initial spectral information obtained in the above step S200.

[0065] For nozzle selection, the coordinates of the center of gravity of the bottle body can be detected by a laser, and then the coordinates of the center of gravity of the bottle body can be substituted into the valve number = the horizontal coordinate of the center of gravity of the bottle body / the vertical coordinate of the center of gravity of the bottle body + 1 according to the preset nozzle selection work. Then, considering the position prediction error caused by the deformation of the bottle body, the valve number is corrected by the motion tracking algorithm to determine which nozzle needs to be used for the work.

[0066] To determine the working angle, the offset in the vertical coordinate of the bottle's center of gravity is substituted into the formula: nozzle angle = arctan(offset in the vertical coordinate of the bottle's center of gravity / (main conveyor speed * airflow duration)) to obtain the initial working angle. The airflow duration can be determined based on actual conditions; in this application, 50ms is preferred. Next, to account for the torque effect of the airflow on the flattened bottle, the initial working angle is substituted into the angle correction algorithm to determine the working angle. By considering the bottle's state to determine the operating information required for the nozzle, the bottle can be more accurately sorted to the corresponding position, achieving fast and accurate bottle sorting.

[0067] When the confidence level is the reflux sorting level, it indicates that the label area of ​​the bottle is too large. In this case, a spiral elevator is needed to transport the bottle to the buffer bin, that is, a reflux sorting instruction corresponding to the reflux channel is generated to open the reflux channel to realize bottle sorting.

[0068] In addition, when transporting the bottles to the buffer bin, microwave sensors detect the water content of the bottles and X-rays detect metal residues, which prepares the bottles for subsequent sorting, facilitating the subsequent sorting of the bottles.

[0069] When the confidence level reaches manual sorting, indicating that an unknown material or a matching material has been detected, X-ray metal composition detection can be initiated, and high-resolution images and spectral data can be uploaded to the cloud. Once a certain amount of data has been accumulated, manual sorting can be performed, generating intervention sorting instructions corresponding to the unknown type to implement bottle sorting. In this way, by determining the plastic type of the bottle and the corresponding confidence level, and then further executing different sorting operations based on different situations, the bottles can be sorted quickly and accurately.

[0070] Figure 5 This is a connection diagram of a multi-category plastic automated return sorting system provided by this application. Figure 5 As shown, a multi-category plastic automated reflow sorting system includes: an information acquisition module, an information processing module and a plastic sorting module.

[0071] The information acquisition module is used to obtain point cloud data of the bottle, process the point cloud data using laser triangulation to obtain the bottle's morphological characteristics, and obtain the bottle's initial spectral information and the working temperature and humidity of the working environment. The information processing module is used to determine whether spectral compensation is required based on the working temperature and humidity and initial spectral information. If necessary, the working temperature and humidity and / or initial spectral information are substituted into a preset compensation model to obtain spectral compensation information. The actual spectral information of the bottle is determined based on the initial spectral information and spectral compensation information. The plastic sorting module is used to substitute the bottle's morphological characteristics and actual spectral information into a hybrid neural network model to obtain the bottle's plastic type and corresponding confidence level, determine the confidence level corresponding to the confidence level, and generate corresponding sorting instructions based on the confidence level to implement bottle sorting.

[0072] The other functions performed by the above-mentioned information acquisition module, information processing module and plastic sorting module, as well as the technical details of each function, are the same or similar to the corresponding features in the previously described multi-category plastic automated reflow sorting method, so they will not be repeated here.

[0073] An embodiment of the present application also provides a computer storage medium having a computer program stored thereon. When the computer storage medium is run on a computer, the computer can execute the steps of the aforementioned method for automated reflow sorting of multiple categories of plastics.

[0074] It should be understood that although the steps in the flowcharts of the accompanying drawings are shown in sequence as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps and they may be executed in other orders.

[0075] The above description is only part of the implementation methods of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A multi-category plastic automated reflow sorting method, characterized in that: The method comprises: Acquiring point cloud data of the bottle body, and processing the point cloud data using a laser triangulation method to obtain bottle body morphological features of the bottle body; Obtaining the initial spectrum information of the bottle and the working temperature and humidity of the working environment, and determining whether spectrum compensation is required based on the working temperature and humidity and the initial spectrum information, If necessary, substituting the working temperature and humidity and / or the initial spectrum information into a preset compensation model to obtain spectrum compensation information, and determining the actual spectrum information of the bottle based on the initial spectrum information and the spectrum compensation information; The bottle morphological characteristics and the actual spectral information are substituted into a hybrid neural network model to obtain the plastic type of the bottle and the corresponding confidence level, determine the confidence level corresponding to the confidence level, and generate corresponding sorting instructions based on the confidence level to achieve sorting of the bottles.

2. The method according to claim 1, characterized in that The method is applied to a sorting device comprising three lasers, wherein the three lasers are evenly distributed in a ring shape on the sorting device. The step of obtaining point cloud data of the bottle body comprises: Point cloud data of the bottle body is synchronously acquired based on the trigger signal sent by the sorting device.

3. The method according to claim 1, characterized in that The preset compensation model includes a temperature and humidity layer compensation model and a pollution layer compensation model, the spectral compensation information includes temperature and humidity layer spectral compensation information and / or pollution layer spectral compensation information, and substituting the operating temperature and humidity and / or the initial spectral information into the preset compensation model to obtain the spectral compensation information includes: Substituting the working temperature and humidity into the temperature and humidity layer compensation model to obtain temperature and humidity layer spectrum compensation information; And / or, the initial spectrum information is substituted into a pollution layer compensation model to obtain pollution layer spectrum compensation information.

4. The method according to claim 2, characterized in that The sorting device includes a conveyor belt, and a set of standard reflective plates are installed on both sides of the conveyor belt, each set of standard reflective plates includes a white plate. The method further includes: Obtaining the actual reflectivity of the whiteboard at every preset time interval, determining whether the actual reflectivity deviation between the actual reflectivity and the set reflectivity of the whiteboard is greater than the trigger compensation reflection deviation, and if not, generating a reflective plate no deviation instruction; If it is greater, spectrum compensation information of the reflector is generated based on the actual reflectivity deviation, and the spectrum compensation information is used to update the actual spectrum information.

5. The method according to claim 3, characterized in that The hybrid neural network model is a CNN-Transformer hybrid neural network, and substituting the bottle morphological characteristics and the actual spectral information into the hybrid neural network model to obtain the plastic type of the bottle and the corresponding confidence level includes: Using a three-layer CNN to process the bottle morphological features to obtain first intermediate output information; Performing attention weighting on the actual spectral information through a Transformer to obtain second intermediate output information; Performing tensor splicing on the first intermediate output information and the second intermediate output information to obtain information to be processed; The information to be processed is processed using a first loss function to obtain the plastic type of the bottle and the corresponding confidence level.

6. The method according to claim 5, characterized in that The method further comprises: The information to be processed is processed using a second loss function to obtain the pollution type and pollution degree value of the bottle.

7. The method according to claim 1, characterized in that Generating corresponding sorting instructions based on the confidence level to implement sorting of the bottles includes: If the confidence level is a direct sorting level, determining the working information of the nozzle based on the morphological characteristics of the bottle, and generating a sorting instruction corresponding to the working information to implement sorting of the bottle; If the confidence level is a reflux sorting level, generating a reflux sorting instruction corresponding to a reflux channel to open the reflux channel to implement sorting of the bottles; If the confidence level is an intervention sorting level, an intervention sorting instruction corresponding to the unknown type is generated to implement sorting of the bottles.

8. The method according to claim 2, characterized in that The sorting device includes a hyperspectral camera, and obtaining initial spectral information of the bottle includes: The hyperspectral camera synchronously acquires initial spectral information of the bottle body based on the trigger signal sent by the sorting device.

9. A multi-category plastic automated return sorting system, characterized in that: The system includes: an information acquisition module, an information processing module and a plastic sorting module; wherein, The information acquisition module is used to obtain point cloud data of the bottle body, process the point cloud data using laser triangulation to obtain the bottle body morphological characteristics of the bottle body, and obtain initial spectral information of the bottle body and the working temperature and humidity of the working environment; The information processing module is configured to determine whether spectrum compensation is required based on the operating temperature and humidity and the initial spectrum information; if required, substitute the operating temperature and humidity and / or the initial spectrum information into a preset compensation model to obtain spectrum compensation information; and determine actual spectrum information of the bottle based on the initial spectrum information and the spectrum compensation information; The plastic sorting module is used to substitute the bottle morphological characteristics and the actual spectral information into the hybrid neural network model to obtain the plastic type of the bottle and the corresponding confidence level, determine the confidence level corresponding to the confidence level, and generate corresponding sorting instructions based on the confidence level to achieve sorting of the bottles.

10. A computer-readable storage medium storing a computer program that can be run on a processor, characterized in that: When the computer program is executed by the processor, the computer program implements the method for automated reflow sorting of multiple categories of plastics as described in any one of claims 1 to 8.

Images