Multi-class material sorting method and system based on hyperspectrum, medium and product

Abstract

The invention discloses a multi-class material sorting method and system based on hyperspectrum, a medium and a product, and relates to the technical field of intelligent sorting. According to the method, the sorting system collects the hyperspectral data of different types of materials on the conveying belt, and the characteristic wavelength point with the highest contribution degree to material classification is selected by combining the spectral response differences of the different types of materials at a plurality of wavelength points; and the transverse center positions and the longitudinal center positions of the to-be-sorted materials and the characteristic spectral response values at the characteristic wavelength points are extracted, a characteristic data set is generated and input into the recognition model for recognition, the sorting rule database is inquired according to the recognition result to determine the sorting parameters corresponding to the to-be-sorted materials, and finally accurate sorting of the materials is achieved. Compared with a traditional sorting method which only depends on RGB images, the method can capture spectral characteristics of the surfaces of the materials, is not affected by surface pollution, color fading, light reflection and shadow, can distinguish materials with similar appearances but different materials, and greatly improves the sorting accuracy and stability.

Description

Technical Field

[0001] This application relates to the field of intelligent sorting technology, and in particular to a method, system, medium and product for sorting multiple types of materials based on hyperspectral imaging. Background Technology

[0002] With the continuous improvement of industrial automation, material sorting, as a key link in the production process, directly determines the overall operating performance of the production line through its efficiency and accuracy. Especially in fields such as waste recycling, ore sorting, and agricultural product grading, achieving rapid and accurate classification of different types of materials has become a core requirement for ensuring production efficiency.

[0003] Currently, most material sorting methods rely on RGB cameras to capture material images and then analyze the material's appearance features, such as color and shape, to complete the classification. The specific process is as follows: The sorting system first captures RGB images of the materials on the conveyor belt, then extracts the material's color and shape features, next uses a classification model to complete material identification, and finally uses a pneumatic device to sort different types of materials to their corresponding bins.

[0004] However, this multi-material sorting method that relies on RGB images has obvious limitations: when the material surface is contaminated or faded, it is difficult to accurately extract effective features using only RGB images; in complex lighting environments, the reflections and shadows generated on the material surface will seriously interfere with the extraction effect of image features; for materials that look similar but have different materials, RGB images cannot capture the subtle differences between them, which can easily lead to sorting errors and reduce overall sorting efficiency. Summary of the Invention

[0005] This application provides a method, system, medium, and product for sorting multiple types of materials based on hyperspectral imaging, which can improve the accuracy of material sorting.

[0006] Firstly, this application provides a hyperspectral-based method for sorting multiple types of materials, applied to a sorting system. The method includes: acquiring conveyor belt encoder position signals and hyperspectral data of different types of materials on the conveyor belt; the conveyor belt encoder position signals are used to determine the position information of different types of materials on the conveyor belt; the hyperspectral data refers to the spectral response values ​​of multiple wavelength points continuously acquired within a preset wavelength range at preset wavelength intervals; analyzing the differences in spectral response of different types of materials at multiple wavelength points, calculating the contribution of each wavelength point to material classification, and determining a preset number of wavelength points with the highest contribution as characteristic wavelength points; and extracting the items to be sorted from the hyperspectral data on the conveyor belt. The system identifies the characteristic spectral response values ​​of materials at characteristic wavelength points and determines the lateral and longitudinal center positions of the materials to be sorted on the conveyor belt based on location information to generate a feature dataset. This feature dataset is then input into the recognition model to obtain the recognition results for each material to be sorted. Based on the recognition results of each material, the system queries the sorting rule database to match the corresponding sorting parameters for each material. These parameters include sorting angle, sorting air pressure, and hopper position. Based on the sorting parameters, conveyor belt speed, and the location information of each material, the system determines the material sorting execution sequence. According to the material sorting execution sequence, the system controls the pneumatic actuator to sort each material.

[0007] By adopting the above technical solution, the sorting system collects hyperspectral data of different types of materials on the conveyor belt. Combining the differences in spectral response of different material types at multiple wavelengths, it selects the characteristic wavelengths that contribute most to material classification, extracts the characteristic spectral response values ​​of the materials to be sorted at these wavelengths, determines the lateral and longitudinal center positions of the materials, generates a feature dataset, and inputs it into the recognition model for identification. Based on the recognition results, it queries the sorting rule database to determine the corresponding sorting parameters for each material, ultimately achieving accurate material sorting. Compared to traditional sorting methods that rely solely on RGB images, this method can capture the spectral characteristics of the material surface, is unaffected by surface contamination, fading, reflection, and shadows, and can also distinguish materials that look similar but have different materials, greatly improving the accuracy and stability of sorting.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, the differences in spectral response of different types of materials at multiple wavelength points are analyzed, and the contribution of each wavelength point to material classification is calculated. Specifically, this includes: based on a preset material spectral feature library, determining the average response value corresponding to each type of material at each wavelength point, and calculating the overall average response value of all types of materials at each wavelength point; calculating the average of the squared differences between the average response value corresponding to each type of material at each wavelength point and the overall average value to obtain the inter-class variance of each wavelength point; based on the material spectral feature library, determining the response standard deviation corresponding to each type of material at each wavelength point, calculating the average of the response standard deviations to obtain the intra-class variance of each wavelength point; determining the discrimination index of each wavelength point by the ratio of the inter-class variance to the intra-class variance of each wavelength point; and weighted summing the discrimination index and signal-to-noise ratio of each wavelength point to obtain the contribution index of each wavelength point.

[0009] By employing the above technical solution, the sorting system calculates the ratio of inter-class variance to intra-class variance to determine the discriminative power of wavelength points. This is then weighted by combining the signal-to-noise ratio of the wavelength points, effectively identifying the characteristic wavelength points that contribute most to material classification. This method automatically identifies the most discriminative spectral features, reduces data redundancy, and improves the efficiency and accuracy of feature extraction.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, calculating the signal-to-noise ratio (SNR) at each wavelength point specifically includes: acquiring background spectral data when there is no material on the conveyor belt to determine multiple background response values ​​at each wavelength point; calculating the arithmetic mean and standard deviation of the multiple background response values ​​at each wavelength point to obtain the background reference value and background noise value at each wavelength point; acquiring reference spectral data when a standard reflectance whiteboard is placed on the conveyor belt to determine multiple reference response values ​​at each wavelength point; calculating the arithmetic mean of the multiple reference response values ​​at each wavelength point to obtain the reference signal value at each wavelength point; and substituting the reference signal value, background reference value, and background noise value into the SNR calculation formula to obtain the SNR at each wavelength point.

[0011] By employing the above technical solution, the sorting system collects background spectral data when the conveyor belt is empty and reference spectral data when a standard reflectance whiteboard is placed on the conveyor belt. This data is then used to calculate the signal-to-noise ratio (SNR) at each wavelength point, effectively assessing the reliability of data at different wavelengths. Wavelength points with higher SNR have more reliable spectral data and are more suitable for material classification. This method can eliminate wavelengths severely affected by noise, improving the quality of feature data and thus enhancing the accuracy of classification and recognition.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, the feature dataset is input into the recognition model to obtain the recognition result of each material to be sorted. Specifically, this includes: inputting the feature dataset into a fast recognition model to obtain the initial recognition result and initial confidence level of each material to be sorted; if the initial confidence level is greater than or equal to a preset threshold, the initial recognition result is determined to be the recognition result of each material to be sorted; if the initial confidence level is less than the preset threshold, the feature dataset is input into a precise recognition model to obtain the final recognition result and final confidence level, and the final recognition result is determined to be the recognition result of each material to be sorted. The computational complexity of the fast recognition model is lower than that of the precise recognition model.

[0013] By adopting the above technical solution, the sorting system employs a two-level identification strategy combining a fast identification model and a precise identification model: for samples with high initial confidence, the fast identification model with lower computational complexity is used directly to obtain the results, improving processing efficiency; for samples with low initial confidence, a precise identification model with higher computational complexity but higher identification accuracy is used for secondary identification to ensure identification accuracy. This hierarchical identification method ensures both the overall efficiency of material sorting and the reliability of identification results. In particular, for materials with similar appearance characteristics that are difficult to distinguish, the in-depth analysis of the precise identification model can significantly improve the classification accuracy.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, based on the identification results of each material to be sorted, a sorting rule database is queried to match the sorting parameters corresponding to each material to be sorted. The sorting parameters include sorting angle, sorting air pressure, and hopper position. Specifically, this includes: based on the identification results of each material to be sorted, querying the hopper position corresponding to each material to be sorted from the preset sorting rule database; determining the sorting angle based on the relative positional relationship between the hopper position and the lateral center position of the material to be sorted; and matching the corresponding sorting air pressure from the sorting rule database based on the quality and size characteristics of the material to be sorted.

[0015] By adopting the above technical solution, the sorting system analyzes the relative positional relationship between the bin location and the lateral center position of the material to be sorted to determine the sorting angle, and matches the appropriate sorting air pressure according to the quality and size characteristics of the material to be sorted, thus achieving precise configuration of sorting parameters. This parameter matching method based on material characteristics and positional relationship allows the pneumatic actuator to generate the most suitable force and direction of action for the current material, ensuring that the material can accurately enter the target bin while avoiding sorting failures caused by excessive or insufficient force.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, the material sorting execution sequence is determined based on sorting parameters, conveyor belt speed, and the position information of each material to be sorted. Specifically, this includes: calculating the estimated time when each material to be sorted will arrive at the air nozzle position based on the conveyor belt speed and the longitudinal center position of each material to be sorted; determining the number of air nozzle opening groups and the opening position based on the lateral center position and sorting angle of each material to be sorted; calculating the execution duration of the air nozzles based on the sorting air pressure and the quality characteristics of each material to be sorted; and combining the estimated time, the number of opening groups, the opening position, and the execution duration to generate the material sorting execution sequence.

[0017] By adopting the above technical solution, the sorting system comprehensively considers multiple factors such as conveyor belt speed, material position, sorting angle, and sorting air pressure. It accurately calculates the estimated arrival time of each material to be sorted at the air nozzle position, the number of air nozzles to be opened, their opening positions, and the execution duration, generating a detailed material sorting execution sequence. This precise timing control ensures that the pneumatic actuators apply force to the materials at the optimal time, avoiding sorting failures caused by improper timing. Furthermore, by rationally planning the sorting sequence of multiple materials, the continuous operation capability of the sorting system can be improved, achieving efficient and stable material sorting.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, after the step of controlling the pneumatic actuator to sort each material to be sorted according to the material sorting execution sequence, the method further includes: obtaining the actual sorting trajectory and the expected sorting trajectory of each material to be sorted; when the deviation between the actual sorting trajectory and the expected sorting trajectory exceeds a preset deviation threshold, determining the sorting status of the target material to be sorted as sorting failure; and performing secondary sorting on the target material to be sorted based on the target material information, wherein the target material information includes the target material type and target location information.

[0019] By adopting the above technical solution, the sorting system tracks the actual sorting trajectory of each material in real time and compares it with the expected trajectory. When a deviation exceeds a preset deviation threshold, a second sorting is performed promptly, establishing a closed-loop sorting quality control mechanism. This real-time monitoring and timely correction method can effectively reduce the sorting error rate and improve the reliability of the sorting system. Especially for some special materials, even if the first sorting fails, a second sorting can ensure that they eventually enter the correct hopper.

[0020] In a second aspect, embodiments of this application provide a sorting system comprising: one or more processors and a memory; the memory is coupled to the one or more processors and is used to store computer program code, the computer program code including computer instructions, wherein the one or more processors invoke the computer instructions to cause the sorting system to perform the method described in the first aspect and any possible implementation thereof.

[0021] Thirdly, embodiments of this application provide a computer program product containing instructions that, when the computer program product is run on a sorting system, cause the sorting system to perform the method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a sorting system, cause the sorting system to perform the method described in the first aspect and any possible implementation thereof.

[0023] Understandably, the sorting system provided in the second aspect, the computer program product provided in the third aspect, and the computer storage medium provided in the fourth aspect are all used to execute the methods provided in the embodiments of this application. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0024] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. By adopting the above technical solution, the sorting system collects hyperspectral data of different types of materials on the conveyor belt. Combining the differences in spectral response of different types of materials at multiple wavelengths, it selects the characteristic wavelengths that contribute most to material classification, extracts the characteristic spectral response values ​​of the materials to be sorted at these wavelengths, determines the lateral and longitudinal center positions of the materials, generates a feature dataset, and inputs it into the recognition model for identification. Based on the recognition results, it queries the sorting rule database to determine the sorting parameters corresponding to each material, ultimately achieving accurate material sorting. Compared to traditional sorting methods that rely solely on RGB images, this method can capture the spectral characteristics of the material surface, is unaffected by surface contamination, fading, reflection, and shadows, and can also distinguish materials that look similar but have different materials, greatly improving the accuracy and stability of sorting.

[0025] 2. By adopting the above technical solution, the sorting system calculates the ratio of inter-class variance to intra-class variance to determine the discriminative power of wavelength points. Combined with the signal-to-noise ratio of these wavelength points for weighting, the system can effectively filter out the characteristic wavelength points that contribute most to material classification. This method can automatically identify the most discriminative spectral features, reduce data redundancy, and improve the efficiency and accuracy of feature extraction.

[0026] 3. By adopting the above technical solution, the sorting system tracks the actual sorting trajectory of each material in real time and compares it with the expected trajectory. When a deviation exceeds a preset deviation threshold, a second sorting is performed promptly, establishing a closed-loop sorting quality control mechanism. This real-time monitoring and timely correction method can effectively reduce the sorting error rate and improve the reliability of the sorting system. Especially for some special materials, even if the first sorting fails, a second sorting can ensure that they eventually enter the correct hopper. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating a hyperspectral-based multi-material sorting method in an embodiment of this application. Figure 2 This is another flowchart illustrating the hyperspectral-based multi-material sorting method in this application embodiment; Figure 3 This is a schematic diagram of the physical device structure of a sorting system in an embodiment of this application. Detailed Implementation

[0028] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0029] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0030] The following describes the process of the method provided in this implementation. Please refer to [link / reference]. Figure 1 This is a flowchart illustrating a multi-material sorting method based on hyperspectral imaging in an embodiment of this application.

[0031] S101. Collect the position signal of the conveyor belt encoder and the hyperspectral data of different types of materials on the conveyor belt. The position signal of the conveyor belt encoder is used to determine the position information of different types of materials on the conveyor belt. The hyperspectral data refers to the spectral response values ​​of multiple wavelength points continuously collected within a preset wavelength range at preset wavelength intervals. Among them, the conveyor belt encoder position signal refers to the digital or analog signal used to measure the position and motion state of the conveyor belt. It is usually generated by an encoder installed on the drive shaft of the conveyor belt and can reflect the motion parameters such as displacement and speed of the conveyor belt in real time. Hyperspectral data refers to the spectral response values ​​of multiple wavelength points continuously collected within a preset wavelength range (such as 400-2500nm) at preset wavelength intervals (such as 1nm). Each wavelength point corresponds to a spectral response value to form the spectral characteristic curve of the material surface. Position information represents the spatial position parameters of the material on the conveyor belt, including the coordinate values ​​relative to the edge of the conveyor belt or a preset reference point. Different types of materials refer to various materials that need to be distinguished and classified by the sorting system, such as waste, ore, agricultural products, etc. of different materials, compositions, or categories.

[0032] Specifically, firstly, the sorting system uses an encoder mounted on the conveyor belt drive shaft to collect the conveyor belt's position signal in real time, tracking its movement. Simultaneously, a hyperspectral camera mounted above the conveyor belt continuously scans the various materials being transported on it. Within a preset wavelength range (e.g., visible light to near-infrared 400-2500nm), at preset wavelength intervals (e.g., 1nm), the system collects the spectral response value of the material surface at each wavelength point. The sorting system synchronizes the conveyor belt encoder position signal with the hyperspectral scan data in time, establishing a correspondence between the material's position and its spectral characteristics, providing a data foundation for subsequent feature extraction and classification.

[0033] Assuming a plastic recycling sorting line with a conveyor belt speed of 0.5 m / s, a conveyor belt width of 1 m, an encoder resolution of 1000 pulses / revolution, a hyperspectral camera with a wavelength range of 400-2500 nm, a wavelength interval of 2 nm, and a scanning frequency of 100 lines / second, in the actual data acquisition process: Suppose that at a certain moment (t=10.5 seconds), there are 3 plastic items waiting to be sorted on the conveyor belt, and the collected data is as follows: Material 1 (PE plastic bottle): Encoder position: 5250 pulses; Lateral position: 0.3 meters from the left edge; Longitudinal position: 2.1 meters from the scan line; Spectral response values ​​(partial): 400nm: 0.45; 402nm: 0.47; 404nm: 0.52; ... 2498nm: 0.38; 2500nm: 0.35; Material 2 (PET plastic box): Encoder position: 5300 pulses; Lateral position: 0.6 meters from the left edge; Longitudinal position: 1.8 meters from the scan line; Spectral response values ​​(partial): 400nm: 0.38; 402nm: 0.41; 404nm: 0.44; ... 2498nm: 0.29; 2500nm: 0.27; Material 3 (PP plastic bag): Encoder position: 5400 pulses; Lateral position: 0.8 meters from the left edge; Vertical position: 1.5 meters from the scan line; Spectral response values ​​(partial): 400nm: 0.51; 402nm: 0.53; 404nm: 0.56; ... 2498nm: 0.42; 2500nm: 0.40; The sorting system records this data synchronously, forming a database of time-location-spectral correspondences. For example, it can be seen that at a wavelength of 400nm, the spectral response value of PE plastic bottles is 0.45, that of PET plastic boxes is 0.38, and that of PP plastic bags is 0.51.

[0034] S102. Analyze the differences in spectral response of different types of materials at multiple wavelength points, calculate the contribution of each wavelength point to the material classification, and determine the wavelength points with the highest contribution as the characteristic wavelength points. Among them, spectral response difference represents the numerical difference between the spectral response values ​​of different types of materials at various wavelength points, which can be quantitatively analyzed through statistical methods; contribution refers to the importance of a certain wavelength point in distinguishing different types of materials, which is usually measured by the dispersion of the spectral response values ​​of different materials at that wavelength point; characteristic wavelength points represent several discrete wavelength points that have a high contribution to material classification, and the spectral response values ​​at these wavelength points best reflect the distinguishing characteristics of different types of materials; preset quantity refers to the number of characteristic wavelength points predetermined according to actual classification needs, which is used to reduce data dimensionality while ensuring classification effect.

[0035] Specifically, the sorting system collects spectral response data for multiple known material types at various wavelengths. Based on this data, the system calculates the between-group variance (differences between different material types) and the within-group variance (fluctuations within the same material type) for each wavelength to determine the F-statistic (calculated as between-group variance / within-group variance). A larger F-statistic indicates a stronger ability of that wavelength to distinguish different materials and a higher contribution to material classification. The sorting system can determine a preset number of wavelengths as characteristic wavelengths based on actual conditions.

[0036] Optionally, under normal circumstances, analyzing the differences in spectral response of different types of materials at multiple wavelength points and calculating the contribution of each wavelength point to material classification can be achieved in the following ways, without limitation: Based on a preset material spectral feature library, determine the average response value corresponding to each type of material at each wavelength point, and calculate the overall average response value of all types of materials at each wavelength point; calculate the average of the squared differences between the average response value corresponding to each type of material at each wavelength point and the overall average value to obtain the inter-class variance of each wavelength point; based on the material spectral feature library, determine the standard deviation of the response corresponding to each type of material at each wavelength point, calculate the average of the standard deviations of the response to obtain the intra-class variance of each wavelength point; determine the discrimination index of each wavelength point by the ratio of the inter-class variance to the intra-class variance of each wavelength point; and perform a weighted summation of the discrimination index and the signal-to-noise ratio of each wavelength point to obtain the contribution index of each wavelength point.

[0037] Optionally, in general, the signal-to-noise ratio (SNR) at each wavelength can be calculated in the following ways, without limitation: Collect background spectral data when there is no material on the conveyor belt to determine multiple background response values ​​at each wavelength; calculate the arithmetic mean and standard deviation of the multiple background response values ​​at each wavelength to obtain the background reference value and background noise value at each wavelength; collect reference spectral data when a standard reflectance whiteboard is placed on the conveyor belt to determine multiple reference response values ​​at each wavelength; calculate the arithmetic mean of the multiple reference response values ​​at each wavelength to obtain the reference signal value at each wavelength; substitute the reference signal value, background reference value, and background noise value into the SNR calculation formula to obtain the SNR at each wavelength.

[0038] The following example of plastic sorting illustrates this calculation process: 1. Pre-set material spectral feature library (simplified to 3 types of plastics, 4 wavelength points): Wavelengths: 800nm, 1200nm, 1600nm, 2000nm; Spectral response values ​​of 10 PE plastic samples: 800nm: [0.45, 0.47, 0.44, 0.46, 0.45, 0.48, 0.46, 0.45, 0.47, 0.44]; 1200nm: [0.62, 0.63, 0.61, 0.64, 0.62, 0.63, 0.62, 0.61, 0.63, 0.62]; 1600nm: [0.38, 0.39, 0.37, 0.38, 0.39, 0.38, 0.37, 0.38, 0.39, 0.38]; 2000nm: [0.25, 0.26, 0.24, 0.25, 0.26, 0.25, 0.24, 0.25, 0.26, 0.25]; Spectral response values ​​of 10 PET plastic samples: 800nm: [0.38, 0.39, 0.37, 0.38, 0.39, 0.38, 0.37, 0.38, 0.39, 0.38]; 1200nm: [0.55, 0.56, 0.54, 0.55, 0.56, 0.55, 0.54, 0.55, 0.56, 0.55]; 1600nm: [0.45, 0.46, 0.44, 0.45, 0.46, 0.45, 0.44, 0.45, 0.46, 0.45]; 2000nm: [0.32, 0.33, 0.31, 0.32, 0.33, 0.32, 0.31, 0.32, 0.33, 0.32]; Spectral response values ​​of 10 PP plastic samples: 800nm: [0.51, 0.52, 0.50, 0.51, 0.52, 0.51, 0.50, 0.51, 0.52, 0.51]; 1200nm: [0.48, 0.49, 0.47, 0.48, 0.49, 0.48, 0.47, 0.48, 0.49, 0.48]; 1600nm: [0.42, 0.43, 0.41, 0.42, 0.43, 0.42, 0.41, 0.42, 0.43, 0.42]; 2000nm: [0.28, 0.29, 0.27, 0.28, 0.29, 0.28, 0.27, 0.28, 0.29, 0.28]; 2. Calculate the average response value at each wavelength point: PE average response value: 800nm: 0.457; 1200nm: 0.623; 1600nm: 0.381; 2000nm: 0.251; PET average response value: 800nm: 0.381; 1200nm: 0.551; 1600nm: 0.451; 2000nm: 0.321; PP average response value: 800nm: 0.511; 1200nm: 0.481; 1600nm: 0.421; 2000nm: 0.281; Overall average response value: 800nm: 0.450; 1200nm: 0.552; 1600nm: 0.418; 2000nm: 0.284; 3. Calculate the between-class variance: The inter-class variance at 800nm ​​= [(0.457-0.450)² + (0.381-0.450)² + (0.511-0.450)²] / 3 = 0.00266; The inter-class variance at 1200nm = [(0.623-0.552)² + (0.551-0.552)² + (0.481-0.552)²] / 3 = 0.00401; The inter-class variance at 1600nm = [(0.381-0.418)² + (0.451-0.418)² + (0.421-0.418)²] / 3 = 0.00097; The inter-class variance at 2000nm = [(0.251-0.284)² + (0.321-0.284)² + (0.281-0.284)²] / 3 = 0.00087; 4. Calculate the within-class variance: Standard deviation at 800nm: PE: 0.013, PET: 0.007, PP: 0.007; Within-class variance = (0.013 + 0.007 + 0.007) / 3 = 0.009; Standard deviation of 1200nm: PE: 0.009, PET: 0.007, PP: 0.008; Within-class variance = (0.009 + 0.007 + 0.008) / 3 = 0.008; Standard deviation of 1600nm: PE: 0.007, PET: 0.007, PP: 0.007; Within-class variance = (0.007 + 0.007 + 0.007) / 3 = 0.007; Standard deviation of 2000nm: PE: 0.007, PET: 0.007, PP: 0.007; Within-class variance = (0.007 + 0.007 + 0.007) / 3 = 0.007; 5. Calculate the discrimination index: The resolution at 800nm ​​= 0.00266 / 0.009 = 0.296; The discrimination index at 1200nm = 0.00401 / 0.008 = 0.501; The resolution at 1600nm = 0.00097 / 0.007 = 0.139; The discrimination index at 2000nm = 0.00087 / 0.007 = 0.124; 6. Calculate the signal-to-noise ratio: Background response values ​​(10 measurements): 800nm: [0.02, 0.03, 0.02, 0.03, 0.02, 0.03, 0.02, 0.03, 0.02, 0.03]; 1200nm: [0.03, 0.04, 0.03, 0.04, 0.03, 0.04, 0.03, 0.04, 0.03, 0.04]; 1600nm: [0.04, 0.05, 0.04, 0.05, 0.04, 0.05, 0.04, 0.05, 0.04, 0.05]; 2000nm: [0.05, 0.06, 0.05, 0.06, 0.05, 0.06, 0.05, 0.06, 0.05, 0.06]; Background baseline value: 800nm: 0.025; 1200nm: 0.035; 1600nm: 0.045; 2000nm: 0.055; Background noise level (standard deviation): 800nm: 0.005; 1200nm: 0.005; 1600nm: 0.005; 2000nm: 0.005; Standard whiteboard response value (average of 10 measurements): 800nm: 0.95; 1200nm: 0.93; 1600nm: 0.90; 2000nm: 0.88; Signal-to-noise ratio calculation: 800nm: (0.95-0.025) / 0.005=185; 1200nm: (0.93-0.035) / 0.005=179; 1600nm: (0.90-0.045) / 0.005=171; 2000nm: (0.88-0.055) / 0.005=165; 7. Final contribution calculation (assuming a weighting ratio of 0.7:0.3): Contribution of 800nm ​​= 0.7 × 0.296 + 0.3 × (185 / 185) = 0.507; Contribution of 1200nm = 0.7 × 0.501 + 0.3 × (179 / 185) = 0.642; The contribution of 1600nm = 0.7 × 0.139 + 0.3 × (171 / 185) = 0.374; Contribution at 2000nm = 0.7 × 0.124 + 0.3 × (165 / 185) = 0.354; The calculation results show that the 1200nm wavelength point has the highest contribution (0.642), indicating that the spectral characteristics of different types of plastics are most significantly different at this wavelength point. The 800nm ​​wavelength point is the second highest (0.507). The 1600nm and 2000nm wavelength points have lower contributions, indicating that these wavelength points have a relatively small effect on distinguishing different types of plastics. The sorting system can select the most characteristic wavelength points for subsequent classification and identification based on these contributions. For example, in this case, the 1200nm and 800nm ​​wavelength points can be selected as characteristic wavelength points.

[0039] It should be noted that in the above examples, the signal-to-noise ratio (SNR) was calculated using the formula: SNR = (Standard whiteboard response value - Background reference value) / Background noise value. Alternatively, in general, the noise measured using the background response value can be replaced with the dark noise value measured with the lens cap on or the noise value measured against a blackboard (blackboard response value). The formula would then be: SNR = (Standard whiteboard response value - Background reference value) / Dark noise value, where the dark noise value is obtained by measuring with the lens cap on.

[0040] The reason for this approach is that noise measured using the lens cap more accurately reflects the camera's own system noise, avoids interference from ambient light on background measurements, and the resulting signal-to-noise ratio better reflects the true performance of the camera system.

[0041] Therefore, the signal-to-noise ratio calculation method in the above example can be further improved. Using camera dark noise instead of background noise as the denominator would be more accurate.

[0042] S103. Extract the characteristic spectral response values ​​of the materials to be sorted on the conveyor belt at characteristic wavelength points from the hyperspectral data, and determine the lateral center position and longitudinal center position of the materials to be sorted on the conveyor belt based on the position information to generate a feature dataset. Input the feature dataset into the recognition model to obtain the recognition results of each material to be sorted. Among them, the characteristic spectral response value represents the spectral reflection or absorption intensity of the material at the characteristic wavelength point, which is a key parameter characterizing the spectral characteristics of the material; the lateral center position refers to the coordinates of the center point of the material in the width direction of the conveyor belt; the longitudinal center position refers to the coordinates of the center point of the material in the length direction of the conveyor belt; the feature dataset refers to the set of data samples composed of the spectral and positional characteristics of the material; the recognition model represents the machine learning model used for material classification, which can be a fast recognition model or an accurate recognition model; the recognition result refers to the material category label and corresponding confidence level output by the recognition model.

[0043] Specifically, firstly, the sorting system extracts the spectral response value of each material to be sorted at a characteristic wavelength point from the complete hyperspectral data, as spectral features. Simultaneously, based on synchronously acquired conveyor belt encoder position signals and image processing results, the sorting system calculates the centroid position of each material to be sorted, obtaining its lateral and longitudinal coordinates on the conveyor belt, as positional features. The sorting system combines the extracted spectral and positional features to form a feature dataset, which is then input into a pre-trained recognition model to obtain the recognition results for each material to be sorted.

[0044] The steps to build a deep learning-based recognition model are as follows: First, the sorting system collects the spectral characteristics of multiple different types of materials, using the spectral characteristics as input features and the material type as output features.

[0045] Then, the sorting system constructs an LSTM-based recurrent neural network, including an input layer, LSTM hidden layers, fully connected layers, and an output layer. The input layer takes in spectral features, and the number of layers and nodes in the LSTM hidden layers and fully connected layers is set according to the actual situation. The output layer outputs the material type.

[0046] Next, the sorting system uses the Adam optimizer with a learning rate of 0.001 and a training batch size of 32. These settings can be adjusted based on actual conditions and are not limited here. The sorting system divides 80% of the data into a training set and 20% into a validation set, training for 100 epochs. The model with the highest accuracy on the validation set is saved. These settings can also be adjusted based on actual conditions and are not limited here. An epoch is the process of the training dataset being completely processed by the neural network once. In machine learning and deep learning, an epoch is a unit used to measure the number of times the entire training set has been repeatedly learned. Specifically, an epoch is completed when the neural network completes one forward computation and one backward propagation, meaning all data has been processed by the network once. The sorting system uses binary cross-entropy as the loss function and employs Early Stopping to prevent overfitting. When the value of the loss function exceeds a preset threshold, the model training is considered complete, and a recognition model is obtained. Early Stopping is a technique in deep learning and machine learning to prevent model overfitting; it determines when to stop training by monitoring the model's performance on the validation set.

[0047] Finally, the sorting system inputs the input features from the validation set into the recognition model, obtaining the model's predicted output features. The predicted output features are then compared with the actual output features in the validation set, and performance metrics such as accuracy, precision, recall, F1 score, and mean squared error (MSE) are used to evaluate the model's performance. Based on the model's performance on the validation set, the model's parameters are adjusted, including adjusting the learning rate, changing model complexity (e.g., increasing or decreasing the number of layers or nodes in the neural network), and modifying the regularization strength. This process may require multiple iterations, each based on the previous learning results, to optimize the recognition model.

[0048] Optionally, in general, inputting the feature dataset into the recognition model to obtain the recognition results of each material to be sorted can be achieved in the following ways, without limitation: input the feature dataset into the fast recognition model to obtain the initial recognition result and initial confidence level of each material to be sorted; if the initial confidence level is greater than or equal to a preset threshold, the initial recognition result is determined to be the recognition result of each material to be sorted; if the initial confidence level is less than the preset threshold, input the feature dataset into the precise recognition model to obtain the final recognition result and final confidence level, and determine the final recognition result as the recognition result of each material to be sorted. The computational complexity of the fast recognition model is lower than that of the precise recognition model.

[0049] Specifically, the sorting system can first input the feature dataset into a fast recognition model (which has high computational efficiency) to obtain initial recognition results and initial confidence levels. If the initial confidence level is greater than or equal to a preset threshold, it means that the initial recognition result is reliable and no further recognition is needed; if the initial confidence level is less than the preset threshold, it means that the initial recognition result is unreliable and further recognition is needed. The sorting system then uses a more accurate and precise recognition model for secondary recognition, outputting the final recognition result and final confidence level.

[0050] The main differences between fast recognition models and accurate recognition models during construction can be: 1. Differences in network structure: The fast recognition model uses a simple feedforward neural network, usually with only 2-3 shallow layers, which has a small computational cost; the accurate recognition model uses LSTM combined with attention mechanism and deep fully connected network, which has a complex structure and stronger feature extraction capability. 2. Training parameter settings: The fast recognition model uses a large learning rate (0.01), few training epochs (around 50 epochs), large batch size (64), and does not use the early stopping mechanism; the accurate recognition model uses a small learning rate (0.001), many training epochs (more than 100 epochs), small batch size (32), and uses the early stopping mechanism to prevent overfitting. ...I will not go into detail here.

[0051] S104. Based on the identification results of each material to be sorted, query the sorting rule database and match the sorting parameters corresponding to each material to be sorted. The sorting parameters include sorting angle, sorting air pressure and hopper position. Among them, the sorting rule database refers to a structured data set that stores the configuration rules for material sorting parameters, including sorting parameter templates and matching rules for different types of materials; the sorting angle refers to the direction angle of the airflow injected by the pneumatic actuator, used to ensure that the material is blown into the correct hopper; the sorting air pressure refers to the airflow pressure generated by the pneumatic actuator, which needs to be adjusted according to the quality and size characteristics of the material (the quality and size characteristics are obtained from the sorting rule database based on the material type); the hopper location refers to the spatial coordinates of the hopper used to collect specific types of materials next to the conveyor belt; and the sorting parameters refer to the set of key control parameters that control the pneumatic actuator to sort materials.

[0052] Specifically, firstly, the sorting system determines the material category based on the identification results, queries the hopper allocation scheme in the sorting rule database, and determines the target hopper location for each material to be sorted. Then, based on the relative positional relationship between the material's lateral center position and the target hopper location, combined with the material's movement on the conveyor belt, the sorting system calculates the optimal sorting angle to ensure the material accurately enters the target hopper. Simultaneously, the sorting system matches the most suitable sorting air pressure from the sorting rule database based on the material's quality characteristics (such as weight and density) and dimensional characteristics (such as volume and shape). This ensures sufficient force to blow the material off the conveyor belt while avoiding excessive air pressure that could cause splashing. The sorting system combines the matched sorting angle, sorting air pressure, and hopper location parameters to form complete sorting parameters.

[0053] Optionally, under normal circumstances, based on the identification results of each material to be sorted, the sorting rule database is queried to match the sorting parameters corresponding to each material to be sorted. The sorting parameters include sorting angle, sorting air pressure, and hopper position. These can be achieved in the following ways, without limitation: Based on the identification results of each material to be sorted, the hopper position corresponding to each material to be sorted is queried from the preset sorting rule database; the sorting angle is determined based on the relative positional relationship between the hopper position and the lateral center position of the material to be sorted; and the corresponding sorting air pressure is matched from the sorting rule database based on the quality and size characteristics of the material to be sorted.

[0054] S105. Determine the material sorting execution sequence based on sorting parameters, conveyor belt speed, and the location information of each material to be sorted. Among them, the conveyor belt speed represents the linear motion speed of the conveyor belt, which can be calculated by encoder signals; the position information refers to the real-time spatial coordinates of the materials to be sorted on the conveyor belt; the material sorting execution sequence represents the time sequence and control parameter sequence for controlling the pneumatic actuator to sort each material.

[0055] Specifically, firstly, the sorting system calculates the estimated time when each piece of material to be sorted will reach the air nozzle position based on real-time measured conveyor belt speed and longitudinal position information of the materials to be sorted, combined with the distance from the conveyor belt to the air nozzle position. Secondly, based on the lateral position information of the materials to be sorted and the predetermined sorting angle, the sorting system calculates the number and specific locations of the air nozzles to be activated, typically activating multiple adjacent air nozzles to form a sufficiently wide airflow field. Simultaneously, the sorting system calculates the optimal execution duration of the air nozzles based on the sorting air pressure and the quality characteristics of the materials to be sorted, ensuring that the materials to be sorted are completely blown off the conveyor belt while avoiding unnecessary waste of compressed air. Finally, the sorting system organizes the estimated trigger time, air nozzle opening configuration, and execution duration of each material to be sorted in chronological order to form a complete sorting execution sequence and optimizes it to handle possible timing conflicts.

[0056] Optionally, under normal circumstances, the material sorting execution sequence can be determined based on sorting parameters, conveyor belt speed, and the position information of each material to be sorted, in the following ways, without limitation: Calculate the estimated time when each material to be sorted will arrive at the air nozzle position based on the conveyor belt speed and the longitudinal center position of each material to be sorted; determine the number of air nozzle opening groups and the opening position based on the lateral center position and sorting angle of each material to be sorted; calculate the execution duration of the air nozzles based on the sorting air pressure and the quality characteristics of each material to be sorted; combine the estimated time, number of opening groups, opening position, and execution duration to generate the material sorting execution sequence.

[0057] Among them, the estimated time refers to the time when the material is expected to arrive at the nozzle position; the number of opening groups indicates the number of nozzles that need to be started at the same time; the opening position refers to the specific position of the nozzle that needs to be started in the nozzle array; and the execution duration indicates the length of time the nozzle remains in the open state.

[0058] S106. According to the material sorting execution sequence, control the pneumatic actuator to sort each material to be sorted.

[0059] Among them, the material sorting execution sequence refers to the time sequence and parameter configuration for controlling the pneumatic actuator to perform sorting operations; the pneumatic actuator refers to an actuator consisting of multiple independently controllable air nozzles, used to generate directional airflow to blow materials into the target silo.

[0060] Specifically, firstly, the sorting system converts the material sorting execution sequence into a specific sequence of control parameters, including the trigger time, air pressure, and duration of each air nozzle. Then, the sorting system uses a real-time clock for precise timing; when material is about to reach the predetermined position, it controls the corresponding air nozzle to open according to the timing sequence, while simultaneously adjusting the air pressure valve to output the specified air pressure. The sorting system monitors the working status of the pneumatic actuators in real time to ensure that the control signals are executed correctly. After the material is blown off the conveyor belt, the sorting system closes the air nozzles for the predetermined duration, completing the sorting operation for a single material. For multiple materials being continuously conveyed, the sorting system strictly follows the material sorting execution sequence, coordinating the operation of multiple air nozzles to ensure the continuity and accuracy of the sorting process. If an execution anomaly is detected, the sorting system immediately handles and records the fault, and if necessary, activates a backup air nozzle or adjusts the control parameters.

[0061] By adopting the above technical solution, the sorting system collects hyperspectral data of different types of materials on the conveyor belt. Combining the differences in spectral response of different material types at multiple wavelengths, it selects the characteristic wavelengths that contribute most to material classification, extracts the characteristic spectral response values ​​of the materials to be sorted at these wavelengths, determines the lateral and longitudinal center positions of the materials, generates a feature dataset, and inputs it into the recognition model for identification. Based on the recognition results, it queries the sorting rule database to determine the corresponding sorting parameters for each material, ultimately achieving accurate material sorting. Compared to traditional sorting methods that rely solely on RGB images, this method can capture the spectral characteristics of the material surface, is unaffected by surface contamination, fading, reflection, and shadows, and can also distinguish materials that look similar but have different materials, greatly improving the accuracy and stability of sorting.

[0062] The following provides a more detailed description of the process of the method provided in this implementation. Please refer to [link / reference]. Figure 2 This is another flowchart illustrating the hyperspectral-based multi-material sorting method in this application.

[0063] The following steps may or may not be performed after step S106; this is not limited here: S201. Obtain the actual sorting trajectory and expected sorting trajectory of each material to be sorted.

[0064] The actual sorting trajectory refers to the actual movement path of the material to be sorted after it is blown away from the conveyor belt by the airflow, which can be obtained by a high-speed camera or trajectory tracking sensor; the expected sorting trajectory refers to the theoretical movement path of the material to be sorted calculated based on sorting parameters and physical motion model.

[0065] Specifically, firstly, the sorting system uses a high-speed camera array installed in the sorting area to continuously acquire motion image sequences of the materials being sorted after they are blown off the conveyor belt at a high frame rate (e.g., 1000fps). Simultaneously, the system uses multiple photoelectric sensor arrays to detect the spatiotemporal position of the materials. The system synchronizes the image data and sensor data in time and registers them spatially, then uses a trajectory reconstruction algorithm to obtain the actual sorting trajectory of the materials. At the same time, based on the initial state (position, attitude, etc.), sorting parameters (air pressure, angle, etc.), and environmental parameters (gravity, air resistance, etc.), the system calculates the expected motion trajectory of the materials using a physical motion model. The system then registers and interpolates the actual and expected sorting trajectories to ensure that the two sets of trajectory data have the same sampling point distribution, facilitating subsequent deviation analysis.

[0066] S202. When the deviation between the actual sorting trajectory and the expected sorting trajectory exceeds the preset deviation threshold, the sorting status of the target material to be sorted is determined as sorting failure.

[0067] Among them, deviation represents the spatial distance difference between the actual sorting trajectory and the expected sorting trajectory, which can be calculated using Euclidean distance or other measurement methods; the preset deviation threshold refers to the maximum allowable trajectory deviation, and sorting is considered to have failed if the deviation exceeds the preset deviation threshold.

[0068] Specifically, firstly, the sorting system calculates the spatial distance between the actual and expected positions at each sampling point, obtaining a time series of trajectory deviations. Then, the sorting system compares the calculated trajectory deviations with a preset deviation threshold (the preset deviation threshold is set considering factors such as material type and sorting speed). If the trajectory deviation exceeds the preset deviation threshold at any given time, or if the cumulative value of the trajectory deviation exceeds the allowable range, the sorting system marks the sorting status of the material to be sorted as a sorting failure. For materials determined to have failed to sort, the sorting system records its complete trajectory data, sorting parameters, and environmental conditions for subsequent failure analysis and parameter optimization.

[0069] S203. Based on the target material information of the target material to be sorted, perform secondary sorting on the target material to be sorted. The target material information includes the target material type and target location information.

[0070] Among them, target material information refers to the set of characteristic data of target materials that failed to be sorted, including target material type and target location information; target material type refers to the type of target materials to be sorted; target location information represents the current spatial coordinates of the target materials to be sorted; secondary sorting means re-sorting the target materials that failed to be sorted.

[0071] Specifically, firstly, the sorting system uses a sensor network to locate the target position of the failed sorting material and determine its complete characteristics, including type, quality, and size. Then, the system analyzes possible causes of the initial sorting failure, such as insufficient air pressure, angle deviation, or timing delay, and optimizes the sorting parameters accordingly. The system then replans the secondary sorting scheme, which may include adjusting air pressure, changing the sorting angle, or selecting different nozzle combinations. Simultaneously, the system may activate backup sorting channels or adopt special sorting modes to improve the success rate of the secondary sorting.

[0072] The sorting system in the embodiments of this invention is described below from the perspective of hardware processing. Please refer to [link / reference]. Figure 3 This is a schematic diagram of the physical device structure of a sorting system in an embodiment of this application.

[0073] It should be noted that, Figure 3 The structure of the sorting system shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0074] like Figure 3 As shown, the sorting system includes a CPU 301, which can perform various appropriate actions and processes based on a program stored in the read-only memory ROM 302 or a program loaded from the storage section 308 into the random access memory RAM 303, such as performing the methods described in the above embodiments. The RAM 303 also stores various programs and data required for system operation. The CPU 301, ROM 302, and RAM 303 are interconnected via a bus 304. An I / O interface 305 is also connected to the bus 304.

[0075] The following components are connected to I / O interface 305: input section 306 including audio input devices, push-button switches, etc.; output section 307 including a liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 308 including a hard disk, etc.; and communication section 309 including a network interface card such as a LAN (Local Area Network) card, modem, etc. Communication section 309 performs communication processing via a network such as the Internet. Drive 310 is also connected to I / O interface 305 as needed. Removable media 311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 310 as needed so that computer programs read from them can be installed into storage section 308 as needed.

[0076] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311. When the computer program is executed by CPU 301, it performs the various functions defined in the present invention.

[0077] It should be noted that specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0078] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those shown in the drawings.

[0079] Specifically, the sorting system in this embodiment includes a processor and a memory. The memory stores a computer program, and when the computer program is executed by the processor, it implements the hyperspectral-based multi-material sorting method provided in the above embodiment.

[0080] In another aspect, the present invention also provides a computer-readable storage medium, which may be included in the sorting system described in the above embodiments; or it may exist independently and not assembled into the sorting system. The storage medium carries one or more computer programs that, when executed by a processor of the sorting system, cause the sorting system to implement the hyperspectral-based multi-material sorting method provided in the above embodiments.

[0081] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0082] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as meaning "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0083] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for sorting multiple types of materials based on hyperspectral imaging, characterized in that, Applied to a sorting system, the method includes: The system collects the position signal of the conveyor belt encoder and hyperspectral data of different types of materials on the conveyor belt. The position signal of the conveyor belt encoder is used to determine the position information of the different types of materials on the conveyor belt. The hyperspectral data refers to the spectral response values ​​of multiple wavelength points continuously collected within a preset wavelength range at preset wavelength intervals. Analyze the differences in spectral response of different types of materials at multiple wavelength points, calculate the contribution of each wavelength point to material classification, and determine the wavelength points with the highest contribution as characteristic wavelength points. The characteristic spectral response values ​​of the materials to be sorted on the conveyor belt at the characteristic wavelength points are extracted from the hyperspectral data, and the lateral center position and longitudinal center position of the materials to be sorted on the conveyor belt are determined based on the position information to generate a feature dataset. The feature dataset is then input into the recognition model to obtain the recognition results of each material to be sorted. Based on the identification results of each material to be sorted, the sorting rule database is queried to match the sorting parameters corresponding to each material to be sorted. The sorting parameters include sorting angle, sorting air pressure and hopper position. Based on the sorting parameters, conveyor belt speed, and the location information of each material to be sorted, the material sorting execution sequence is determined; According to the material sorting execution sequence, the pneumatic actuator is controlled to sort each material to be sorted.

2. The method according to claim 1, characterized in that, The analysis of the spectral response differences of different types of materials at multiple wavelength points, and the calculation of the contribution of each wavelength point to material classification, specifically includes: Based on a pre-defined material spectral feature library, the average response value corresponding to different types of materials at each wavelength point is determined, and the overall average response value of all types of materials at each wavelength point is calculated. The average of the squared differences between the average response values ​​of different types of materials at each wavelength point and the overall average value is calculated to obtain the inter-class variance of each wavelength point. Based on the material spectral feature library, the response standard deviation corresponding to different types of materials at each wavelength point is determined, the average value of the response standard deviation is calculated, and the intra-class variance of each wavelength point is obtained. The ratio of the inter-class variance to the intra-class variance at each wavelength point is determined as the discrimination index at each wavelength point. The contribution of each wavelength point is obtained by weighted summation of the discrimination and signal-to-noise ratio.

3. The method according to claim 2, characterized in that, The calculation of the signal-to-noise ratio at each wavelength point specifically includes: Background spectral data is collected when there is no material on the conveyor belt to determine multiple background response values ​​for each wavelength point; The arithmetic mean and standard deviation of the multiple background response values ​​at each wavelength point are calculated to obtain the background reference value and background noise value at each wavelength point; Reference spectral data were collected when a standard reflectance whiteboard was placed on the conveyor belt to determine multiple reference response values ​​for each wavelength point; The arithmetic mean of the plurality of reference response values ​​at each wavelength point is calculated to obtain the reference signal value at each wavelength point; Substituting the reference signal value, the background reference value, and the background noise value into the signal-to-noise ratio calculation formula, the signal-to-noise ratio at each wavelength point is obtained.

4. The method according to claim 1, characterized in that, The step of inputting the feature dataset into the recognition model to obtain the recognition results for each material to be sorted specifically includes: The feature dataset is input into the fast recognition model to obtain the initial recognition results and initial confidence levels of each material to be sorted. If the initial confidence level is greater than or equal to a preset threshold, the initial identification result is determined to be the identification result of each material to be sorted. If the initial confidence level is less than the preset threshold, the feature dataset is input into the accurate recognition model to obtain the final recognition result and the final confidence level. The final recognition result is determined to be the recognition result of each material to be sorted. The computational complexity of the fast recognition model is lower than that of the accurate recognition model.

5. The method according to claim 1, characterized in that, The process involves querying a sorting rule database based on the identification results of each material to be sorted, and matching the sorting parameters corresponding to each material. These sorting parameters include sorting angle, sorting air pressure, and hopper location, specifically including: Based on the identification results of each material to be sorted, the location of the corresponding silo for each material to be sorted is queried from the preset sorting rule database; The sorting angle is determined based on the relative positional relationship between the location of the hopper and the lateral center of the material to be sorted. Based on the quality and size characteristics of the materials to be sorted, the corresponding sorting air pressure is obtained by matching from the sorting rule database.

6. The method according to claim 5, characterized in that, The determination of the material sorting execution sequence based on the sorting parameters, conveyor belt speed, and the position information of each material to be sorted specifically includes: Based on the conveyor belt speed and the longitudinal center position of each material to be sorted, calculate the estimated time when each material to be sorted will reach the air nozzle position; Based on the lateral center position of each material to be sorted and the sorting angle, determine the number of air nozzles to be opened and the opening position. The execution duration of the air nozzle is calculated based on the sorting air pressure and the quality characteristics of each material to be sorted. The material sorting execution sequence is generated by combining the expected time, the number of activation groups, the activation position, and the execution duration.

7. The method according to claim 1, characterized in that, After the step of controlling the pneumatic actuator to sort each material to be sorted according to the material sorting execution sequence, the method further includes: Obtain the actual sorting trajectory and expected sorting trajectory of each material to be sorted; When the deviation between the actual sorting trajectory and the expected sorting trajectory exceeds a preset deviation threshold, the sorting status of the target material to be sorted is determined to be sorting failure. Based on the target material information of the target material to be sorted, the target material to be sorted is sorted a second time. The target material information includes the target material type and target location information.

8. A sorting system, characterized in that, The sorting system includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the sorting system to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is executed on the sorting system, the sorting system performs the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product is run on the sorting system, the sorting system performs the method as described in any one of claims 1-7.

Images