A method and system for intelligent classification and identification of recycled materials

Abstract

The present invention relates to the field of image recognition technology, and in particular to a method and system for intelligent classification and identification of recycled materials. A garbage area is divided into a first material garbage area and a second material garbage area according to the identification results of the material coding identification. A first membership degree can be obtained by comparing the contour similarity between the second material garbage area and the first material garbage area of ​​each material category. The first similarity between the representative garbage area and the second material garbage area in different appearance dimensions is compared, and the second membership degree is obtained by comparing the weight and the first similarity. A random forest decision tree model can be constructed based on the first membership degree and the second membership degree, and the decision conditions in the decision tree can be continuously adjusted according to the classification error rate of the training results during the training process. The present invention combines image recognition and machine learning to obtain accurate material classification results through a trained random forest decision tree model.

Description

Technical Field

[0001] The present invention relates to the field of image recognition technology, and in particular to a method and system for intelligent classification and recognition of recycled materials. Background Art

[0002] Plastic is a common recyclable resource. Plastic waste can be recycled and processed by sorting it according to its material, reducing environmental pollution and conserving production resources. Existing technology utilizes intelligent sorting equipment to intelligently sort a batch of waste, determine the material of each piece of waste, and classify it. Plastic waste comes in a variety of material categories, such as polyester (PET) (beverage bottles, etc.), polyethylene (PE) (plastic bags, plastic wrap, etc.), and polyvinyl chloride (PVC) (toys, containers, etc.). Only after classifying each material category can targeted recycling be carried out. For these material categories, material coding markings are printed or engraved on the surface of the plastic waste. Image recognition can be used to identify these surface material coding markings to achieve plastic waste classification. However, in actual classification, due to surface damage or dirt covering the plastic waste, the material coding markings in the image cannot be accurately identified, resulting in inaccurate plastic waste classification and requiring extensive manual intervention, making efficient and automatic plastic waste classification impossible. Summary of the Invention

[0003] In order to solve the technical problem in the prior art that plastic waste cannot be accurately identified due to the inability to accurately identify material coding identifiers in images, the purpose of the present invention is to provide a method and system for intelligent classification and identification of recycled materials. The technical solutions adopted are as follows:

[0004] The present invention proposes a method for intelligent classification and identification of recycled materials, which includes:

[0005] Obtaining a plastic waste image when the intelligent waste sorting device processes the same batch of plastic waste; the plastic waste image contains multiple plastic waste regions; identifying a material coding identifier on each plastic waste region in the plastic waste image, and defining the successfully identified plastic waste region as a first material waste region, and the unidentified plastic waste region as a second material waste region;

[0006] For each second material garbage region, obtaining a first degree of membership between each second material garbage region and each material category according to a contour similarity between the second material garbage region and the first material garbage region in each material category;

[0007] For each material category, a representative garbage region in each material category is obtained; a first similarity between the second material garbage region and the representative garbage region in different appearance dimensions is obtained, and a second similarity between the representative garbage regions in each appearance dimension is obtained; a comparison weight of the material category is obtained based on the second similarity; and a second degree of membership between the second material garbage region and the material category is obtained based on the comparison weight and the first similarity;

[0008] Using the second material garbage area as training data, constructing a random forest decision tree model based on the first membership degree and the second membership degree; during the training process of the random forest model, continuously updating the judgment thresholds of the decision tree for the first membership degree and the second membership degree based on the classification error rates of the decision tree in the two membership degree dimensions;

[0009] The trained random forest decision tree model is used to classify the materials of plastic waste areas where the material coding labels have not been successfully identified.

[0010] Furthermore, the method for obtaining the first degree of membership includes:

[0011] A contour-based template matching algorithm is used to obtain the contour similarity between the second material garbage area and the first material garbage area in each material category; in each material category, the mean of the contour similarities is used as the first membership between the second material garbage area and the material category.

[0012] Furthermore, the appearance dimension includes a color histogram and a gray-level co-occurrence matrix.

[0013] Furthermore, the method for obtaining the comparison weight includes:

[0014] For each appearance dimension, the mean of the second similarities between all representative garbage regions is used as the comparison weight of the corresponding material category under the appearance dimension.

[0015] Furthermore, the method for obtaining the second degree of membership includes:

[0016] The comparison weight is used as a weight to perform weighted summation on the first similarities in all appearance dimensions to obtain the second membership degree.

[0017] Furthermore, the method for updating the judgment thresholds of the first membership degree and the second membership degree includes:

[0018] For the judgment threshold corresponding to each membership in the first membership and the second membership, the difference between the judgment threshold and the preset maximum judgment threshold is used as the basic value, the product of the classification error rate under the corresponding membership dimension and the basic value is used as the adjustment amount, and the sum of the judgment threshold and the adjustment amount is used as the judgment threshold after this update.

[0019] Furthermore, an OCR algorithm is used to identify the material coding mark on each plastic waste area in the plastic waste image.

[0020] Furthermore, the plastic waste areas that have not been successfully identified are divided into unmarked waste areas and partially marked waste areas; for the partially marked waste areas, the highly correlated categories of the partially marked waste areas are determined based on the image similarity between the partial identification in the partially marked waste areas and the standard identification of each material category; in the random forest decision tree model, the partially marked waste areas are input into the decision tree corresponding to the highly correlated category for material classification.

[0021] Furthermore, the method for obtaining the image similarity includes:

[0022] The image corresponding to the partial logo and the standard logo are converted into vector form, and the Pearson correlation coefficient between the two image vectors is used as the image similarity.

[0023] The present invention proposes an intelligent classification system for recycled materials, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the system implements any one of the steps of the intelligent classification and identification method for recycled materials.

[0024] The present invention has the following beneficial effects:

[0025] The embodiment of the present invention first divides the garbage area into a first material garbage area and a second material garbage area according to the identification results of the material coding identifier. The first material garbage area is a garbage area of ​​a known material category, and therefore can be used as a reference for the second material garbage area. Because there is garbage with the same appearance between a batch of plastic garbage, the first membership can be obtained by comparing the contour similarity between the second material garbage area and the first material garbage area of ​​each material category. Further considering that each material category has relatively common garbage, representative garbage areas are screened out, and the first similarities between the representative garbage areas and the second material garbage areas in different appearance dimensions are further compared. Further considering that the more uniform the appearance features between the representative garbage areas, the stronger the reference and the more representative the representative garbage areas in the appearance dimension, the further comparative weight is obtained, and the second membership is obtained by comparing the comparative weight and the first similarity. A random forest decision tree model can be constructed based on the first and second membership degrees. Within this model, the material classification of the second material waste area can be determined individually based on the first and second membership degrees. During the training process, the decision conditions in the decision tree can be continuously adjusted based on the classification error rate of the training results. Ultimately, an accurate random forest decision tree model is obtained and used to classify the materials of plastic waste areas where the material coding identifiers are not successfully identified. Specifically, the present invention combines image recognition and machine learning. After obtaining the first and second membership degrees for the plastic waste area to be classified, these are input into the random forest decision tree model to obtain accurate material classification results. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. 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 any creative work.

[0027] Figure 1 A flow chart of a method for intelligent classification and identification of recycled materials provided by one embodiment of the present invention;

[0028] Figure 2 A schematic diagram of a working scenario of an intelligent garbage sorting device provided by one embodiment of the present invention;

[0029] Figure 3 A schematic diagram of a material coding identification provided by one embodiment of the present invention;

[0030] Figure 4 A decision tree diagram provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0031] To further illustrate the technical means and effectiveness of the present invention in achieving its intended objectives, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail the specific implementation, structure, features, and effectiveness of a method and system for intelligently classifying and identifying recycled materials according to the present invention. In the following description, references to "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.

[0032] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0033] The embodiment of the present invention aims to improve the garbage material classification effect based on the material coding identification results by improving the image recognition technology and machine learning technology in the working scenario of intelligent garbage sorting equipment. Figure 2 , which shows a schematic diagram of the working scenario of an intelligent waste sorting device provided by one embodiment of the present invention. During operation, the intelligent waste sorting device continuously collects multiple images of plastic waste through the production line and preliminarily identifies the material coding. It should be noted that before intelligent waste sorting equipment can operate, the waste to be processed generally requires certain pretreatment and cleaning to remove surface contaminants such as stains, grease, residue, and labels. After drying, it can enter the production line for intelligent sorting.

[0034] The specific scheme of the intelligent classification and identification method and system for recycled materials provided by the present invention is described in detail below with reference to the accompanying drawings.

[0035] See also Figure 1 , which shows a flow chart of a method for intelligent classification and identification of recycled materials provided by one embodiment of the present invention, the method comprising:

[0036] Step S1: Obtain a plastic waste image of the same batch of plastic waste processed by the intelligent waste sorting device; the plastic waste image contains multiple plastic waste regions. Identify the material coding identifier on each plastic waste region in the plastic waste image, and define the successfully identified plastic waste region as a first material waste region, while the unidentified plastic waste region as a second material waste region.

[0037] In this embodiment of the present invention, to maximize the acquisition of material coding identification on the surface of plastic waste, the intelligent waste sorting equipment should be equipped with a multi-view camera. This captures panoramic images of the waste, facilitating subsequent image recognition. The resulting plastic waste image is a spliced ​​panoramic image, representing the surface information of the waste within the current capture area. Furthermore, the camera's image acquisition frequency should be aligned with the production line's movement speed to ensure that each piece of plastic waste appears in only one plastic waste image.

[0038] The embodiment of the present invention uses a semantic segmentation network to segment and mark the plastic waste areas in the plastic waste image. The marking facilitates the subsequent comparison between the waste areas, and each plastic waste has only one marking number.

[0039] For plastic products, each type of plastic has a relatively single function when making a product. Usually, only one material will appear on a product, and a material code mark will be printed or engraved on the surface of the product. By identifying the material code mark, the material type of each plastic product can be determined. Figure 3 , which shows a schematic diagram of a material coding identification provided by one embodiment of the present invention. However, for plastic waste, material coding identification may be missing or incomplete due to factors such as damage and dirt, making accurate identification impossible. Therefore, it is necessary to effectively distinguish between successfully identified and unidentified waste areas. In this embodiment of the present invention, the successfully identified plastic waste area is designated as the first material waste area, and the unidentified plastic waste area is designated as the second material waste area. Because the first material waste area is the successfully identified waste area, it can be used as comparative information for subsequent analysis.

[0040] In one embodiment of the present invention, an OCR algorithm is used to identify the material coding mark on each plastic waste area in the plastic waste image. The OCR algorithm is an optical character recognition algorithm, such as Figure 3 As shown, the material coding identification area has obvious character information, so the algorithm can accurately identify the complete material coding identification.

[0041] Step S2: for each second material garbage region, obtain a first degree of membership between each second material garbage region and each material category according to the contour similarity between the second material garbage region and the first material garbage region in each material category.

[0042] Because the first material garbage area is a garbage area whose material category has been successfully identified, and it and the second material garbage area that has not been successfully identified are from the same batch of plastic garbage, the first material garbage area has a strong reference degree for the second material garbage area, and can therefore be used as reference information for each material category for comparative analysis of the second material garbage area.

[0043] Because the first and second material waste areas represent the same batch of waste on the intelligent waste sorting equipment, there's a high probability that they contain the same type of plastic waste. Therefore, for each second material waste area, the first degree of membership between each second material waste area and each material category can be determined by obtaining the contour similarity between the first material waste areas within each material category. That is, the greater the contour similarity between the first and second material waste areas within a material category, the more likely the second material waste area belongs to that material category.

[0044] Preferably, in one embodiment of the present invention, the method for obtaining the first degree of membership includes:

[0045] A contour-based template matching algorithm is used to obtain the contour similarity between the second material garbage region and the first material garbage region in each material category. In an embodiment of the present invention, a contour-based template matching algorithm is used to compare the contour similarity scores between the two regions. It should be noted that the smaller the contour similarity score of the algorithm, the more similar the two contours are. Therefore, the reciprocal of the contour similarity score is used as the contour similarity. It should be noted that the contour-based template matching algorithm can use algorithms such as the Hausdorff distance method and the Hu moment invariant method. The specific technical means are numerical methods used by those skilled in the art and are not limited or elaborated on here.

[0046] In each material category, each first material garbage region has a contour similarity with the second material garbage region, so the mean of the contour similarities is used as the first membership degree between the second material garbage region and the material category.

[0047] Step S3: For each material category, obtain a representative garbage area in each material category; obtain a first similarity between the second material garbage area and the representative garbage area in different appearance dimensions, and obtain a second similarity between the representative garbage areas in each appearance dimension; obtain a comparison weight of the material category based on the second similarity, and obtain a second degree of membership between the second material garbage area and the material category based on the comparison weight and the first similarity.

[0048] The first membership analysis is based on a comparison of plastic waste from the same batch. For each material category, there are relatively common products, such as polyester, which is commonly used in beverage bottles, and polyethylene, which is commonly used in plastic bags and plastic wrap. Therefore, this embodiment of the present invention further obtains representative garbage areas for each material category. These representative garbage areas are the most common garbage products within that material category. These products are also highly relevant for that material category and can be used as additional comparison information for the second material garbage area.

[0049] For a representative garbage area of ​​a material category, it has special appearance features such as color and texture. Therefore, the first similarity between the second material garbage area and the representative garbage area in different appearance dimensions can be obtained. That is, the greater the first similarity, the more similar the second material garbage area is to the representative garbage area in this appearance dimension. If the first similarity in all appearance dimensions is large, it means that the second material garbage area belongs more closely to this material category.

[0050] This embodiment of the present invention further considers that the reference degrees of representative garbage areas determined for a material category may vary. For multiple representative garbage areas within a material category, the more common characteristics they share and the more uniform their appearance, the more representative the appearance features of the representative garbage areas of that material category are of that material category, and the stronger the reference degree within that appearance dimension. Therefore, this embodiment of the present invention further determines a second similarity between the representative garbage areas within each appearance dimension. Based on this second similarity, a comparison weight for the material category is obtained. Based on this comparison weight and the first similarity, a second degree of membership between the second material garbage area and the material category is obtained. Specifically, the comparison weight is used as a reference weight to weight the first similarity within each appearance dimension, ultimately obtaining an accurate second degree of membership.

[0051] Preferably, in one embodiment of the present invention, the appearance dimension includes a color histogram and a gray-level co-occurrence matrix. Therefore, the method for obtaining the first similarity may include: converting the color histogram into a curve form, obtaining a difference curve between the two curves, performing negative correlation mapping on the average value of the difference value on the difference curve and normalizing it to obtain the first similarity under the appearance dimension; obtaining various texture features of the gray-level co-occurrence matrix, averaging the differences between the various texture features to obtain the overall difference, performing negative correlation mapping on the overall difference and normalizing it to obtain the first similarity under the appearance dimension. It should be noted that the second similarity is obtained in the same process as the first similarity, except that the objects of comparison are different, so the method for obtaining the second similarity will not be repeated.

[0052] It should be noted that the negative correlation mapping and normalization in the embodiment of the present invention can be achieved through basic mathematical means. In the embodiment of the present invention, the opposite number of the data is used as the power of an exponential function with a natural constant as the base, and the output result of the exponential function is the result of negative correlation mapping and normalization.

[0053] Preferably, in one embodiment of the present invention, the method for obtaining the comparison weight includes:

[0054] For each appearance dimension, there are multiple representative garbage areas in a material category. Multiple representative garbage areas can be compared with each other to obtain second similarities. Therefore, the second similarities obtained from all comparisons are counted, and the average of the second similarities between all representative garbage areas is used as the comparison weight of the corresponding material category in the appearance dimension.

[0055] Preferably, in one embodiment of the present invention, the method for obtaining the second degree of membership includes:

[0056] The contrast weight is used as the weight, and the first similarities under all appearance dimensions are weighted summed to obtain the second membership.

[0057] Step S4: Using the second material garbage area as training data, a random forest decision tree model is constructed based on the first membership and the second membership. During the training process, the random forest model continuously updates the judgment thresholds of the decision tree for the first membership and the second membership based on the classification error rate of the decision tree.

[0058] The first membership and the second membership are two-dimensional membership information obtained through different contrast information. It is necessary to judge the second material garbage area based on the two-dimensional membership information to obtain the final material category. The embodiment of the present invention takes into account that directly setting a fixed threshold will lead to inaccurate judgment due to unreasonable threshold setting. Based on the principle of machine learning, the second material garbage area is used as training data, and a random forest decision tree model is constructed based on the first membership and the second membership. That is, the decision conditions in the random forest decision tree model are the first membership and the second membership. Please refer to Figure 4 , which shows a decision tree diagram provided by an embodiment of the present invention, wherein P i,j_k represents the first membership of the jth second material garbage area in the i-th plastic garbage image in the k-th material category, Y1 is the judgment threshold of the first membership, Q i,j_k Y2 represents the second membership of the jth second-material waste region in the i-th plastic waste image to the k-th material category, and Y2 is the judgment threshold for the second membership. This embodiment of the present invention first sets two initial judgment thresholds and then continuously updates the judgment thresholds based on the classification error rate of the decision tree until the classification error rate reaches the expected result, at which point training is considered complete.

[0059] It should be noted that because this step is a model training step, the real material of the second material garbage area needs to be manually calibrated in advance. The calibration result is compared with the decision tree output result to determine whether it is a classification error. The ratio of the number of classification errors to the total number of second material garbage areas is the classification error rate. Figure 4 As shown in the figure, there are two judgment processes in a decision tree, that is, each membership degree corresponds to a decision result. The classification error rate of each membership degree dimension can be obtained by comparing the membership degree decision results of the two dimensions with the actual material results of the second material garbage area.

[0060] Preferably, in one embodiment of the present invention, the method for updating the judgment thresholds of the first membership degree and the second membership degree includes:

[0061] For each of the first and second membership degrees, the difference between the judgment threshold and the preset maximum judgment threshold is used as the base value. The product of the classification error rate for the corresponding membership dimension and the base value is used as the adjustment amount. The sum of the judgment threshold and the adjustment amount is used as the updated judgment threshold. In other words, the base value represents the maximum adjustment amount for this update; a larger classification error rate indicates that a larger adjustment amount is required to adjust the judgment threshold.

[0062] In the embodiment of the present invention, when the classification error rate of the membership is less than 0.1, it indicates that the judgment threshold of the membership dimension does not need to be updated again, until the judgment thresholds of the two membership dimensions do not need to be updated, and the model training is completed.

[0063] It should be noted that in the embodiment of the present invention, the first membership and the second membership are both normalized, so the value range of the judgment threshold is also between 0 and 1. The initial judgment thresholds of the two memberships are set to 0.6, and the maximum judgment threshold is 1.

[0064] Step S5: Use the trained random forest decision tree model to classify the plastic waste areas where the material coding identification is not successfully identified.

[0065] The trained random forest decision tree model includes a decision tree for each material category. For unidentified plastic waste areas with the material coding identifier, the first and second memberships are obtained and then input into the model to output the material category. It should be noted that the second material waste area in the above step is used as training data, so the actual material must be manually determined. However, the plastic waste area to be identified can be directly input into the random forest decision tree model to obtain the classification result.

[0066] Preferably, in one embodiment of the present invention, to reduce the amount of data computation, unidentified plastic waste areas are divided into unmarked waste areas and partially marked waste areas. For partially marked waste areas, the highly correlated categories are determined based on the image similarity between the partial markings within the partially marked waste areas and the standard markings for each material category. In a random forest decision tree model, the partially marked waste areas are input into the decision tree corresponding to the highly correlated category for material classification. This preprocessing method eliminates the meaningless membership degree acquisition process and decision tree classification process, reducing the amount of data computation.

[0067] Furthermore, the method for obtaining image similarity in the embodiment of the present invention includes:

[0068] The images corresponding to the partial logos and the standard logos were converted into vector form, and the Pearson correlation coefficient between the two image vectors was used as the image similarity. The material categories corresponding to the standard logos with a Pearson correlation coefficient greater than 0.6 were considered highly correlated categories.

[0069] After identifying the material of each plastic waste, the robotic arm of the intelligent waste sorting equipment can place the plastic waste of each material category in a centralized location for recycling. Specific recycling and reuse operations may include:

[0070] (1) Mechanical recycling: The plastic is crushed into small particles by mechanical equipment, and then the crushed particles are cleaned to remove pollutants and impurities. Finally, the clean particles are processed into plastic particles or products again.

[0071] (2) Chemical recycling: using chemical reactions to convert waste plastics into chemicals or energy, for example, breaking down plastics into low molecular weight compounds under high temperature and high pressure conditions, and then further converting the cracking products into solid products, such as synthetic resins or additives.

[0072] (3) Pyrolysis: Pyrolysis of long-chain polymer molecules into chemicals with smaller molecular weight and relatively simple structure under high temperature and high pressure conditions.

[0073] The recycling methods of plastic waste also include catalytic pyrolysis, hydrolysis, alcoholysis, etc. The details will not be explained here. You can choose the appropriate treatment method according to the specific material.

[0074] In summary, the embodiment of the present invention divides the garbage area into a first material garbage area and a second material garbage area according to the recognition results of the material coding identifier. The first membership degree can be obtained by comparing the contour similarity between the second material garbage area and the first material garbage area of ​​each material category. The first similarity between the representative garbage area and the second material garbage area in different appearance dimensions is compared, and the second membership degree is obtained by comparing the weight and the first similarity. A random forest decision tree model can be constructed based on the first membership degree and the second membership degree. During the training process, the decision conditions in the decision tree can be continuously adjusted according to the classification error rate of the training results, and finally a random forest decision tree model with accurate decision-making is obtained and used to classify the materials of plastic garbage areas where the material coding identifier is not successfully identified. The present invention combines image recognition and machine learning. After obtaining the first membership degree and the second membership degree of the plastic garbage area to be classified, the accurate material classification results can be obtained by inputting them into the random forest decision tree model.

[0075] Based on the same inventive concept, the present invention also discloses a system for intelligent classification of recycled materials, including a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the computer program, it implements any one of the steps of a method for intelligent classification and identification of recycled materials.

[0076] It should be noted that the order in which the embodiments of the present invention are described above is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0077] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

Claims

1. A method for intelligent classification and identification of recycled materials, characterized in that: The method comprises: Obtaining a plastic waste image when the intelligent waste sorting device processes the same batch of plastic waste; the plastic waste image contains multiple plastic waste regions; identifying a material coding identifier on each plastic waste region in the plastic waste image, and defining the successfully identified plastic waste region as a first material waste region, and the unidentified plastic waste region as a second material waste region; For each second material garbage region, obtaining a first degree of membership between each second material garbage region and each material category according to a contour similarity between the second material garbage region and the first material garbage region in each material category; For each material category, a representative garbage region in each material category is obtained; a first similarity between the second material garbage region and the representative garbage region in different appearance dimensions is obtained, and a second similarity between the representative garbage regions in each appearance dimension is obtained; a comparison weight of the material category is obtained based on the second similarity; and a second degree of membership between the second material garbage region and the material category is obtained based on the comparison weight and the first similarity; Using the second material garbage area as training data, constructing a random forest decision tree model based on the first membership degree and the second membership degree; during the training process of the random forest model, continuously updating the judgment thresholds of the decision tree for the first membership degree and the second membership degree based on the classification error rates of the decision tree in the two membership degree dimensions; The trained random forest decision tree model is used to classify the materials of plastic waste areas where the material coding labels have not been successfully identified.

2. The method for intelligent classification and identification of recycled materials according to claim 1, characterized in that: The method for obtaining the first degree of membership includes: A contour-based template matching algorithm is used to obtain the contour similarity between the second material garbage area and the first material garbage area in each material category; in each material category, the mean of the contour similarities is used as the first membership between the second material garbage area and the material category.

3. The method for intelligent classification and identification of recycled materials according to claim 1, characterized in that: The appearance dimension includes a color histogram and a gray-level co-occurrence matrix.

4. A method for intelligent classification and identification of recycled materials according to claim 1 or 3, characterized in that: The method for obtaining the comparison weight includes: For each appearance dimension, the mean of the second similarities between all representative garbage regions is used as the comparison weight of the corresponding material category under the appearance dimension.

5. The method for intelligent classification and identification of recycled materials according to claim 1, characterized in that: The method for obtaining the second degree of membership includes: The comparison weight is used as a weight to perform weighted summation on the first similarities in all appearance dimensions to obtain the second membership degree.

6. The method for intelligent classification and identification of recycled materials according to claim 1, characterized in that: The method for updating the judgment thresholds of the first membership degree and the second membership degree includes: For the judgment threshold corresponding to each membership in the first membership and the second membership, the difference between the judgment threshold and the preset maximum judgment threshold is used as the basic value, the product of the classification error rate under the corresponding membership dimension and the basic value is used as the adjustment amount, and the sum of the judgment threshold and the adjustment amount is used as the judgment threshold after this update.

7. The method for intelligent classification and identification of recycled materials according to claim 1, characterized in that: The OCR algorithm is used to identify the material coding mark on each plastic waste area in the plastic waste image.

8. The method for intelligent classification and identification of recycled materials according to claim 1, characterized in that: The unidentified plastic waste areas are divided into unmarked waste areas and partially marked waste areas. For the partially marked waste areas, the highly relevant categories of the partially marked waste areas are determined based on the image similarity between the partial marks in the partially marked waste areas and the standard marks of each material category. In the random forest decision tree model, the partially identified garbage areas are input into the decision tree corresponding to the highly correlated category for material classification.

9. The method for intelligent classification and identification of recycled materials according to claim 8, characterized in that: The method for obtaining the image similarity includes: The image corresponding to the partial logo and the standard logo are converted into vector form, and the Pearson correlation coefficient between the two image vectors is used as the image similarity.

10. A system for intelligent classification of recycled materials, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method for intelligent classification and identification of recycled materials as described in any one of claims 1 to 9 are implemented.

Images