Recycled resource automatic sorting method and system combined with intelligent identification

Abstract

The invention relates to a method and a system for automatically sorting recycled resources in combination with intelligent identification. A multi-layer sorting assembly line is constructed and comprises a coarse sorting assembly line and a fine sorting assembly line; the rough sorting assembly line obtains and preprocesses garbage data; according to the preprocessed data, based on a coarse classification model, performing coarse classification on garbage, and detecting the position of a single article in combination with a target detection algorithm; according to the rough classification result and the article position, sorting the articles to a corresponding fine sorting assembly line through a mechanical arm; each fine sorting assembly line is used for carrying out secondary classification on the articles; each fine sorting assembly line is finally connected to an independent conveying belt, the materials are directly conveyed to the corresponding storage areas, weight recording is carried out before the materials enter the storage areas, RFID tags are adopted, and material types, source areas, weight and processing arrangement information are added for each pile of garbage. The garbage management cost is effectively reduced, and the resource recovery efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent sorting, and particularly to an automatic sorting method and system for recycling resources combined with intelligent recognition. Background Art

[0002] Traditional manual sorting methods are inefficient, inaccurate in classification, and have a high risk of pollution, making it difficult to meet the requirements of automation, precision, and high efficiency in modern waste treatment.

[0003] Although existing automated sorting systems have improved the efficiency of waste classification to a certain extent, they still face multiple technical challenges. For example, many systems classify based on a single sensor (such as an RGB camera) or simple rules, resulting in insufficient recognition accuracy for complex waste components and difficulty in precisely sorting various materials of waste. In addition, due to the complexity of waste types, different shapes and sizes, and the presence of pollutants on the surface of some waste, the stability and adaptability of existing detection algorithms and equipment are insufficient to cope with complex and changing sorting scenarios. At the same time, there is also a lack of an efficient optimization scheme for obtaining the position information of waste items and the linkage control of the sorting pipeline, and the collaborative efficiency of robotic arm sorting equipment needs to be improved urgently. Summary of the Invention

[0004] In order to solve the above problems, the purpose of the present invention is to provide an automatic sorting method and system for recycling resources combined with intelligent recognition, which can effectively reduce the waste management cost and improve the resource recycling efficiency.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions: An automatic sorting method for recycling resources combined with intelligent recognition, comprising the following steps; Construct a multi-layer sorting pipeline, including a rough sorting pipeline and a fine sorting pipeline; The rough sorting pipeline uses an RGB recognition camera, an infrared sensor, and a spectral analyzer to obtain waste data and preprocess it; based on the preprocessed data, the waste is roughly classified into plastic, glass, metal, and paper categories using a rough classification model, and the position of each item is detected using a target detection algorithm; According to the rough classification result and the position of the item, the item is sorted to the corresponding fine sorting pipeline by a robotic arm; The fine sorting pipeline includes a plastic sorting line, a metal sorting line, a glass sorting line, and a paper sorting line; each fine sorting pipeline performs secondary classification on the items; Each fine sorting pipeline is finally connected to an independent conveyor belt, and the materials are directly transported to the corresponding storage area. Before the materials enter the storage area, the weight is recorded, and an RFID tag is used to add information such as the material category, source area, weight, and processing arrangement for each pile of waste.

[0006] Furthermore, garbage data is acquired and preprocessed through an RGB recognition camera, an infrared sensor, and a spectral analyzer, as follows: The RGB recognition camera is used to collect RGB images, obtaining the color and texture features of the garbage surface, and outputting the RGB image feature F RGB ; The infrared sensor is used to obtain the temperature and reflectivity information of the material, constructing an infrared feature vector F IR ; The spectral analyzer is used to obtain the absorption or reflection spectra of the garbage at different wavelengths, constructing a spectral feature vector F Spectral ; The RGB image feature F RGB , the infrared feature vector F IR , and the spectral feature vector F Spectral are fused to obtain the final multi-modal feature based on a deep neural network: The input layer of the deep neural network is constructed, and the features of the three modalities are respectively dimension-reduced through the embedding layer. The formula is as follows: ; ; ; where, W RGB , W IR , W Spectral are the weight coefficients of the RGB image, infrared, and spectrum; b RGB , b IR , b Spectral are the bias matrices of the RGB image, infrared, and spectrum respectively; Z RGB , Z IR , Z Spectral are the features of the RGB image, infrared, and spectrum modalities after dimension reduction; The fusion layer constructs the final multi-modal feature F final : F final =σ(W fuse [Z RGB ,Z IR ,Z Spectral +b fuse ); where, W fuse and b fuse are the weight and bias matrices of the fusion layer; σ() is the activation function.

[0007] Furthermore, the RGB recognition camera is used to collect RGB images, obtaining the color and texture features of the garbage surface, and outputting the RGB image feature FRGB , specifically as follows: Convert the RGB image to the HSV color space and extract the color information of the garbage. The formula is as follows: ; ; ; where R, G, and B are the pixel values of the red, green, and blue channels respectively, H is the hue, S is the saturation, and V is the brightness; statistically analyze the distributions of H, S, and V as histogram features; Extract texture information through the gray-level co-occurrence matrix : ; where P(i,j∣Δx,Δy) represents the probability of the occurrence of gray values i and j between pixel points, and Δx and Δy are the distances in the horizontal and vertical directions respectively; Calculate the contrast Contrast, energy Energy, and homogeneity Homogeneity: ; ; ; Combine the color histogram and texture features to form an RGB image feature vector: F RGB = {H, S, V, Contrast, Energy, Homogeneity}.

[0008] Furthermore, based on the coarse classification model, classify the garbage into plastic, glass, metal, and paper categories roughly, and combine the object detection algorithm to detect the position of a single item. Specifically as follows: The coarse classification model adopts a classification model based on deep learning. According to the final multi-modal feature F final , obtain the probability output P coarse (c∣F final ), where c represents the category c; The object detection algorithm uses the RGB image as the input, combines the multi-modal feature F final to optimize the category information and simultaneously predict the category and the position of the bounding box ; Combine the probability output P coarse of the coarse classification and the detection classification probability P det， to calculate the final classification probability: ; Among them, α is the fusion weight.

[0009] Furthermore, the classification model based on deep learning includes an input layer, a feature extraction network, an Attention module, and an output layer, which are specifically as follows: The input layer inputs the final multi-modal feature F final , and the feature extraction network processes the input multi-modal feature F final through a weighted linear transformation: ; where W1 is the weight matrix of the weighted linear transformation; b1 is the bias vector of the weighted linear transformation; Z1 is the output of the weighted linear transformation; Enhance the non-linear feature representation through the ReLU activation function: ; where A1 is the output enhanced by the ReLU activation function; Extract features further through n - 1 hidden layers: ; ; ; where W2, W n are the weight matrices of the first hidden layer and the n - 1 hidden layer respectively; b2, b n are the bias vectors of the first hidden layer and the n - 1 hidden layer respectively; Z 2， Z n are the outputs of the first hidden layer and the n - 1 hidden layer respectively; A2, A n are the outputs enhanced by the ReLU activation function of the first hidden layer and the n - 1 hidden layer respectively; The Attention module captures important features by learning the weight distribution W of the multi-modal features attn : ; where Softmax is the activation function; Calculate the feature weighted vector to obtain the enhanced feature : ; Pass the enhanced feature through the hidden layer: ; where W attn-hidden , b attn-hidden are the attention weights and biases; is the representation of the hidden layer; The final attention feature representation is combined with the hidden layer representation A n : ; where λ is the combination ratio; The output layer maps the final feature F enhanced to the garbage classification space: ; where W class , b class are the mapping weights and biases; Z class is the feature after mapping; The probability distribution of each garbage category is output using the Softmax function: .

[0010] Furthermore, the object detection algorithm uses an improved YOLO model, including multi-scale image feature extraction, a feature fusion module, a bounding box regression module, and a classification head, as follows: Using the RGB image I RGB as the input, multi-scale features are extracted through the backbone network CSPDarknet: ; where is the convolutional feature; Backbone is the backbone network; At the same time, the multi-modal feature F final is combined to obtain the further fused feature :

[0011] where Concat represents the concatenation operation in the channel dimension; The bounding box regression module outputs the predicted normalized bounding box position B=(x,y,w,h), where (x,y) are the coordinates of the bounding box, and w,h are the width and height of the bounding box respectively. The classification head part uses a position encoding mechanism to optimize the positioning accuracy of the bounding box and outputs the center coordinates as well as the width and height : ; ; where t x , t y , t w , t h are the center coordinates and as well as the width and height The offset; σ is the activation function; c x 、c y is the offset of the center coordinates of the predicted grid cell; p w 、p h are the reference width and height of the corresponding grid cell; The final coordinates of the bounding box are: ; According to the fused features , the class probability distribution of each target is predicted through the classification head : ; where, W c b c are the weights and biases of the classification head.

[0012] Furthermore, according to the rough classification result and the item position, the item is sorted to the corresponding sub-sorting pipeline by the robotic arm, specifically as follows: According to the final classification result c output by the classification model, the actual three-dimensional coordinates (Xreal, Yreal, Zreal) of the item position are obtained by the object detection model; Grasping stage: Calculate the grasping point Ppick=(Xreal, Yreal, Zreal) and the safety point Psafe-pick; use inverse kinematics to solve the joint angles and drive the robotic arm to complete the grasping; Sorting stage: Find the target sorting point Ptarget=(Xtarget, Ytarget, Ztarget) according to the classification result c, (Xtarget, Ytarget, Ztarget) are the three-dimensional coordinates of the target sorting point; path planning moves the item from the grasping point to the target sorting point; Releasing stage: Place the item on the corresponding pipeline at the position where the height Psafe-targe is reduced at the target point.

[0013] Preferably, each sub-sorting pipeline performs secondary classification on the item, specifically as follows: for the plastic sorting line, an NIR spectrometer is used to subdivide different categories; for the metal sorting line, an eddy current separator and a magnetic separation device are used to separate non-ferromagnetic and ferromagnetic metals; for the glass sorting line, RGB cameras are used to assist in color sorting; for the paper sorting line, clean paper and contaminated paper are classified.

[0014] An automatic sorting system for recycling resources combined with intelligent recognition, including a processor, a memory, and a computer program stored on the memory. When the processor executes the computer program, it specifically executes the steps in the above-mentioned automatic sorting method for recycling resources combined with intelligent recognition.

[0015] A computer storage medium stores multiple instructions, and the instructions are suitable for being loaded and executed by a processor to execute the steps in the above-mentioned automatic sorting method for recycling resources combined with intelligent recognition.

[0016] The present invention has the following beneficial effects: 1. The present invention gradually extracts the features of the surface color, texture, temperature, material reflectivity, and spectral characteristics of garbage through RGB images, infrared sensors, and spectrometers, and completes multi-modal feature fusion through hybrid strategies such as normalization, weighting, and dimensionality reduction. Finally, a compact feature vector that can be used for subsequent classification and analysis is formed; 2. Based on the multi-modal fusion features, the present invention constructs a deep learning coarse classification model to quickly classify garbage into four categories: plastic, glass, metal, and paper. At the same time, it combines the object detection algorithm to detect the position of a single item, realizing the synchronous processing of garbage classification and positioning, reducing the operation complexity, and providing accurate information for subsequent robotic arm operations; 3. The present invention divides the sorting task into two parts: rough sorting (large category classification) and fine sorting (sub-category classification), effectively improving the sorting efficiency and accuracy. Through the rough sorting pipeline, the large category classification of garbage (plastic category, glass category, metal category, paper category) is quickly completed, reducing the complexity of fine processing; the fine sorting pipeline further performs refined processing to improve the resource recycling rate. Description of the Drawings

[0017] Figure 1 It is a flowchart of the method of the present invention. Detailed Embodiment

[0018] The following further describes the present invention in detail with reference to the drawings and specific embodiments: Refer to Figure 1 , in this embodiment, an automatic sorting method for recycling resources combined with intelligent recognition is provided, including the following steps; Construct a multi-layer sorting pipeline, including a rough sorting pipeline and a fine sorting pipeline; The rough sorting pipeline obtains garbage data through an RGB recognition camera, an infrared sensor, and a spectrometer, and preprocesses it; based on the preprocessed data, the garbage is roughly classified into plastic category, glass category, metal category, and paper category based on the coarse classification model, and the position of a single item is detected by combining the object detection algorithm; According to the rough classification result and the location of the item, the item is sorted to the corresponding sub-sorting pipeline by the robotic arm; The sub-sorting pipeline includes a plastic sorting line, a metal sorting line, a glass sorting line and a paper sorting line; each sub-sorting pipeline conducts secondary classification on the items; Each sub-sorting pipeline is finally connected to an independent conveyor belt, and the materials are directly transported to the corresponding storage area. Before the materials enter the storage area, the weight is recorded, and RFID tags are used to add information such as the material category, source area, weight and treatment arrangement to each pile of garbage.

[0019] In this embodiment, garbage data is obtained and preprocessed through an RGB recognition camera, an infrared sensor and a spectral analyzer, specifically as follows: The RGB recognition camera is used to collect RGB images, obtain the surface color and texture features of the garbage, and output the RGB image feature F RGB ; The infrared sensor is used to obtain the temperature and reflectivity information of the material, and construct the infrared feature vector F IR ; The spectral analyzer is used to obtain the absorption or reflection spectrum of the garbage at different wavelengths, and construct the feature vector F of the spectrum Spectral ; The RGB image feature F RGB , the infrared feature vector F IR and the feature vector F of the spectrum Spectral are fused, and the final multi-modal feature is obtained based on the deep neural network: The input layer of the deep neural network is constructed, and the features of the three modalities are respectively dimension-reduced through the embedding layer. The formula is as follows: ; ; ; Among them, W RGB , W IR , W Spectral are the weight coefficients of the RGB image, infrared and spectrum; b RGB , b IR , b Spectral are the bias matrices of the RGB image, infrared and spectrum respectively; Z RGB , Z IR , Z Spectral are the features of the RGB image, infrared and spectrum modalities after dimension reduction respectively; The fusion layer constructs the final multi-modal feature F through feature weighted combination final : F final =σ(W fuse[Z RGB ,Z IR ,Z Spectral ) + b fuse ); Among them, W fuse and b fuse are the weight and bias matrices of the fusion layer; σ() is the activation function.

[0020] In this embodiment, an RGB image is collected by an RGB recognition camera to obtain the color and texture features of the garbage surface, and the RGB image feature F RGB is output as follows: The RGB image is converted to the HSV color space to extract the color information of the garbage. The formula is as follows: ; ; ; Among them, R, G, and B are the pixel values of the red, green, and blue channels respectively, H is the hue, S is the saturation, and V is the brightness; the distributions of H, S, and V are statistically counted as histogram features; The texture information is extracted through the gray-level co-occurrence matrix (GLCM) : ; Among them, P(i,j∣Δx,Δy) represents the probability of the occurrence of gray values i and j between pixel points, and Δx and Δy are the distances in the horizontal and vertical directions respectively; Calculate the contrast Contrast, energy Energy, and homogeneity Homogeneity: ; ; ; Combine the color histogram and texture features to form an RGB image feature vector: F RGB = {H, S, V, Contrast, Energy, Homogeneity}.

[0021] In this embodiment, the garbage is roughly classified into plastic, glass, metal, and paper categories based on a rough classification model, and the position of a single item is detected in combination with an object detection algorithm as follows: The rough classification model uses a classification model based on deep learning. According to the final multi-modal feature F final , the probability output P coarse (c∣F final), where c represents class c; The target detection algorithm uses an RGB image as input and combines multi-modal feature F final to optimize the class information and simultaneously predict the class of the item and the bounding box position ; Combining the probability output P of the coarse classification coarse and the detection classification probability P det， calculate the final classification probability: ; where α is the fusion weight.

[0022] In this embodiment, the classification model based on deep learning includes an input layer, a feature extraction network, an Attention module, and an output layer, specifically as follows: The input layer inputs the final multi-modal feature F final , and the feature extraction network transforms the input multi-modal feature F final through a layer of weighted linear transformation: ; where W1 is the weight matrix of the weighted linear transformation; b1 is the bias vector of the weighted linear transformation; Z1 is the output of the weighted linear transformation; Enhance the non-linear feature expression through the ReLU activation function: ; where A1 is the output enhanced by the ReLU activation function; Extract features further through n-1 hidden layers: ; ; ; where W2, W n are the weight matrices of the first hidden layer and the n-1 hidden layer respectively; b2, b n are the bias vectors of the first hidden layer and the n-1 hidden layer respectively; Z 2， Z n are the outputs of the first hidden layer and the n-1 hidden layer respectively; A2, A n are the outputs enhanced by the ReLU activation function of the first hidden layer and the n-1 hidden layer respectively; The Attention module captures important features by learning the weight distribution W of the multi-modal features attn : ; Among them, Softmax is the activation function; Perform feature weighted vector calculation to obtain enhanced features : ; The enhanced features Pass through the hidden layer: ; Among them, W attn-hidden , b attn-hidden Are the attention weights and biases; Is the hidden layer representation; The final attention feature representation is combined with the hidden layer representation A n : ; Among them, λ is the combination ratio; The output layer maps the final feature F enhanced To the garbage classification space: ; Among them, W class , b class Are the mapping weights and biases; Z class Is the mapped feature; Use the Softmax function to output the probability distribution of each garbage category: .

[0023] In this embodiment, the object detection algorithm adopts an improved YOLO model, including multi-scale image feature extraction, a feature fusion module, a bounding box regression module, and a classification head, specifically as follows: Use the RGB image I RGB As the input, pass through the backbone network CSPDarknet to extract multi-scale features: ; Among them, Is the convolutional feature; Backbone is the backbone network; At the same time, combine the multi-modal feature F final To obtain the further fused feature :

[0024] Among them, Concat represents the concatenation operation on the channel dimension; The bounding box regression module outputs the predicted normalized bounding box position =(x, y, w, h), where (x, y) are the coordinates of the bounding box, and w and h are the width and height of the bounding box respectively. The classification head part uses a position encoding mechanism to optimize the positioning accuracy of the bounding box and outputs the center coordinates and the width and height : ; ; where t x 、t y 、t w 、t h are the offsets of the center coordinates and the width and height regressed by the bounding box regression network; σ is the activation function; c x 、c y are the offsets of the center coordinates of the predicted grid cell; p w 、p h are the reference width and height of the corresponding grid cell; The final coordinates of the bounding box are: ; According to the fused features , the classification head predicts the class probability distribution of each target : ; where W c 、b c are the weights and biases of the classification head.

[0025] In this embodiment, according to the rough classification result and the item position, the item is sorted to the corresponding sub-sorting pipeline by the robotic arm, specifically as follows: According to the final classification result c output by the classification model, the target detection model obtains the actual three-dimensional coordinates (Xreal, Yreal, Zreal) of the item position; Grasping stage: Calculate the grasping point Ppick = (Xreal, Yreal, Zreal) and the safety point Psafe - pick; Use inverse kinematics to solve the joint angles and drive the robotic arm to complete the grasping; Sorting stage: Find the target sorting point Ptarget = (Xtarget, Ytarget, Ztarget) according to the classification result c, where (Xtarget, Ytarget, Ztarget) are the three-dimensional coordinates of the target sorting point; Path planning moves the item from the grasping point to the target sorting point; Releasing stage: At the height Psafe - targe reduction position of the target point, place the item onto the corresponding pipeline.

[0026] Each of the sub - sorting pipelines classifies the items secondarily as follows: For the plastic sorting line, different categories are subdivided using an NIR spectrometer; for the metal sorting line, an eddy current separator and a magnetic separation device are used to separate non - ferromagnetic and ferromagnetic metals; for the glass sorting line, color sorting is assisted by an RGB camera; for the paper sorting line, clean paper and contaminated paper are classified.

[0027] An automatic sorting system for recycling resources combined with intelligent recognition includes a processor, a memory, and a computer program stored on the memory. When the processor executes the computer program, it specifically executes the steps in the above - mentioned automatic sorting method for recycling resources combined with intelligent recognition.

[0028] A computer storage medium stores multiple instructions, and these instructions are suitable for being loaded and executed by a processor to perform the steps in the above - mentioned automatic sorting method for recycling resources combined with intelligent recognition. Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer - usable storage media (including but not limited to disk storage, CD - ROM, optical storage, etc.) containing computer - usable program code.

[0029] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general - purpose computer, a special - purpose computer, an embedded processor, or other programmable data - processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data - processing devices generate a device for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0030] These computer program instructions can also be stored in a computer - readable memory that can direct a computer or other programmable data - processing device to work in a specific manner, so that the instructions stored in the computer - readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions in the flow Figure 1One or more processes and / or boxes Figure 1 The functions specified in one box or more boxes.

[0031] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 One or more processes and / or boxes Figure 1 One box or more boxes.

[0032] As described above, it is only the preferred embodiment of the present invention, and it is not a limitation to the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes. However, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the technical solution content of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. An automatic sorting method for recycled resources combined with intelligent recognition, characterized in that, Including the following steps; Construct a multi-layer sorting pipeline, including a rough sorting pipeline and a fine sorting pipeline; The rough sorting pipeline uses an RGB recognition camera, an infrared sensor, and a spectral analyzer to obtain garbage data and preprocess it; based on the preprocessed data, the garbage is roughly classified into plastic, glass, metal, and paper categories using a rough classification model, and the position of each individual item is detected using an object detection algorithm; According to the rough classification results and the item positions, the items are sorted to the corresponding fine sorting pipeline by a robotic arm; The fine sorting pipeline includes a plastic sorting line, a metal sorting line, a glass sorting line, and a paper sorting line; each fine sorting pipeline performs secondary classification on the items; Each fine sorting pipeline is finally connected to an independent conveyor belt, and the materials are directly transported to the corresponding storage area. Before the materials enter the storage area, the weight is recorded, and an RFID tag is used to add information such as the material category, source area, weight, and processing arrangement for each pile of garbage.

2. The automatic sorting method for recycled resources combined with intelligent recognition according to claim 1, wherein The process of using the RGB recognition camera, infrared sensor, and spectral analyzer to obtain garbage data and preprocess it is as follows: Collect an RGB image through an RGB recognition camera, obtain the color and texture features of the garbage surface, and output the RGB image feature F RGB ; Obtain the temperature and reflectivity information of the material through an infrared sensor, and construct an infrared feature vector F IR ; Obtain the absorption or reflection spectra of the garbage at different wavelengths through a spectral analyzer, and construct the feature vector F of the spectra Spectral ; For the RGB image feature F RGB , the infrared feature vector F IR and the spectral feature vector F Spectral are fused to obtain the final multimodal feature based on a deep neural network: Construct the input layer of a deep neural network, and reduce the dimensions of the features of the three modalities through an embedding layer respectively. The formula is as follows: ； ； ； Among them, W RGB , W IR , W Spectral are the weight coefficients of RGB images, infrared, and spectra; b RGB , b IR , b Spectral are the bias matrices of RGB images, infrared, and spectra respectively; Z RGB , Z IR , Z Spectral are the features of the RGB image, infrared, and spectral modalities after dimensionality reduction respectively; The fusion layer constructs the final multi-modal feature F through feature weighted combination final : F final = σ(W fuse [Z RGB , Z IR , Z Spectral + b fuse ); Among them, W fuse and b fuse are the weight and bias matrices of the fusion layer; σ() is the activation function.

3. An automatic sorting method for recycled resources combined with intelligent recognition according to claim 2, characterized in that, The RGB recognition camera is used to collect RGB images, obtain the color and texture features of the garbage surface, and output the RGB image feature F RGB , which is specifically as follows: Convert the RGB image to the HSV color space and extract the color information of the garbage. The formula is as follows: ； ； ； Where R, G, and B are the pixel values of the red, green, and blue channels respectively, H is the hue, S is the saturation, and V is the brightness; the distributions of H, S, and V are statistically calculated as histogram features; Extract texture information through gray-level co-occurrence matrix : ； Where P(i,j∣Δx,Δy) represents the probability of the gray values i and j appearing between pixel points, and Δx and Δy are the distances in the horizontal and vertical directions respectively; Calculate the contrast, energy, and homogeneity: ； ； ； Combine the color histogram and texture features to form an RGB image feature vector: F RGB ={H, S, V, Contrast, Energy, Homogeneity}。 4. The automatic sorting method for recycled resources combined with intelligent recognition according to claim 2, characterized in that The process of roughly classifying the garbage into plastic, glass, metal, and paper categories using the rough classification model and detecting the position of each individual item using an object detection algorithm is as follows: The coarse classification model uses a classification model based on deep learning, and according to the final multi-modal feature F final , obtains the probability output P coarse (c|F final ), where c represents the category c; The target detection algorithm uses RGB images as input and combines multi-modal features F final to optimize the class information and simultaneously predict the class of the item and the bounding box position ; Combine the probability output P of the coarse classification coarse and the detection classification probability P det， Calculate the final classification probability: ； Where α is the fusion weight.

5. An automatic sorting method for recycled resources combined with intelligent recognition according to claim 4, characterized in that, The classification model based on deep learning includes an input layer, a feature extraction network, an Attention module, and an output layer, as follows: The input layer inputs the final multi-modal feature F final , and the feature extraction network processes the input multi-modal feature F final through a layer of weighted linear transformation: ； Where W1 is the weight matrix of the weighted linear transformation; b1 is the bias vector of the weighted linear transformation; Z1 is the output of the weighted linear transformation; Enhance the non-linear feature expression through the ReLU activation function: ； Where A1 is the output enhanced by the ReLU activation function; Extract features further through n - 1 hidden layers: ； ; ; Among them, W2, W n are the weight matrices of the first hidden layer and the (n - 1)-th hidden layer respectively; b2, b n are the bias vectors of the first hidden layer and the (n - 1)-th hidden layer respectively; Z 2， Z n are the outputs of the first hidden layer and the (n - 1)-th hidden layer respectively; A2, A n are the outputs enhanced by the ReLU activation function of the first hidden layer and the (n - 1)-th hidden layer respectively; The Attention module captures important features by learning the weight distribution W of multimodal features attn : ； Where Softmax is the activation function; Perform feature weighted vector calculation to obtain enhanced features : ； The enhanced features through the hidden layer: ； Among them, W attn-hidden , b attn-hidden are the attention weights and biases; is the hidden layer representation; The final attention feature representation is combined with the hidden layer representation A n to combine: ； Where λ is the combination ratio; The output layer maps the final feature F enhanced to the waste classification space: ； Among them, W class , b class are the mapping weights and biases; Z class is the mapped feature; Use the Softmax function to output the probability distribution of each garbage category: 。 6. The automatic sorting method for recycled resources combined with intelligent recognition according to claim 4, characterized in that, The object detection algorithm uses an improved YOLO model, including multi-scale image feature extraction, a feature fusion module, a bounding box regression module, and a classification head, as follows: Use the RGB image I RGB as the input, pass it through the backbone network CSPDarknet to extract multi-scale features: ； Among them, is the convolutional feature; Backbone is the backbone network; Meanwhile, combine with the multi-modal feature F final Obtain the further fused feature : ； Where Concat represents the concatenation operation in the channel dimension; The bounding box regression module outputs the predicted normalized bounding box position B=(x,y,w,h), where (x,y) are the coordinates of the bounding box, and w,h are the width and height of the bounding box respectively. The classification head part uses a positional encoding mechanism to optimize the positioning accuracy of the bounding box and outputs the center coordinates as well as the width and height : ； ； Among them, t x 、t y 、t w 、t h are the offset values of the center coordinates and the width and height regressed by the bounding box regression network; σ is the activation function; c x 、c y are the offset values of the center coordinates of the predicted grid cell ; p w 、p h are the reference width and height of the corresponding grid cell; The final coordinates of the bounding box are: ； According to the fused features , predict the class probability distribution of each target through the classification head : ； Among them, W c b c are the weights and biases of the classification head.

7. An automatic sorting method for recycling resources combined with intelligent recognition according to claim 1, characterized in that According to the rough classification results and the location of the items, the items are sorted to the corresponding sub-sorting lines by the robotic arm, as follows: According to the final classification result c output by the classification model, the target detection model obtains the actual three-dimensional coordinates (Xreal, Yreal, Zreal) of the object location; Crawl phase: Calculate the grasping point Ppick=(Xreal, Yreal, Zreal) and the safety point Psafe-pick; use inverse kinematics to solve the joint angle and drive the robot arm to complete the grasping; Sorting stage: According to the classification result c, the target sorting point Ptarget=(Xtarget, Ytarget, Ztarget) is found, where (Xtarget, Ytarget, Ztarget) is the three-dimensional coordinate of the target sorting point; the path planning moves the item from the grabbing point to the target sorting point; Release phase: At the height of the target point Psafe-tage is lowered and the items are placed in the corresponding assembly line.

8. An automatic sorting method for recycling resources combined with intelligent recognition according to claim 1, characterized in that, The sub-sorting lines perform secondary classification of items, as follows: the plastic sorting line uses an NIR spectrometer to subdivide different categories; the metal sorting line uses an eddy current separator and a magnetic separation device to separate non-ferromagnetic and ferromagnetic metals; the glass sorting line uses an RGB camera to assist in color sorting; the paper sorting line classifies clean paper and contaminated paper.

9. An automatic sorting system for recycled resources combined with intelligent recognition, characterized in that, It includes a processor, a memory and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps in the automatic sorting method for recycling and reuse resources combined with intelligent identification as described in any one of claims 1 to 8.

10. A computer storage medium, characterized in that, The computer storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor and executing the steps in the automatic sorting method for recycling and reuse resources combined with intelligent identification as described in any one of claims 1 to 8.

Images