Intelligent treatment system for decoration garbage

Abstract

The invention relates to the technical field of automation, in particular to an intelligent treatment system for decoration garbage. The system integrates a preliminary screening module, a combined screening module, a magnetic separation module, a negative pressure winnowing module, a light separation module, a positive pressure winnowing module and a robot sorting module. Through combination of the 3D drum screen and the relaxation screen, the screening efficiency and precision are improved, and various decoration wastes including bagged decoration wastes and bulk decoration wastes are effectively treated; the magnetic separation and winnowing module is used for separating waste metal and light materials, and the light separation module is used for further subdividing the materials by utilizing a spectrum technology; the robot sorting module adopts a deep learning algorithm, various materials can be intelligently recognized and accurately classified, and the resource recycling rate is remarkably increased. The whole system realizes intellectualization and automation of decoration garbage treatment, and solves the problems of low screening efficiency and poor classification precision in the prior art.

Description

Technical Field

[0001] The invention relates to the field of automation technology, and in particular to an intelligent processing system for decoration waste. Background Art

[0002] With the acceleration of urbanization, the amount of decoration waste has increased dramatically, bringing great pressure to the environment. Traditional methods of disposing decoration waste are mainly landfill and incineration, which not only occupies a large amount of land resources, but also may cause secondary pollution. Although some existing waste disposal systems have achieved partial classification, they often have problems such as low screening efficiency and poor classification accuracy. Therefore, it is particularly important to develop an efficient, accurate and environmentally friendly decoration waste disposal system. Summary of the invention

[0003] In view of the above-mentioned technical deficiencies, the purpose of the present invention is to provide an intelligent processing system for decoration waste to solve the problems of low screening efficiency and poor classification accuracy in the prior art.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions: In a first aspect, an intelligent processing system for decoration waste includes: Preliminary screening module: including the first plate feeder, the second plate feeder, the first 3D drum screen and the second 3D drum screen, which are used to process the input decoration waste; bagged waste is input through the first plate feeder, and bulk waste is input through the second plate feeder. The input bagged waste is broken and screened through the first 3D drum screen to obtain materials of 0-100mm and above 100mm; 0-100mm materials are directly sent to the combined screening module; materials above 100mm are mixed with bulk waste and sent to the second 3D drum screen for further screening to obtain materials of 0-90mm and above 90mm, of which 0-90mm materials are sent to the combined screening module, and materials above 90mm are sent to the second magnetic separation module; Combined screening module: It includes a combined step screen with a double-layer screening structure. It receives the materials output by the preliminary screening module, further screens them, and classifies them by size. 12-60mm materials are sent to the first magnetic separation module, materials above 60mm are sent to the second magnetic separation module, and 0-12mm materials are sent to the material yard as finished powders. The first magnetic separation module: comprising the first magnetic separation equipment, used to separate scrap metal from the 12-60 mm material output from the combined screening module, and the remaining material is sent to the negative pressure air separation module; Negative pressure air separation module: It includes negative pressure air separation equipment, which is used to separate light materials and waste-derived fuels through negative pressure air separation, and the remaining materials are sent to the optical separation module; Optical sorting module: It includes optical sorting equipment, which uses near-infrared spectroscopy technology to classify the materials after negative pressure air separation, separate the wood blocks, and the remaining materials enter the aggregate finished material yard; The second magnetic separation module: includes the second magnetic separation equipment, which processes the materials larger than 60 mm output from the combined screening module and the materials larger than 90 mm output from the preliminary screening module, separates the scrap metal and sends it to the positive pressure air separation module; Positive pressure air separation module: includes positive pressure air separation equipment, which separates light materials and waste-derived fuels through positive pressure air separation, and the remaining materials are sent to the robot sorting module; Robot sorting module: includes robot sorting equipment, which uses deep learning-based algorithms to detect materials and control the robot to take out materials according to preset categories. The remaining materials enter the aggregate finished product yard.

[0005] Preferably, in a possible implementation manner of the first aspect, the first 3D drum screen and the second 3D drum screen have a structure for preventing strip-shaped objects from passing through, and the structure includes strip screens alternately arranged in a wavy shape and a spiral shape and an internal baffle design; The first 3D drum screen and the second 3D drum screen also have a composite drum motion mechanism, which includes three motion modes of the drum: tilting, vibrating and swinging.

[0006] Preferably, in a possible implementation manner of the first aspect, the strip-shaped object includes wood strips and waste steel bars.

[0007] Preferably, in a possible implementation manner of the first aspect, the double-layer screening structure of the combined screening module includes a 3D design layer and a tension and relaxation screen layer.

[0008] Preferably, in a possible implementation manner of the first aspect, the tension and relaxation screen layer has elastic deformation characteristics, can deform appropriately when materials pass through, and remove attachments through vibration and material flow.

[0009] Preferably, in a possible implementation manner of the first aspect, the sorting algorithm adopted by the robot sorting module comprises the following steps: Constructing a dataset using historical material image data; Use the dataset to train the deep learning model and deploy the trained model to the robot control system; Based on the real-time material image data transmitted by the camera, a deep learning model is used to output the category and bounding box of each material; Calculate the position of each material, and through the robot control system, control the robot to take out the materials according to the preset classification and the position of each material.

[0010] Preferably, in a possible implementation manner of the first aspect, the network structure of the deep learning model specifically includes: Input layer: input real-time material images; Feature extraction layer: The CSPDarknet53 network is used as the backbone network, multiple stages are constructed through group convolution and residual connection, and deep separable convolution and channel shuffle operations are introduced; Feature fusion layer: PANet network is used for multi-scale feature fusion; Detection head: A dual allocation strategy is used to perform target classification and localization; Output layer: Generate detection boxes and category probabilities.

[0011] The beneficial effects of the present invention are as follows: the intelligent decoration waste processing system provided by the present invention realizes the efficient classification and resource utilization of decoration waste through multi-stage screening, magnetic separation, wind separation and intelligent robot sorting, and significantly improves the processing efficiency and resource recovery rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0013] Figure 1 A structural schematic diagram of a decoration waste intelligent processing system is provided for this application.

[0014] Explanation of the accompanying drawings: 1-first plate feeder, 2-second plate feeder, 3-first 3D drum screen, 4-second 3D drum screen, 5-combined step screen, 6-first magnetic separation equipment, 7-second magnetic separation module, 8-negative pressure air separation equipment, 9-positive pressure air separation equipment, 10-optical separation equipment, 11-robot sorting equipment. DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0016] Embodiment 1: Figure 1 As shown, the present invention provides an intelligent processing system for decoration waste, comprising: Preliminary screening module: includes the first plate feeder, the second plate feeder, the first 3D drum screen and the second 3D drum screen, which are used to process the input renovation waste; bagged waste is input through the first plate feeder, and bulk waste is input through the second plate feeder. The input bagged waste is broken and then screened through the first 3D drum screen to obtain materials of 0-100mm and above 100mm; 0-100mm materials are directly sent to the combined screening module; materials above 100mm are mixed with bulk waste and sent to the second 3D drum screen for further screening to obtain materials of 0-90mm and above 90mm, of which 0-90mm materials are sent to the combined screening module, and materials above 90mm are sent to the second magnetic separation module.

[0017] Specifically, the plate feeder in this embodiment adopts GBZ1100×12500 heavy-duty plate feeder, which has the advantages of compact structure, small footprint, and is suitable for sites with limited space; stable and reliable operation ensures the continuity and efficiency of garbage transportation; and is equipped with an advanced control system to realize automated operation and improve work efficiency.

[0018] The 3D drum screen adopts 3D design to prevent strips such as wood strips and scrap steel bars from mixing into the 0-100mm screen. The inclination angle and speed of the drum screen are calculated to achieve the best screening effect. The inclination angle helps the material to slide naturally in the drum, while the appropriate speed can promote the tumbling and scattering of the material, improving the screening efficiency.

[0019] The 3D drum screen adopted in this embodiment uses wavy and spiral staggered strip screens to increase the contact opportunities between the material and the screen and improve the screening efficiency. By controlling the screen aperture and gap size, materials smaller than 100 mm can pass through and materials larger than this size are blocked. The composite motion of tilting, vibrating and swinging is combined to make the material tumble and collide in the drum, which helps to separate strip-shaped objects. The internal baffle design is adopted to lift the material to a certain height and then drop it, guiding the material to move along a specific path, increasing the chance of strip-shaped objects being intercepted by the screen.

[0020] Combined with the complex drum movement mechanism, including tilting, vibration and swinging, the materials are rolled and collided in the drum, improving the material separation effect.

[0021] Combined screening module: It includes a combined step screen with a double-layer screening structure. It receives the materials output by the preliminary screening module, further screens them, and classifies them by size. 12-60mm materials are sent to the first magnetic separation module, and materials above 60mm are sent to the second magnetic separation module. 0-12mm materials enter the material yard as finished powders.

[0022] Specifically, the combined screen adopts a double-layer screening design. The first layer adopts a 3D design to effectively prevent the iron wire and hooks in the material from hooking the screen. The second layer adopts a tension and relaxation screen design to ensure that there is no blockage even when the material is relatively wet.

[0023] The surface of the screen is designed with tiny protrusions or grooves. These structures can change the flow trajectory of the material on the screen and reduce the chance of wires, hooks, etc. directly hooking the screen. Wear-resistant, corrosion-resistant and elastic materials such as stainless steel or special alloys are selected to improve the durability and anti-hooking performance of the screen. The tension screen design has the function of elastic deformation and self-cleaning of the screen. It can deform appropriately when the material passes through, preventing the wet material from clogging the screen, and shaking off or flushing away the attachments through vibration and material flow.

[0024] In this embodiment, the screen uses laser micromachining technology on stainless steel to form a series of tiny hemispherical protrusions (diameter 0.5mm, height 0.2mm), which are evenly distributed on the surface of the screen. It is designed to guide the material to flow in a more complex path, reducing the risk of direct impact and hooking the screen, especially for iron wire and slender hooks. At the same time, subtle V-shaped grooves (depth of about 0.1mm) are designed in some areas prone to clogging. These grooves can guide small particles and liquids in the wet material to flow in a specific direction, reduce accumulation, and improve screening efficiency.

[0025] The tension and relaxation screen allows the screen to undergo a slight elastic deformation (deformation between 0.1% and 0.5%) when impacted by materials. This deformation can effectively alleviate the direct impact of the material on the screen, while increasing the screening area and improving the screening efficiency. A spring is installed at the bottom of the screen as an elastic support system to further enhance the adaptability and shock resistance of the screen. Combined with the vibration screening technology, a high-frequency vibration motor is installed under the screen. By adjusting the vibration frequency and amplitude, the screen generates high-frequency vibration when the material passes through. This vibration not only promotes the rapid passage of materials through the screen, but also effectively shakes off the fine particles and wet materials attached to the screen to achieve a self-cleaning effect. A spray device is set at the outlet of the screen to spray clean water regularly, combined with vibration, to further wash away the residue on the screen, keeping the screen clean and the screening efficiency.

[0026] The first magnetic separation module: includes the first magnetic separation equipment, which is used to separate metal from the 12-60mm material output by the combined screening module, and the remaining material is sent to the negative pressure air separation module.

[0027] Specifically, the first magnetic separation device performs magnetic separation on the 12-60 mm material transported by the combined screening module. In this embodiment, the first magnetic separation device adopts an electromagnetic drum magnetic separator, and the working principle of the electromagnetic drum magnetic separator is based on the magnetic force exerted on the magnetic material in the magnetic field. When the material passes through the magnetic field area of ​​the magnetic separator, the magnetic material therein will be adsorbed on the surface of the drum, and will be brought to the non-magnetic area as the drum rotates. At this time, the magnetic material loses its magnetic force and falls off, thereby achieving separation from the non-magnetic material.

[0028] Negative pressure air separation module: It includes negative pressure air separation equipment, which is used to separate light materials and waste-derived fuels through negative pressure air separation, and the remaining materials are sent to the optical separation module.

[0029] Specifically, decoration waste materials containing light substances (particle size range of 12-60mm) are fed into the feed port of the equipment. These materials may include bricks, concrete fragments, plastics, paper and other components.

[0030] The equipment is equipped with a negative pressure fan and an air flow distribution device. When the material enters the sorting area, the negative pressure fan generates a strong suction force to form a negative pressure air flow. The air flow passes through a specific distribution device to form an air flow field in the sorting area. Light materials (such as plastics and paper) are blown up and float upwards under the action of the air flow, while heavy materials (such as bricks and concrete fragments) remain in place due to gravity.

[0031] The light materials blown up are then captured by a collection device, such as a bag filter, while the heavy materials are discharged through the discharge port below the equipment and sent to the optical sorting module.

[0032] Optical sorting module: It includes optical sorting equipment, which uses near-infrared spectroscopy technology to classify the materials after negative pressure air separation, separate the wood blocks, and the remaining materials enter the aggregate finished product yard.

[0033] Specifically, the material output from the negative pressure air separation module is spectrally measured using a NIR spectrometer, which emits near-infrared light and receives light signals reflected or transmitted by the material to obtain spectral data of the material.

[0034] NIR spectrometer, or near-infrared spectrometer, is a critical precision analytical equipment in the optical sorting module. It uses near-infrared spectroscopy technology to achieve non-contact, fast and non-destructive material classification. By emitting near-infrared light and receiving light signals reflected or transmitted by the material, the unique spectral data of the material is obtained. These data are then compared and analyzed with the pre-built spectral database. During the optical sorting process, the NIR spectrometer can accurately identify small-sized wood blocks in the material after negative pressure air separation. The application advantage of the NIR spectrometer lies in its non-contact measurement method. In addition, its fast and efficient measurement speed and high-accuracy analysis capabilities also enable the optical sorting module to process a large amount of material in a short time, improving the overall processing efficiency.

[0035] The measured spectral data is compared and analyzed with the pre-established spectral database. The spectral characteristic information of materials of different materials is stored in the spectral database. Through the comparison and analysis, small-sized wood pieces in the material are identified. According to the results, the identified small-sized wood pieces are separated.

[0036] The second magnetic separation module: includes the second magnetic separation equipment, which processes the materials larger than 60 mm output by the combined screening module and the materials larger than 90 mm output by the preliminary screening module, and sends the metals to the positive pressure air separation module after separation.

[0037] Specifically, the second magnetic separation device magnetically separates materials larger than 60 mm conveyed by the combined screening module and materials larger than 90 mm conveyed by the second 3D drum screen. In this embodiment, the second magnetic separation device adopts a suspended magnetic separator. The working principle of the suspended magnetic separator is similar to that of the electromagnetic drum magnetic separator, and is also based on the magnetic force exerted on the magnetic material in the magnetic field. The difference is that the magnetic field of the suspended magnetic separator is generated by a magnet suspended in the air. When the material passes under the magnet, the magnetic material therein will be adsorbed on the magnet, and then the magnetic material will be removed from the magnet by a mechanical device.

[0038] Positive pressure air separation module: contains positive pressure air separation equipment, which separates light materials and waste-derived fuels through positive pressure air separation, and the remaining materials are sent to the robot sorting module.

[0039] Specifically, the positive pressure air separation module is mainly based on the effect of airflow. When the material is fed into the module, the fan inside the module generates positive pressure airflow, and the airflow acts on the material through a specific air duct or nozzle. Due to the difference in the behavior of light and heavy objects in the airflow, light objects (such as paper, plastic film, etc.) will be blown up by the airflow and suspended in the air, while heavy objects (such as metal, glass, etc.) will remain under the airflow or be discharged along a specific path due to inertia.

[0040] Inside the module, the fan generates a strong positive pressure airflow, which acts on the material through the air duct. At this time, light objects are blown up by the airflow and suspended in the air, while heavy objects remain below the airflow. The blown light objects are then captured by the bag dust collector. The heavy objects are discharged through the discharge port below the equipment and sent to the subsequent robot sorting module for further processing and classification.

[0041] Robot sorting module: includes robot sorting equipment, which uses deep learning-based algorithms to detect materials and control the robot to take out materials according to preset categories. The remaining materials enter the aggregate finished product yard.

[0042] Specifically, the process of building a deep learning-based detection algorithm includes: Use historical material image data to build a dataset. Deploy a camera above the robot sorting area to obtain historical material image data, use LabelImg to annotate, and after starting LabelImg, open the dataset folder containing the image data.

[0043] In LabelImg, create rectangular boxes by dragging the mouse and assign a label to each rectangular box, making sure the label name is consistent with the category name used during model training. After labeling is completed, save the labeling information. LabelImg saves the labeling information as a txt file with the same name as the image, which contains the category and coordinate information of the target box.

[0044] Train deep learning models and deploy the trained models to the robot control system.

[0045] First, install the necessary libraries and frameworks. This example uses the PyTorch framework. Set the necessary parameters in the training script, such as learning rate, batch size, training rounds, optimizer, etc. Specify the paths of the training set and validation set. Run the training script to start training the model. During the training process, the model will learn on the training set and be evaluated on the validation set at the end of each epoch. During the training process, the performance of the model is evaluated by observing the performance indicators on the validation set. If the performance of the model on the validation set no longer improves, or there are signs of overfitting, consider stopping training early or using regularization techniques.

[0046] Use the test set to test the trained model to evaluate its generalization ability on unknown data. Write a test script to load the trained model and make predictions for each image in the test set. Calculate and record performance metrics on the test set, such as mAP, precision, recall, etc.

[0047] In this embodiment, the network structure of the deep learning model includes an input layer, a feature extraction layer, a feature fusion layer, a detection head and an output layer.

[0048] The input layer is responsible for inputting real-time material images from various cameras. All images are adjusted to 256×256. Next, the image data is normalized, that is, the pixel values ​​are mapped from the original integer range to a smaller floating point range, which helps to accelerate the training process of the model and improve the convergence speed of the model. Finally, in order to increase the generalization ability of the model, a variety of data enhancement techniques are applied to the images, including but not limited to random rotation, cropping, scaling, and flipping, to simulate various changes that images may encounter in actual applications, thereby helping the model learn more robust features.

[0049] The feature extraction layer uses the CSPDarknet53 backbone network as the core of the model, which is responsible for extracting deep feature representations from the input image. CSPDarknet53 combines the design concepts of CSPNet and Darknet, and achieves high efficiency and accuracy in feature extraction through cross-stage partial connections and Darknet's unique convolution design. The network constructs multiple stages through grouped convolutions and residual connections to gradually extract deep features from the image. Among them, the feature map is divided into two parts for processing, one part is directly transmitted, and the other part is processed by complex convolution blocks and then fused, which effectively reduces the amount of calculation and maintains a high feature learning ability.

[0050] In order to further improve the efficiency and quality of feature extraction, depthwise separable convolution and channel shuffling operations are introduced on the basis of CSPDarknet53. Depthwise separable convolution decomposes the standard convolution into two steps: depthwise convolution and pointwise convolution. The former performs convolution operations on each input channel independently, while the latter is responsible for mixing channel information. The channel shuffling operation complements the grouped convolution, which solves the problem of information not flowing between feature map channels after grouped convolution. By rearranging the order of the grouped channels, channel shuffling enables information to flow between different groups, thereby enhancing the feature extraction capability of the model.

[0051] The feature fusion layer uses feature pyramids and PANet networks to enhance the model's ability to detect objects of different sizes. The feature pyramid fuses feature maps at different levels of the backbone network through top-down paths and lateral connections to form a feature representation that contains rich multi-scale information. PANet further adds bottom-up path aggregation to enrich the spatial and semantic information of the feature map. This multi-scale feature fusion mechanism enables the model to capture both large and small target information in the image, improving the accuracy and efficiency of detection.

[0052] The detection head is the last part of the model, which is responsible for performing accurate classification and localization of the target based on the features of the feature extraction layer and the feature fusion layer. The detection head adopts a dual allocation strategy to optimize the training process and improve the inference speed. In the training phase, a dual label allocation mechanism of one-to-many and one-to-one is used to train the model. The one-to-many allocation strategy helps the model learn more comprehensive target features because it allows one target to correspond to the annotation information of multiple candidate boxes. The one-to-one allocation strategy simplifies the inference process because it requires that each target can only be matched by one candidate box. In the inference phase, in order to improve the speed, only a one-to-one allocation strategy is used and the non-maximum suppression algorithm is avoided to remove overlapping candidate boxes. It can not only reduce the amount of calculation but also improve the speed and accuracy of detection. Finally, the detection head will output the detection box and the corresponding category probability of each target.

[0053] Output layer: The output layer generates the final detection box and category probability based on the prediction results of the detection head. The detection box of each target is generated based on the bounding box parameters predicted by the detection head; at the same time, the probability of each detection box corresponding to each category is output to determine the category of the target in the detection box.

[0054] The performance indicators of this embodiment use mAP, precision, and recall. mAP is the average of the average detection accuracy of multiple categories, which is used to measure the average performance of the model in different categories. It is an important evaluation indicator in the target detection task, which can comprehensively reflect the detection effect of the model under different intersection-over-union thresholds; the precision measures the proportion of instances predicted by the model as positive samples that are actually positive samples, reflecting the accuracy of the model's prediction; it measures the model's coverage of positive samples, reflecting the completeness of the model in finding positive samples in the detection task.

[0055] Deploy the trained model to the robot's control device. Install the PyTorch environment and configure all necessary dependency libraries. Install a high-definition camera on the robot's head and adjust its focus, exposure and other parameters to obtain the best image quality. The camera transmits the real-time video stream to the control device via the RTSP protocol.

[0056] Based on the real-time material data transmitted by the camera, the deep learning model outputs the category and bounding box of each material.

[0057] Specifically, after the control device receives the video stream, it captures images frame by frame and performs preprocessing operations on the images, including resizing and normalization, to meet the input requirements of the model. The preprocessed images are fed into the deep learning model loaded by PyTorch for real-time inference.

[0058] Calculate the position of each material, and through the robot control system, control the robot to take out the materials according to the preset classification and the position of each material.

[0059] Specifically, the model infers the input image and outputs the category and bounding box of each detected material. Based on this output information, the robot system calculates the location of each material and formulates a corresponding sorting strategy. This embodiment presets the classification into three types of materials: wood, non-ferrous metal, and aerated blocks.

[0060] Through the robot control system, instructions are sent to the robot to move to the specified position and perform grabbing or pushing operations. The robot's motion trajectory is planned to ensure that the material can be accurately sorted into the corresponding collection container and enter the finished product yard.

[0061] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. An intelligent processing system for decoration waste, characterized in that: include: Preliminary screening module: including the first plate feeder, the second plate feeder, the first 3D drum screen and the second 3D drum screen, used to process the input decoration waste; Bag-packed garbage is input through the first plate feeder, and bulk garbage is input through the second plate feeder. The bag-packed garbage is broken and screened through the first 3D drum screen to obtain materials of 0-100mm and above 100mm; 0-100mm materials are directly sent to the combined screening module; materials above 100mm are mixed with bulk garbage and sent to the second 3D drum screen for further screening to obtain materials of 0-90mm and above 90mm, of which 0-90mm materials are sent to the combined screening module, and materials above 90mm are sent to the second magnetic separation module; Combined screening module: It includes a combined step screen with a double-layer screening structure. It receives the materials output by the preliminary screening module, further screens them, and classifies them by size. 12-60mm materials are sent to the first magnetic separation module, materials above 60mm are sent to the second magnetic separation module, and 0-12mm materials are sent to the material yard as finished powders. The first magnetic separation module: comprising the first magnetic separation equipment, used to separate scrap metal from the 12-60 mm material output from the combined screening module, and the remaining material is sent to the negative pressure air separation module; Negative pressure air separation module: It includes negative pressure air separation equipment, which is used to separate light materials and waste-derived fuels through negative pressure air separation, and the remaining materials are sent to the optical separation module; Optical sorting module: It includes optical sorting equipment, which uses near-infrared spectroscopy technology to classify the materials after negative pressure air separation, separate the wood blocks, and the remaining materials enter the aggregate finished material yard; The second magnetic separation module: includes the second magnetic separation equipment, which processes the materials larger than 60 mm output from the combined screening module and the materials larger than 90 mm output from the preliminary screening module, separates the scrap metals and sends them to the positive pressure air separation module; Positive pressure air separation module: includes positive pressure air separation equipment, which separates light materials and waste-derived fuels through positive pressure air separation, and the remaining materials are sent to the robot sorting module; Robot sorting module: includes robot sorting equipment, which uses deep learning-based algorithms to detect materials and control the robot to take out materials according to preset categories. The remaining materials enter the aggregate finished product yard.

2. The system according to claim 1, characterized in that The first 3D drum screen and the second 3D drum screen have a structure for preventing strip-shaped objects from passing through, and the structure includes strip screens alternately arranged in a wavy and spiral shape and an internal baffle design; The first 3D drum screen and the second 3D drum screen also have a composite drum motion mechanism, which includes three motion modes of the drum: tilting, vibrating and swinging.

3. The system according to claim 2, characterized in that The strip-shaped objects include wood strips and waste steel bars.

4. The system according to claim 1, characterized in that The double-layer screening structure of the combined screening module includes a 3D design layer and a tension and relaxation screen layer.

5. The system according to claim 4, characterized in that The tension and relaxation screen layer has elastic deformation characteristics, can be deformed appropriately when materials pass through, and can remove attachments through vibration and material flow.

6. The system according to claim 1, wherein: The sorting algorithm adopted by the robot sorting module includes the following steps: Constructing a dataset using historical material image data; Use the dataset to train the deep learning model and deploy the trained model to the robot control system; Based on the real-time material image data transmitted by the camera, a deep learning model is used to output the category and bounding box of each material; Calculate the position of each material, and through the robot control system, control the robot to take out the materials according to the preset classification and the position of each material.

7. The system according to claim 6, characterized in that The network structure of the deep learning model specifically includes: Input layer: input real-time material images; Feature extraction layer: The CSPDarknet53 network is used as the backbone network, multiple stages are constructed through group convolution and residual connection, and deep separable convolution and channel shuffle operations are introduced; Feature fusion layer: PANet network is used for multi-scale feature fusion; Detection head: A dual allocation strategy is used to perform target classification and localization; Output layer: Generate detection boxes and category probabilities.

Images