Intelligent garbage classification system based on X-ray transmission and deep learning

Abstract

The invention relates to the technical field of image recognition, in particular to an intelligent garbage classification system based on X-ray transmission and deep learning. A scanning module of the system adopts multi-angle dual-energy X-rays to obtain high-energy and low-energy projection data, and reconstructs a cross section attenuation image through a filtering back projection algorithm; the image processing module calculates an effective atomic number of a pixel point by using a dual-energy imaging algorithm, and performs mapping based on a color model so as to realize material distinguishing of the organic matter, the mixture and the inorganic matter; the garbage identification module inputs the mapping transmission image into a deep learning model, completes identification of specific categories of metal, plastic and the like, and outputs a classification instruction in combination with a garbage classification library; and the garbage sorting unit moves the garbage carrier to a collection area according to the instruction. The data management unit stores the transmission image and the classification result, and continuously optimizes the recognition model through incremental learning. According to the invention, transmission type identification and intelligent classification of garbage can be realized, and the classification precision and the system stability are improved.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a smart waste sorting system based on X-ray transmission and deep learning. Background Technology

[0002] In recent years, with the continuous increase in the amount of household waste generated, waste sorting has gradually become an important part of urban management and resource recycling. Although the state and local governments have introduced relevant regulations and standards, the implementation of waste sorting is still not ideal. On the one hand, residents' participation in the disposal process is insufficient, and the complex and costly sorting operations result in low-quality sorting at the front end. On the other hand, in the back-end processing stage, problems such as mixed loading and transportation during transportation and disposal further reduce sorting efficiency.

[0003] To improve waste disposal efficiency, existing technologies have proposed various automated sorting devices, most of which use optical cameras, weight sensors, or infrared sensors for identification and sorting. While these devices reduce human intervention to some extent, they still have significant shortcomings: First, optical and infrared sensors rely solely on surface features and cannot penetrate garbage bags to identify internal objects, resulting in low recognition rates for obscured or compressed waste. Second, for irregularly shaped, overlapping, or complex-material waste, existing algorithms struggle to guarantee recognition accuracy, easily leading to sorting errors. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a smart waste sorting system based on X-ray transmission and deep learning. This system employs low-dose X-ray dual-energy imaging technology to perform multi-angle transmission scanning of waste bags, generating cross-sectional transmission images. It then uses a color mapping model to distinguish between organic matter, mixtures, and inorganic matter. Based on this, a deep learning algorithm is combined to identify and classify the specific categories of waste, thereby outputting sorting instructions to drive the sorting mechanism to complete the automated waste sorting.

[0005] In a first aspect, the present invention provides a waste intelligent sorting system based on X-ray transmission and deep learning, comprising:

[0006] The user verification module is used to identify the user who disposes of garbage and control the opening and closing of the garbage disposal port.

[0007] The X-ray transmission scanning module is used to perform multi-angle dual-energy X-ray scanning on the garbage bags to obtain high-energy and low-energy projection data, and reconstruct the cross-sectional attenuation image through a filtered back-projection algorithm.

[0008] The image processing module is used to calculate the effective atomic number of pixels in the cross-sectional transmission image using a dual-energy imaging algorithm, and to perform visualization mapping based on a preset color model to obtain the mapped transmission image.

[0009] The waste identification module is used to distinguish the material type of waste in the mapped transmission image, and analyze the distinction results through a deep learning model to identify the specific category of waste and make a classification judgment to output classification instructions;

[0010] The waste sorting unit is used to move waste carriers to the corresponding waste collection area according to the sorting instructions;

[0011] The data management unit is used to store transmission images, classification records, and user data, and to perform incremental updates to the deep learning model based on the stored data to improve the accuracy of waste classification and identification.

[0012] Preferably, the user verification module includes at least one of a face recognition unit, an NFC sensing unit, a QR code recognition unit, a camera unit, and a human body sensing unit, used for identity verification when a user disposes of garbage.

[0013] Preferably, the filtering back projection algorithm steps include:

[0014] Perform Fourier transform on the projection data from multiple angles to obtain the frequency domain representation of each projection data;

[0015] Convolve each projection frequency domain representation with the filter to enhance high-frequency information and reduce reconstruction artifacts;

[0016] Backprojection is performed on the filtered projection data, and the data are accumulated along the scanning angle direction to form a preliminary reconstructed image.

[0017] The preliminary reconstructed image is normalized to obtain the final cross-sectional transmission image.

[0018] Preferably, the mapping rules of the color model include:

[0019] If the effective atomic number of a pixel is less than 10, then the color of that pixel is mapped to orange.

[0020] If the effective atomic number of a pixel is between 10 and 18, then the color of that pixel is mapped to green.

[0021] If the effective atomic number of a pixel is greater than 18, then the color of that pixel is mapped to blue.

[0022] Preferably, the rules for classifying waste materials include:

[0023] Pixels of the same color in the mapped transmission image are aggregated to extract the contour of each individual object;

[0024] The material category of an object is determined based on the color of the pixels within the outline. Orange corresponds to objects classified as organic materials, green corresponds to objects classified as mixtures or light metals, and blue corresponds to objects classified as inorganic materials.

[0025] Preferably, the steps of identifying the specific category of waste and classifying it include:

[0026] The cross-sectional transmission image obtained based on material category is input into the deep learning model;

[0027] Based on a deep learning model, target detection and classification are performed on cross-sectional transmission images, and the specific category of waste is output, including metal, plastic, glass and other preset categories.

[0028] Based on a pre-set waste sorting database, specific types of waste are mapped to corresponding major waste categories, including recyclable waste, kitchen waste, combustible waste, or hazardous waste.

[0029] Preferably, the incremental learning step includes:

[0030] A training sample set is constructed based on the stored cross-sectional transmission images and the corresponding classification results;

[0031] Optimize deep learning model parameters using online learning or micro-batch training methods;

[0032] Once the model accuracy meets the preset threshold, the updated model is deployed to the waste identification module to improve the accuracy of subsequent classification.

[0033] Preferably, the system further includes a user interaction module equipped with a touch screen and voice prompts, used to provide operation guidance to users during the garbage disposal process and to display the corresponding classification results after the classification is completed.

[0034] Secondly, the present invention provides a waste intelligent classification method based on X-ray transmission and deep learning, for use in any of the above-mentioned waste intelligent classification systems, the method comprising:

[0035] S100: Identifies users who dispose of garbage and controls the opening and closing of the garbage disposal inlet;

[0036] S200: Performs multi-angle dual-energy X-ray scanning on the garbage bags to obtain high-energy and low-energy projection data, and reconstructs the cross-sectional attenuation image through a filtered back-projection algorithm;

[0037] S300: Calculate the effective atomic number of pixels in the cross-sectional transmission image using a dual-energy imaging algorithm, and perform visualization mapping based on a preset color model to obtain the mapped transmission image;

[0038] S400: Distinguishes waste material categories from the mapped transmission image, analyzes the distinction results using a deep learning model, identifies the specific category of waste, makes a classification judgment, and outputs classification instructions;

[0039] S500: Move the waste carrier to the corresponding waste collection area according to the classification instructions;

[0040] S600: Stores transmission images, classification records, and user data, and performs incremental updates to the deep learning model based on the stored data to improve the accuracy of waste classification and identification;

[0041] During the waste disposal process from S100 to S600, user guidance is provided, and the corresponding sorting results are displayed after sorting is completed.

[0042] The present invention has the following beneficial effects:

[0043] 1. In this invention, dual-energy X-ray imaging technology is used to perform multi-angle transmission scanning of garbage bags. A cross-sectional image is reconstructed using a filtered back-projection algorithm, and color mapping is performed based on the effective atomic number, thereby effectively distinguishing organic matter, mixtures, and inorganic matter. This solution overcomes the limitation of traditional optical recognition methods that cannot penetrate garbage bags, improving the accuracy of material identification.

[0044] 2. In this invention, a deep learning model is used to train and recognize transmission images, enabling accurate identification and classification of specific categories such as metal, plastic, and glass in complex scenarios. This solution not only supports simultaneous detection of multiple categories but also has universality in handling special situations such as occlusion, compression, and liquid leakage, further improving the precision of waste sorting.

[0045] 3. In this invention, the data management unit supports the storage of transmission images and classification results, and uses this data to incrementally update the deep learning model. Through online learning or micro-batch training, the model can continuously optimize its recognition accuracy and adaptability over long-term operation. This mechanism enables the system to dynamically adapt to new types of waste and complex scenarios, ensuring the continuous accuracy and stability of waste classification. Attached Figure Description

[0046] Figure 1 This is a structural diagram of a waste intelligent sorting system based on X-ray transmission and deep learning proposed in this invention;

[0047] Figure 2This is a flowchart of a waste intelligent classification method based on X-ray transmission and deep learning proposed in this invention. Detailed Implementation

[0048] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Example 1

[0050] In a first embodiment of the present invention, the present invention provides a waste intelligent sorting system based on X-ray transmission and deep learning, such as... Figure 1 As shown, it includes:

[0051] The user verification module is used to identify the user who disposes of garbage and control the opening and closing of the garbage disposal port.

[0052] Preferably, the user verification module includes at least one of a face recognition unit, an NFC sensing unit, a QR code recognition unit, a camera unit, and a human body sensing unit, used for identity verification when a user disposes of garbage.

[0053] Specifically, when the human body sensor detects a user approaching, the system automatically enters a verification state and wakes up the camera unit or sensing device. If the user chooses facial recognition, the camera unit captures the user's facial image and compares it with user information in the database using a built-in facial recognition algorithm. If a match is found, the user's identity is confirmed. If the user is carrying an NFC card, the user simply needs to bring the card close to the sensing area, and the system reads the card information and compares it with account information in the database to complete identity verification. If the user chooses QR code recognition, a QR code is generated through the mobile application, scanned by the QR code recognition unit, and matched with account information returned by the server to complete verification.

[0054] When identity verification is successful, the system controls the automatic opening of the garbage disposal compartment door and prompts "Please dispose of garbage" on the user interface; when identity verification fails, the system prompts "Authentication failed, please re-authenticate" and keeps the disposal compartment closed to ensure system security and data uniqueness.

[0055] Through the above process, the user verification module can achieve identity recognition in multiple ways, ensuring system security while improving user experience.

[0056] The X-ray transmission scanning module is used to perform multi-angle dual-energy X-ray scanning on the garbage bags to obtain high-energy and low-energy projection data, and reconstruct the cross-sectional attenuation image through a filtered back-projection algorithm.

[0057] Preferably, the filtering back projection algorithm steps include:

[0058] Perform Fourier transform on the projection data from multiple angles to obtain the frequency domain representation of each projection data;

[0059] Convolve each projection frequency domain representation with the filter to enhance high-frequency information and reduce reconstruction artifacts;

[0060] Backprojection is performed on the filtered projection data, and the data are accumulated along the scanning angle direction to form a preliminary reconstructed image.

[0061] The preliminary reconstructed image is normalized to obtain the final cross-sectional transmission image.

[0062] Specifically, the X-ray transmission scanning module scans the garbage bag from multiple angles to acquire projection data at different angles. In one feasible implementation, the dual-energy X-ray emitter rotates 180° around the bag, acquiring a pair of projection data for every 1.5° rotation. This projection data is stored in the form of a two-dimensional matrix, with each row or column corresponding to a different scanning angle. The projection data at each angle are then subjected to a one-dimensional Fourier transform to obtain the corresponding frequency domain representation.

[0063] Pre-defined filter functions, such as the Lamb filter, Shepp-Logan filter, or Hamming window function filter, are applied to the frequency domain representation. The purpose of the filter is to enhance the high-frequency components of the projected data, thereby reducing blurring and fringe artifacts that may occur during backprojection.

[0064] The filtered projection data undergoes backprojection processing, which involves projecting each projection signal in reverse within a two-dimensional space according to the acquisition angle. As the angles accumulate, a preliminary reconstructed image of the cross-section is gradually formed.

[0065] The initial reconstructed image undergoes pixel intensity normalization, adjusting pixel values ​​to a uniform range (e.g., a grayscale range of 0–255) to eliminate brightness differences caused by the superposition of projection data from different angles. The normalized image is the cross-sectional transmission image, which has good clarity and contrast and can be used as input data for subsequent dual-energy imaging analysis and deep learning recognition.

[0066] Through the above process, the filtering back projection algorithm can effectively suppress artifacts and improve image clarity, thus providing a reliable image basis for the identification and classification of waste materials.

[0067] It is important to note that, considering radiation safety, the outer wall of the X-ray transmission scanning module's scanning chamber is shielded with a lead liner to ensure that no radiation leakage occurs to the external environment during the scanning process. Furthermore, if a living organism is detected within the scanning chamber, the controller will immediately terminate X-ray emission, stop the scanning operation, and trigger an audible and visual alarm.

[0068] The image processing module is used to calculate the effective atomic number of pixels in the cross-sectional transmission image using a dual-energy imaging algorithm, and to perform visualization mapping based on a preset color model to obtain the mapped transmission image.

[0069] Preferably, the mapping rules of the color model include:

[0070] If the effective atomic number of a pixel is less than 10, then the color of that pixel is mapped to orange.

[0071] If the effective atomic number of a pixel is between 10 and 18, then the color of that pixel is mapped to green.

[0072] If the effective atomic number of a pixel is greater than 18, then the color of that pixel is mapped to blue.

[0073] Specifically, a dual-energy imaging algorithm is used to compare high-energy and low-energy cross-sectional transmission images. Based on the attenuation coefficient of pixels at different energies, the effective atomic number corresponding to each pixel is calculated. The system pre-defines three mapping intervals: when the effective atomic number of a pixel is <10, it is marked as an organic material and assigned an orange pixel value; when the effective atomic number of a pixel is ∈ [10, 18], it is marked as a mixture or light metal and assigned a green pixel value; when the effective atomic number of a pixel is >18, it is marked as an inorganic material and assigned a blue pixel value. In the specific implementation, the system maps each pixel using a lookup table function or conditional statement and generates a corresponding color matrix. The color matrix is ​​mapped to the coordinates of the original cross-sectional image to output a pseudo-color image, i.e., the mapped transmission image. This image can intuitively distinguish object regions of different materials on the display terminal, facilitating subsequent contour extraction and deep learning recognition.

[0074] Through the above process, color mapping not only improves the visualization of images, but also provides intuitive features for the automatic differentiation of waste materials, ensuring that deep learning models can accurately identify different types of waste.

[0075] Preferably, the rules for classifying waste materials include:

[0076] Pixels of the same color in the mapped transmission image are aggregated to extract the contour of each individual object;

[0077] The material category of an object is determined based on the color of the pixels within the outline. Orange corresponds to objects classified as organic materials, green corresponds to objects classified as mixtures or light metals, and blue corresponds to objects classified as inorganic materials.

[0078] Specifically, the system receives a mapped transmission image, in which the color of each pixel has been mapped according to its effective atomic number. Pixels with the same color are aggregated, that is, adjacent pixels with the same color are combined into a single connected region, forming a preliminary object region. The aggregation method can employ connected component labeling algorithms, region growing algorithms, or other commonly used image segmentation techniques. For each aggregated object region, its contour information is extracted. The contour can be obtained using edge detection algorithms, such as the Canny operator, the Sobel operator, or pixel sequences based on region boundaries. The extracted contour is used to represent the shape, size, and spatial location of each individual object.

[0079] The system determines the material type of an object based on the pixel color within its outline region: objects with orange outlines are classified as organic materials; objects with green outlines are classified as mixtures or light metals; and objects with blue outlines are classified as inorganic materials. The system can also calculate the pixel percentage and average color of each outline region to ensure accuracy.

[0080] In this way, the system can effectively distinguish objects of different materials within complex garbage bags, providing a reliable basis for garbage classification.

[0081] The waste identification module is used to distinguish the material type of waste in the mapped transmission image, and analyze the distinction results through a deep learning model to identify the specific category of waste and make a classification judgment to output classification instructions;

[0082] Preferably, the steps of identifying the specific category of waste and classifying it include:

[0083] The cross-sectional transmission image obtained based on material category is input into the deep learning model;

[0084] Based on a deep learning model, target detection and classification are performed on cross-sectional transmission images, and the specific category of waste is output, including metal, plastic, glass and other preset categories.

[0085] Based on a pre-set waste sorting database, specific types of waste are mapped to corresponding major waste categories, including recyclable waste, kitchen waste, combustible waste, or hazardous waste.

[0086] Specifically, the system first acquires a cross-sectional transmission image based on material category differentiation. The individual object regions in this image have already been obtained through color mapping and contour extraction. This image is used as input data for a deep learning model to ensure that the model can perform object-level recognition. The deep learning model can employ Faster R-CNN, YOLOv5, or VIT object detection models; in a feasible implementation, YOLOv5 is preferred. The model input is the cross-sectional transmission image, and the output is the bounding box and specific category label for each object.

[0087] Specific categories include at least: metals, plastics, glass, paper, and organic residues, and can be extended to include hazardous materials such as batteries and lighters. To improve model robustness, simulated complex scenarios such as occlusion, compression, and liquid leakage were added to the training data to ensure the model's accuracy in identifying waste under actual disposal conditions.

[0088] Based on a pre-set waste classification database, the system maps the specific categories output by the model to the major waste categories in national or regional standards: metal, plastic, glass, and paper → recyclable waste; organic residue → kitchen waste; wood products and some plastic products → combustible waste; batteries, medicines, lighters, etc. → hazardous waste. This mapping method achieves hierarchical classification from "material category → specific category → major waste category".

[0089] By introducing a deep learning model, the specific category of waste can be accurately identified in complex scenarios, and the classification library mapping ensures that the system's output meets the waste classification standards.

[0090] The waste sorting unit is used to move waste carriers to the corresponding waste collection area according to the sorting instructions;

[0091] Specifically, after the waste identification module in the system identifies and classifies the specific type of waste, it outputs a corresponding classification instruction. This instruction includes at least the target waste type and the location information of the target collection area. Upon receiving the classification instruction, the waste sorting unit parses the target area information, such as "recyclable waste area," "kitchen waste area," "combustible waste area," or "hazardous waste area." Based on the preset coordinates or markers of the waste collection area, the system determines the target location for the waste carrier. The waste sorting unit controls the waste carrier to move along a set path to the target location. In one feasible implementation, the waste carrier can be a tray-type structure. The sorting unit maintains the integrity of the waste bag during movement to prevent secondary pollution caused by bag damage. When the waste carrier moves above the target collection area, the system performs a disposal operation, placing the waste into the corresponding waste bin or collection container. After disposal, the waste carrier automatically returns to its initial position, awaiting the next sorting task.

[0092] The data management unit is used to store transmission images, classification records, and user data, and to perform incremental updates to the deep learning model based on the stored data to improve the accuracy of waste classification and identification.

[0093] Preferably, the incremental learning step includes:

[0094] A training sample set is constructed based on the stored cross-sectional transmission images and the corresponding classification results;

[0095] Optimize deep learning model parameters using online learning or micro-batch training methods;

[0096] Once the model accuracy meets the preset threshold, the updated model is deployed to the waste identification module to improve the accuracy of subsequent classification.

[0097] Specifically, the system pairs stored cross-sectional transmission images with corresponding classification results to form a training sample set. The classification results can come from labels automatically identified by the system or combine manual correction results to ensure the accuracy of sample labeling. The training sample set can cover common waste categories as well as some complex scenarios, such as situations where garbage bags are obstructed, compressed, or leaking liquid.

[0098] The data management unit optimizes the parameters of the deployed deep learning model based on the constructed sample set. Parameter optimization can be achieved through online learning, where the model is updated in real time as new samples arrive; or through micro-batch training, where a certain number of new samples are collected periodically and then the model is trained on a small scale. Through parameter optimization, the model's adaptability to new scenarios and categories can be gradually improved without compromising the stability of the original model parameters.

[0099] After incremental training is completed, the system evaluates the accuracy of the updated model using a validation set. If the evaluation result reaches or exceeds a preset threshold (e.g., classification accuracy reaches 95%), the updated model is automatically deployed to the waste identification module. Once deployed, the new model can be used for subsequent waste identification tasks, thereby improving the accuracy of waste classification.

[0100] Through the above process, the data management unit not only has data storage function, but also enables dynamic optimization of deep learning models, ensuring that the system can continuously adapt to changes in waste disposal types and interference from complex environments during long-term operation.

[0101] Preferably, the system also includes a user interaction module, equipped with a touch screen and voice prompts, to provide operation guidance to users during the waste disposal process and to display the corresponding classification results after the classification is completed.

[0102] Example 2

[0103] In real-world waste disposal scenarios, garbage bags often contain a mixture of materials that overlap and obscure each other. For example, plastic bottles may be covered with organic residue, or metal cans may be compressed inside plastic bags. In such cases, traditional waste sorting methods based on a single characteristic (such as weight, color, or human vision) struggle to achieve accurate identification, leading to significant deviations in sorting results and impacting waste recycling efficiency.

[0104] To solve the above problems, the present invention provides the following... Figure 2 The following describes a waste intelligent classification method based on X-ray transmission and deep learning. The specific implementation process of this method is as follows:

[0105] When a user disposes of trash, the system verifies their identity through a user authentication module. For example, it may use a facial recognition unit or an NFC sensor to verify user information. Once verification is successful, the trash can automatically opens, allowing the user to deposit a trash bag.

[0106] After the garbage bag enters the scanning area, the X-ray transmission scanning module activates the dual-energy X-ray source and acquires high-energy and low-energy projection data from multiple angles by rotating 180°. In one feasible approach, a pair of projection data is acquired every 1.5° rotation. The multi-angle projection data obtained from the scan is processed by a filtered back-projection algorithm, specifically including Fourier transform, frequency domain filtering, back-projection accumulation, and normalization, ultimately yielding a cross-sectional transmission image.

[0107] The system calculates the effective atomic number of each pixel based on high-energy and low-energy images and maps them according to a color model: atomic number < 10 → mapped to orange (organic matter); atomic number ∈ [10, 18] → mapped to green (light metals or mixtures); atomic number > 18 → mapped to blue (inorganic matter). This yields a mapped transmission image, facilitating subsequent material classification.

[0108] The system performs pixel clustering and contour extraction on the mapped transmission image, aggregating pixels of the same color to form independent object regions. Based on the dominant color within each region, it distinguishes material categories. The extracted object regions are then input into a deep learning model (such as a YOLOv5-based convolutional neural network) for object detection and specific category identification of waste. Output categories can include metal, plastic, glass, paper, etc.; the model also supports the identification of hazardous materials (such as batteries, sharp objects) and complex scenarios (occlusion, compression, liquid leakage). Based on the identified specific category and a pre-defined waste classification database, the system maps it to the corresponding major category.

[0109] After the classification instructions are generated, the waste sorting unit controls the mechanical actuators (such as conveyor belts or push rods) according to the instructions to transport the waste carriers to the corresponding collection areas.

[0110] Meanwhile, the user interaction module displays the classification results to the user on the screen and provides guidance on how to distribute the product.

[0111] The system stores the transmission image and classification results together. The data management unit performs incremental training based on these samples to update the deep learning model and improve classification accuracy in similar complex scenarios.

[0112] Through the above implementation process, even in complex scenarios where there are multiple materials mixed, obstructed, or compressed in the garbage bag, the system can still achieve high-precision automatic identification and classification, solving the technical problem that traditional methods are easily interfered with by surface features.

[0113] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A waste intelligent sorting system based on X-ray transmission and deep learning, characterized in that, include: The user verification module is used to identify the user who disposes of garbage and control the opening and closing of the garbage disposal port. The X-ray transmission scanning module is used to perform multi-angle dual-energy X-ray scanning on the garbage bags to obtain high-energy and low-energy projection data, and reconstruct the cross-sectional attenuation image through a filtered back-projection algorithm. The image processing module is used to calculate the effective atomic number of pixels in the cross-sectional transmission image using a dual-energy imaging algorithm, and to perform visualization mapping based on a preset color model to obtain the mapped transmission image. The waste identification module is used to distinguish the material type of waste in the mapped transmission image, and analyze the distinction results through a deep learning model to identify the specific category of waste and make a classification judgment to output classification instructions; The waste sorting unit is used to move waste carriers to the corresponding waste collection area according to the sorting instructions; The data management unit is used to store transmission images, classification records, and user data, and to perform incremental updates to the deep learning model based on the stored data to improve the accuracy of waste classification and identification.

2. The intelligent waste sorting system based on X-ray transmission and deep learning according to claim 1, characterized in that, The user verification module includes at least one of a face recognition unit, an NFC sensing unit, a QR code recognition unit, a camera unit, and a human body sensing unit, and is used to identify the user when disposing of garbage.

3. The intelligent waste sorting system based on X-ray transmission and deep learning according to claim 1, characterized in that, The filtering back projection algorithm includes the following steps: Perform Fourier transform on the projection data from multiple angles to obtain the frequency domain representation of each projection data; Convolve each projection frequency domain representation with the filter to enhance high-frequency information and reduce reconstruction artifacts; Backprojection is performed on the filtered projection data, and the data are accumulated along the scanning angle direction to form a preliminary reconstructed image. The preliminary reconstructed image is normalized to obtain the final cross-sectional transmission image.

4. The intelligent waste sorting system based on X-ray transmission and deep learning according to claim 1, characterized in that, The mapping rules of the color model include: If the effective atomic number of a pixel is less than 10, then the color of that pixel is mapped to orange. If the effective atomic number of a pixel is between 10 and 18, then the color of that pixel is mapped to green. If the effective atomic number of a pixel is greater than 18, then the color of that pixel is mapped to blue.

5. The intelligent waste sorting system based on X-ray transmission and deep learning according to claim 1, characterized in that, The rules for classifying waste materials include: Pixels of the same color in the mapped transmission image are aggregated to extract the contour of each individual object; The material category of an object is determined based on the color of the pixels within the outline. Orange corresponds to objects classified as organic materials, green corresponds to objects classified as mixtures or light metals, and blue corresponds to objects classified as inorganic materials.

6. The intelligent waste sorting system based on X-ray transmission and deep learning according to claim 1, characterized in that, The steps for identifying and classifying waste include: The cross-sectional transmission image obtained based on material category is input into the deep learning model; Based on a deep learning model, target detection and classification are performed on cross-sectional transmission images, and the specific category of waste is output, including metal, plastic, glass and other preset categories; Based on a pre-set waste sorting database, specific types of waste are mapped to corresponding major waste categories, including recyclable waste, kitchen waste, combustible waste, or hazardous waste.

7. The intelligent waste sorting system based on X-ray transmission and deep learning according to claim 1, characterized in that, The incremental learning steps include: A training sample set is constructed based on the stored cross-sectional transmission images and the corresponding classification results; Optimize deep learning model parameters using online learning or micro-batch training methods; Once the model accuracy meets the preset threshold, the updated model is deployed to the waste identification module to improve the accuracy of subsequent classification and judgment.

8. The intelligent waste sorting system based on X-ray transmission and deep learning according to claim 1, characterized in that, The system also includes a user interaction module, equipped with a touch screen and voice prompts, which provides operation guidance to users during the garbage disposal process and displays the corresponding classification results after the classification is completed.

9. A waste intelligent classification method based on X-ray transmission and deep learning, characterized in that, The method for the intelligent waste sorting system according to any one of claims 1-8 comprises: S100: Identifies users who dispose of garbage and controls the opening and closing of the garbage disposal chute; S200: Performs multi-angle dual-energy X-ray scanning on the garbage bags to obtain high-energy and low-energy projection data, and reconstructs the cross-sectional attenuation image through a filtered back-projection algorithm; S300: Calculate the effective atomic number of pixels in the cross-sectional transmission image using a dual-energy imaging algorithm, and perform visualization mapping based on a preset color model to obtain the mapped transmission image; S400: Distinguishes waste material categories from the mapped transmission image, analyzes the distinction results using a deep learning model, identifies the specific category of waste, makes a classification judgment, and outputs classification instructions; S500: Move the waste carrier to the corresponding waste collection area according to the classification instructions; S600: Stores transmission images, classification records, and user data, and performs incremental updates to the deep learning model based on the stored data to improve the accuracy of waste classification and identification; During the waste disposal process from S100 to S600, user guidance is provided, and the corresponding sorting results are displayed after sorting is completed.

Images