Regenerated aluminum material quality grading method and system based on image recognition

Abstract

The invention belongs to the technical field of image recognition, and particularly relates to a recycled aluminum material quality grading method and system based on image recognition, the recycled aluminum material quality grading system comprises a feeding module, a preprocessing module, an image acquisition module, an image processing and feature extraction module and a grading module; aiming at the problems of low quality grading efficiency, strong subjectivity, insufficient precision and the like in the recycled aluminum material recovery process, the method obtains the surface and cross-section images of the recycled aluminum material through the image collection module, optimizes the image quality through a preprocessing algorithm, extracts features in combination with a deep learning model, recognizes the related features of defects and components, and improves the quality of the recycled aluminum material. And finally, automatic and precise quality grading of the recycled aluminum materials is achieved.

Description

Technical Field

[0001] This invention belongs to the field of image recognition technology, specifically a method and system for quality grading of recycled aluminum materials based on image recognition. Background Technology

[0002] With the advancement of the global low-carbon economy and the implementation of resource recycling strategies, the recycled aluminum industry, as a core component of the green development of the aluminum industry, has ushered in a period of rapid development. Compared with primary aluminum, recycled aluminum can save approximately 95% of energy consumption and reduce carbon emissions by more than 70%, with wide application demands in various fields such as construction, automobiles, home appliances, and aerospace. According to industry statistics, global recycled aluminum production accounts for more than 50% of total aluminum production. As a major aluminum consumer, my country has maintained an average annual growth rate of over 8% in recycled aluminum production, and the market size continues to expand.

[0003] Quality grading is a crucial link in the recycled aluminum production and processing chain, directly determining the subsequent application scenarios, processing technology selection, and final product quality of recycled aluminum. Currently, quality grading of recycled aluminum mainly relies on traditional manual sorting and simple physical testing methods, which have several prominent problems: First, manual grading is inefficient, heavily influenced by subjective factors such as operator experience and fatigue, making it difficult to guarantee grading accuracy, typically only 75%-85%, which cannot meet the quality stability requirements of large-scale production; second, traditional physical testing methods (such as hardness testing and specific gravity testing) are mostly destructive, leading to raw material waste and long testing cycles, making real-time online grading impossible; third, recycled aluminum comes from complex sources with diverse surface and internal defect types (such as cracks, oxide scale, pores, inclusions, etc.), and significant differences in compositional uniformity, making it difficult for traditional methods to comprehensively and accurately identify these characteristic parameters, resulting in difficulties in uniformly implementing grading standards. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies and solve the aforementioned technical problems, this invention proposes a method and system for quality grading of recycled aluminum materials based on image recognition.

[0005] The technical solution adopted by this invention to solve its technical problem is as follows: This invention proposes a method for quality grading of recycled aluminum materials based on image recognition, and the specific steps of the method are as follows: S1, Pre-treatment of recycled aluminum: The pre-treatment module cleans the surface of the recycled aluminum material of the batch to be tested, removes impurities adhering to the surface of the recycled aluminum material, and reduces the error caused by impurities in the image recognition of the surface of the recycled aluminum material. S2, Image Data Acquisition: The recycled aluminum material is placed on the conveyor belt of the feeding module, which moves the recycled aluminum material through the support frame. The image acquisition module arranged on the support frame acquires image data of the recycled aluminum material from different angles and transmits the image data to the image processing module. S3, Image Processing and Feature Extraction: Based on the image processing and feature extraction module, the acquired image data is preprocessed, and the defect features in the image data are automatically extracted through a pre-trained deep learning model to identify the surface defect features of recycled aluminum materials and form a defect data set for this batch of recycled aluminum materials. S4, Quality Grading: By pre-determining quality grading standards and inputting them into the grading module, the defect data group is used as data support. The data is matched and compared with the data indicators in the quality grading standards, and the grading result of this batch of recycled aluminum is output.

[0006] Preferably, the image data preprocessing method in S3 above includes the following steps: S311, Image Denoising: Gaussian filtering algorithm is used to remove Gaussian noise from the acquired image data, and then median filtering algorithm is used to remove salt-and-pepper noise from the image data; S312, Image Enhancement: The contrast in image data is improved by using a histogram equalization algorithm to highlight the difference between the defect area and the background, which facilitates the subsequent identification and extraction of defect features; S313, Image Segmentation: A threshold segmentation algorithm is used to separate the recycled aluminum material area from the background in the image data to obtain the region of interest; for images with complex backgrounds, an edge detection algorithm is combined to optimize the segmentation effect. S314, Image Normalization: Adjust the segmented image to a uniform size and normalize the pixel grayscale values ​​to the same range, so as to facilitate uniform processing by subsequent deep learning models.

[0007] The preferred method for training deep learning modules includes the following specific steps: S321, Model Selection: Select an improved model based on convolutional neural networks as the core recognition model; S322, Construct a dataset: Collect image samples of recycled aluminum materials of different quality grades and different defect types. The number of samples should be no less than 100,000. Label the relevant data such as the type, location, and size of the defect features in the samples. Then divide them into training set, validation set and test set in a ratio of 7:2:1. S323, Model Training: Train the model using a deep learning framework, setting parameters such as learning rate, batch size, and number of iterations, and optimizing the model parameters using the cross-entropy loss function until the trained deep learning model achieves a recognition accuracy of ≥96% on the validation set. S324, Model Optimization: Data augmentation techniques are used to expand the dataset and improve the robustness of the final deep learning model.

[0008] Preferably, the industrial camera in the image acquisition module is arranged on the lower top surface and the inner walls of both sides of the support frame of the feeding module, and the support frame is also equipped with a positioning sensor to locate the moving aluminum material to be tested.

[0009] A protective cover is installed on the outside of the working end of the industrial camera. An observation port is set in the middle of the protective cover. A protective lens is installed inside the observation port. The protective cover is located on the outside of the working end of the industrial camera, and the outer surface of the protective lens is covered with a protective film.

[0010] Preferably, an installation groove is provided on the inner wall of the protective cover at the location corresponding to the protective film, and the protective film and the surface of the protective lens are in sliding contact. A recycling roller is provided inside the installation groove, and the recycling roller is connected to the output end of the drive device provided on the inner wall of the installation groove.

[0011] Preferably, a cooling fan is provided inside the industrial camera, and a cooling groove is provided on the outer surface of the industrial camera housing. The gap area between the inner wall of the protective cover and the acquisition end of the industrial camera is connected to the cooling groove. A connecting hole is provided in the middle part of the protective lens, and the connecting hole is connected to the gap area between the protective lens and the protective film.

[0012] Preferably, the vertical cross-section of the mounting groove is L-shaped. After the protective film extends into the mounting groove along the vertical part, it passes around the guide roller set at the turning part of the mounting groove, and the movement trajectory changes to the horizontal part. It is then collected on the recovery roller along the horizontal part of the mounting groove. A guide groove is provided on the inner wall of the mounting groove below the guide roller, and the opening of the guide groove points to the outer surface of the protective film in the observation port.

[0013] Preferably, guide blocks are evenly arranged on the lower top surface of the horizontal part of the installation groove, with the side of the guide block near the recovery roller being a vertical surface and the side away from the recovery roller being a horizontal surface.

[0014] A quality grading system for recycled aluminum based on image recognition is provided. The recycled aluminum quality grading system is applicable to the above-mentioned recycled aluminum quality grading method. The recycled aluminum quality grading system includes a feeding module, a preprocessing module, an image acquisition module, an image processing and feature extraction module, and a grading module.

[0015] The beneficial effects of this invention are as follows: The present invention provides a method and system for quality grading of recycled aluminum materials based on image recognition. This method achieves efficient and automated quality grading of recycled aluminum materials. Compared with manual grading and screening, it improves the accuracy and efficiency of quality grading of recycled aluminum materials, thereby improving the recycling efficiency of recycled aluminum materials and ensuring product quality. Attached Figure Description

[0016] The invention will now be further described with reference to the accompanying drawings.

[0017] Figure 1 This is a flowchart of the quality grading method for recycled aluminum materials in this invention; Figure 2 This is a flowchart of the image preprocessing method in this invention; Figure 3 This is a perspective view of the feeding module of the present invention; Figure 4 This is a partial sectional view of the support frame in this invention; Figure 5 yes Figure 4 A magnified view of a section at point A in the middle; Figure 6 yes Figure 5 A magnified view of a section at point B in the middle; Figure 7 yes Figure 5 A magnified view of a section at point C.

[0018] In the diagram: Industrial camera 1, protective cover 11, observation port 111, protective lens 112, protective film 113, mounting groove 114, recovery roller 115, connecting hole 116, guide block 117, guide groove 118, guide hole 119, cooling groove 12, guide roller 13, adjusting protrusion 131, conveyor belt 2, support frame 3. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1:

[0021] A method for quality grading of recycled aluminum based on image recognition, the specific steps of which are as follows: S1, Pre-treatment of recycled aluminum: The pre-treatment module cleans the surface of the recycled aluminum material of the batch to be tested, removes impurities adhering to the surface of the recycled aluminum material, and reduces the error caused by impurities in the image recognition of the surface of the recycled aluminum material. Specifically, the pre-processing module includes high-pressure spray cleaning equipment and hot air dryers. Large non-metallic impurities, such as plastic fragments, wood blocks, and stones, are removed from the recycled aluminum material through manual or automated sorting equipment. The recycled aluminum material is then placed in the high-pressure spray cleaning equipment, where a cleaning solution prepared based on a neutral cleaning agent is used to clean the surface of the recycled aluminum material, removing dirt, oil, and other impurities adhering to the surface. The cleaned recycled aluminum material is then sent to the hot air dryer, where it is dried with hot air to prevent residual water stains on the surface from affecting subsequent surface image data acquisition. To facilitate transportation and handling, the stacked recycled aluminum materials are classified into multiple batches according to the sorting time and the size of the recycled aluminum materials. The recycled aluminum materials in each batch are similar in size and within the same range, which facilitates positioning and shooting during image acquisition. S2, Image Data Acquisition: The recycled aluminum material is placed on the conveyor belt 2 of the feeding module, which moves the recycled aluminum material through the support frame 3. The image acquisition module arranged on the support frame 3 acquires image data of the recycled aluminum material at different angles and transmits the image data to the image processing module. The image acquisition module can use an existing industrial camera 1, which is arranged in various positions on the inner wall of the support frame 3. When the recycled aluminum material passes through the support frame 3 along the conveyor, the industrial camera 1 can acquire surface image data of the recycled aluminum material from multiple angles, which makes it easy to capture defects such as cracks or oxide scale that are not easy to be found on the side of the recycled aluminum material. To address the need to collect data on internal defects in recycled aluminum, several samples can be extracted from the corresponding batch of recycled aluminum, and after cutting to expose the cross-section, cross-sectional image data of these samples can be collected as data support for later analysis of internal defects in aluminum. S3, Image Processing and Feature Extraction: Based on the image processing and feature extraction module, the acquired image data is preprocessed, and the defect features in the image data are automatically extracted through a deep learning model to identify the surface defect features of recycled aluminum materials and form a defect data set for this batch of recycled aluminum materials. The preprocessing of the acquired image data is mainly used to eliminate image noise, enhance defect features, and improve the accuracy and efficiency of subsequent deep learning model recognition. The deep learning model is pre-trained and can identify defect features such as cracks, oxide scale, and pores on the surface of recycled aluminum materials, and mark the location of these defect features and quantify the size of these defect features, such as crack length, oxide scale area, and pore diameter. S4, Quality Grading: By pre-determining quality grading standards and inputting them into the grading module, and using defect data groups as data support, the module matches and compares the data indicators with those in the quality grading standards, automatically outputting the grading results for this batch of recycled aluminum materials and generating a grading report.

[0022] The formulation of quality grading standards for recycled aluminum materials should be based on relevant industry-specific regulations, combined with the processing and market demands of the applicant's company and local area. The standards should be divided into several levels. Different levels of recycled aluminum materials within the same batch should be categorized to achieve in-depth classification. Recycled aluminum materials of poor quality that do not meet the required grades should be remelted and refined to improve their quality. A grading database can be established to record the source, processing time, grading results, and defect details of each batch of recycled aluminum materials, facilitating subsequent quality traceability and process optimization.

[0023] Regarding the specific grading standards, this embodiment provides a possible implementation plan. Specifically, the quality of recycled aluminum materials can be divided into five grades: Grade 1 quality: No cracks, no obvious oxide scale, scratch depth ≤0.1mm, area ≤1cm²; Second-level quality: No cracks, oxide scale area ≤ 5cm², scratch depth ≤ 0.3mm, area ≤ 3cm²; Level 3 quality: minor cracks (length ≤ 1cm), oxide scale area ≤ 10cm², scratch depth ≤ 0.5mm; Level 4 quality: Obvious cracks (length ≤ 3cm), oxide scale area ≤ 20cm², scratch depth ≤ 1mm Level 5 quality: severe cracks (length > 3cm), oxide scale area > 20cm², scratch depth > 1mm; The above grading standards can be deployed in the computing device of the grading module. By comparing and analyzing the defect feature dataset identified and extracted by the deep learning module, the quality grade of the batch of aluminum materials can be determined.

[0024] In summary, this application achieves highly efficient and automated quality grading of recycled aluminum materials through a quality grading method. Compared with manual grading and screening, it improves the accuracy and efficiency of quality grading of recycled aluminum materials, thereby enhancing the recycling efficiency of recycled aluminum materials and effectively ensuring product quality.

[0025] Example 2:

[0026] Based on Embodiment 1, there are various possible implementation schemes for image preprocessing methods. Any scheme that meets the above-mentioned requirements of this application can be applied to this application. This embodiment provides one possible implementation scheme, with specific steps including: S311, Image Denoising: Gaussian filtering algorithm is used to remove Gaussian noise from the acquired image data, and then median filtering algorithm is used to remove salt-and-pepper noise from the image data to avoid noise being misidentified as defects; S312, Image Enhancement: The contrast in image data is improved by using a histogram equalization algorithm to highlight the difference between defect areas and the background, which facilitates the identification and extraction of subsequent defect features. For images with uneven lighting, an adaptive histogram equalization algorithm should be used to avoid local overexposure or underexposure. S313, Image Segmentation: A threshold segmentation algorithm is used to separate the aluminum region from the background in the image data to obtain the region of interest; for images with complex backgrounds, an edge detection algorithm is combined to optimize the segmentation effect. S314, Image Normalization: Adjust the segmented image to a uniform size, such as 800×600 pixels, and normalize the pixel grayscale values ​​to the same range, such as 0-255, so that subsequent deep learning models can process them uniformly.

[0027] Furthermore, regarding the selection and training of deep learning modules, this embodiment provides a possible implementation scheme, with the following specific steps: S321, Model Selection: Based on the requirement to identify defect features in the surface image data of recycled aluminum, this application may select an improved model based on convolutional neural networks as the core identification model, such as ResNet-50 and YOLOv8 detection algorithms. ResNet-50 can be used for feature extraction and composition uniformity analysis, while YOLOv8 can be used for real-time defect detection and localization. S322, Construct a dataset: Collect image samples of recycled aluminum materials of different quality grades and different defect types. The number of samples should be no less than 100,000. Label the relevant data such as the type, location, and size of the defect features in the samples. Then divide them into training set, validation set and test set in a ratio of 7:2:1. S323, Model Training: The model was trained using the PyTorch / TensorFlow deep learning framework with a learning rate of 0.001, a batch size of 32, and 100 iterations. The model parameters were optimized using the cross-entropy loss function until the model achieved a recognition accuracy of ≥96% on the validation set. S324, Model Optimization: Data augmentation techniques, such as random cropping, flipping, rotating, and brightness adjustment, are used to expand the dataset and improve the robustness of the final deep learning model; transfer learning techniques can also be used to shorten the training time by utilizing pre-trained model parameters.

[0028] Example 3:

[0029] Based on Embodiment 2, the image acquisition module includes an industrial camera 1, which is arranged on the conveyor belt 2 of the feeding module. Specifically, the conveyor belt 2 includes a feeding trough, and feeding rollers are evenly arranged on the inner wall of the feeding trough. The feeding rollers are used to push the recycled aluminum material along the feeding trough to the acquisition position. An inverted U-shaped support frame 3 is arranged on the feeding trough above the acquisition position. The industrial camera 1 is arranged on the lower surface of the top and the inner walls of both sides of the support frame 3, surrounding and pointing to the recycled aluminum material that has moved to the acquisition position. A protective cover 11 is provided on the outside of the working end of the industrial camera 1. An observation port 111 is provided in the middle of the protective cover 11. A protective lens 112 is provided inside the observation port 111. The protective cover 11 is located on the outside of the working end of the industrial camera 1, and the outer surface of the protective lens 112 is covered with a protective film 113. During operation, the industrial camera 1 configured on the support frame 3 located at the acquisition position is activated. The industrial cameras 1 installed at multiple positions can acquire surface images of the recycled aluminum material at the central acquisition position from multiple angles, and transmit the corresponding image data to the image data processing module to provide data support for analyzing the quality standards of recycled aluminum material. Furthermore, to better protect the working lens of the industrial camera 1 and prevent processing impurities stirred up by the recycled aluminum material during movement from adhering to the working lens of the industrial camera 1 and affecting the surface quality of the working lens, as well as the adhesion of dust and impurities will also be reflected in the acquired images, interfering with image analysis and affecting the accuracy of quality grading of the recycled aluminum material, this application chooses to nest a protective cover 11 on the outside of the working lens of the industrial camera 1. When the industrial camera 1 is started, the working end can capture the surface image of the external recycled aluminum material through the protective lens 112 of the observation port 111 in the middle of the protective cover 11. The stirred-up dust and impurities are also intercepted on the outside of the protective film 113 of the protective lens 112, preventing these processing impurities from affecting the safety of the working lens of the industrial camera 1.

[0030] Furthermore, an installation groove 114 is provided on the inner wall of the protective cover 11 at the location corresponding to the protective film 113, and the protective film 113 and the surface of the protective lens 112 are in sliding contact. A recovery roller 115 is provided inside the installation groove 114, and the recovery roller 115 is connected to the output end of the micro motor drive device provided on the inner wall of the installation groove 114. A cooling fan is provided inside the industrial camera 1, and a cooling groove 12 is provided on the outer surface of the housing of the industrial camera 1. A filter screen is provided inside the cooling groove 12, and the gap area between the inner wall of the protective cover 11 and the acquisition end of the industrial camera 1 is connected to the cooling groove 12. A connecting hole 116 is provided in the middle part of the protective lens 112, and the connecting hole 116 is connected to the gap area between the protective lens 112 and the protective film 113. A cooling fan located inside the industrial camera 1 is activated periodically, sending external cooling airflow into the cooling tank 12. This airflow acts on the electrical components inside the industrial camera 1, cooling them down. After cooling, some of the airflow enters the area surrounded by the protective cover 11, and then enters the gap area between the protective lens 112 and the protective film 113 through the connecting hole 116 in the middle of the protective lens 112. This increases the air pressure in the gap area, pushing the protective film 113 to separate from the protective lens 112. Simultaneously, the micro motor connected to the recovery roller 115 is activated, driving the recovery roller to... Roller 115 rotates slowly, causing the protective film 113 facing the protective lens 112 to detach and be wound onto the upper recovery roller 115. Meanwhile, a new protective film 113 is released from the bottom and re-covers the protective lens 112. This allows for the replacement of the original protective film 113 when impurities adhere to its surface. The new protective film 113 has a clean surface and can transmit images normally, reducing interference from surface impurities during image acquisition. As the original protective film 113 moves upward, the airflow in the gap between the protective film 113 and the protective lens 112 flows upward, causing the surface of the moving protective film 113 to vibrate. The airflow enters the mounting groove 114 and then flows out of the opening of the mounting groove 114, bypassing the recovery roller 115, to wash the outer surface of the protective film 113. This causes both sides of the protective film 113 to be washed by the airflow, accelerating the removal of impurities adhering to the surface of the protective film 113, thereby effectively cleaning the protective film 113. The cleaned protective film 113 is collected on the surface of the recovery roller 115 on the upper side of the groove. In this way, when it is necessary to replace the protective film 113 corresponding to the protective lens 112 later, the micro motor can be started in reverse to drive the upper recovery roller 115 to release the protective film 113, so that it moves down to replace the protective film 113 corresponding to the protective lens 112, ensuring the accuracy of image acquisition by the internal industrial camera 1.

[0031] Example 4:

[0032] Based on Embodiment 3, the vertical cross-section of the mounting groove 114 is L-shaped. After the protective film 113 extends into the mounting groove 114 along the vertical part, it passes around the guide roller 13 provided at the turning part of the mounting groove 114, and its movement trajectory changes to a horizontal part. It is then collected on the recovery roller 115 along the horizontal part of the mounting groove 114. Furthermore, guide blocks 117 are evenly provided on the lower surface of the top of the horizontal part of the mounting groove 114. The side of the guide block 117 near the recovery roller 115 is a vertical surface, and the side facing away from the recovery roller 115 is a horizontal surface. In addition, a guide groove 118 is provided on the inner wall of the mounting groove 114 at the part below the guide roller 13. The opening of the guide groove 118 points towards the outer surface of the protective film 113 in the observation port 111. Regarding the structural features of the mounting groove 114, when the airflow flows along the gap area between the protective film 113 and the protective lens 112 to the mounting groove 114 on both the upper and lower sides, it first enters the gap area between the inner surface of the mounting groove 114 and the inner surface of the protective film 113 and flows into the interior of the mounting groove 114. Then, after bypassing the horizontal part where the recovery roller 115 is located inside the mounting groove 114, it flows again to the opening of the vertical part of the mounting groove 114 and flows to the outside from the guide groove 118 at the opening of the mounting groove 114. During this process, both sides of the protective film 113 inside the mounting groove 114 are scoured by the airflow, causing the impurity particles adhering to the outer surface of the protective film 113 to fall off and flow to the outside with the airflow, leaving the surface of the protective film 113. Furthermore, a guide roller 13 is provided at the turning point where the movement trajectory of the protective film 113 inside the mounting groove 114 changes from vertical to horizontal. The guide roller 13 is rotatably connected and is located on the airflow path inside the mounting groove 114. This allows the airflow to be guided to the gap area between the guide roller 13 and the protective film 113 when it passes the position of the guide roller 13. The protective film 113 deforms significantly at the bending point. As the protective film 113 deforms, the adhesion between it and the dust and impurities adhering to its surface changes. Combined with the concentrated impact caused by the change in the flow path of the outflowing airflow at the bending point, the particulate impurities adhering to the surface of the protective film 113 are detached under the combined action of deformation vibration and airflow scouring at the bending point. Furthermore, in the horizontal portion of the mounting groove 114, guide blocks 117 are evenly arranged near the inner surface of the protective film 113. The airflow flowing into the mounting groove 114 flows along the gap between the guide blocks 117 and the inner surface of the protective film 113, forming a continuously zigzagging flow trajectory along the inclined surface of the guide blocks 117. During this process, the impact on the surface of the protective film 113 continuously changes, causing the protective film 113 to vibrate. The vibration from the inner part affects the outer surface, causing dust and impurities adhering to the outer surface of the protective film 113 to fall off and leave the protective film 113 with the airflow exiting the mounting groove 114. This thoroughly purifies the protective film 113, ensuring that when this part of the protective film 113 is subsequently moved back to the observation port 111, the interference of dust and impurities adhering to the protective film 113 on the image data is reduced, thereby improving the accuracy of extracting defect features from the image data later.

[0033] Furthermore, the protective film 113 is provided with multiple sets of guide holes 119, and the outer surface of the guide roller 13 is uniformly provided with arc-shaped adjustment protrusions 131, and the number of adjustment protrusions 131 is odd. The guide hole 119 on the protective film 113 has a small aperture, and the protective film 113 is usually tightly attached to the surface of the protective lens 112. This can prevent external dust and impurities from penetrating inward and threatening the surface safety of the protective lens 112. Only when the purified cooling airflow flowing in from the cooling tank 12 fills the gap area between the protective lens 112 and the protective film 113, does the air pressure in the gap area increase, and some airflow permeates outward from the guide hole 119. This causes the area near the outer surface of the protective film 113 in the observation port 111 to form an airflow trend away from the protective film 113. Combined with the transverse impact airflow flowing out from the guide tank 118 from the edge of the observation port 111, the airflow impacts each other and forms a multi-directional impact on the surface of the protective film 113. This can prevent particulate impurities mixed in the airflow from adhering to the surface of the protective film 113 again, thereby ensuring the cleanliness of the surface of the protective film 113. Furthermore, an odd number of adjustment protrusions 131 are provided on the surface of the guide roller 13, and the adjustment protrusions 131 can be made of a flexible material such as sponge, specifically for cleaning the surface of the protective film 113. The limitation on the number ensures that when the protruding parts of the guide roller 13, such as the adjustment protrusions 131, contact the surface of the protective film 113 in the vertical direction, the recessed parts on the guide roller 13 located in the gap between the adjustment protrusions 131 move to the inner wall of the mounting groove 114 near the guide channel 118. At this time, the gap between the guide roller 13 and the inner wall of the mounting groove 114 increases, and airflow with a flow tendency flows from the guide roller 13 and the mounting groove 114. The airflow is concentrated in the gap area between the inner walls; conversely, when the concave part of the guide roller 13 rotates to a position close to the surface of the protective film 113, the convex part of the guide roller 13 rotates to a position close to the inner wall of the mounting groove 114, which intercepts the airflow and causes more airflow to flow out from the gap area between the concave part of the guide roller 13 and the protective film 113. As the guide roller 13 rotates, the above process is repeated. The periodic airflow impact on the surface of the protective film 113 can effectively remove the dust and impurities adhering to the protective film 113, improve the cleaning efficiency of the protective film 113, and thus ensure the normal operation of the industrial camera 1.

[0034] Example 5:

[0035] Based on the above embodiments, a quality grading system for recycled aluminum based on image recognition is provided. The quality grading system for recycled aluminum is applicable to the above-mentioned quality grading method for recycled aluminum. The quality grading system for recycled aluminum includes a feeding module, a preprocessing module, an image acquisition module, an image processing and feature extraction module, and a grading module.

[0036] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for quality grading of recycled aluminum materials based on image recognition, characterized in that, The specific steps of the method for quality grading of recycled aluminum materials are as follows: S1, Pre-treatment of recycled aluminum: The pre-treatment module cleans the surface of the recycled aluminum material of the batch to be tested, removes impurities adhering to the surface of the recycled aluminum material, and reduces the error caused by impurities in the image recognition of the surface of the recycled aluminum material. S2, Image Data Acquisition: The recycled aluminum material is placed on the conveyor belt of the feeding module, which moves the recycled aluminum material through the support frame. The image acquisition module arranged on the support frame acquires image data of the recycled aluminum material from different angles and transmits the image data to the image processing module. S3, Image Processing and Feature Extraction: Based on the image processing and feature extraction module, the acquired image data is preprocessed, and the defect features in the image data are automatically extracted through a pre-trained deep learning model to identify the surface defect features of recycled aluminum materials and form a defect data set for this batch of recycled aluminum materials. S4, Quality Grading: By pre-determining quality grading standards and inputting them into the grading module, and using defect data groups as data support, the module matches and compares the data indicators with those in the quality grading standards to output the grading results for this batch of recycled aluminum.

2. The method for quality grading of recycled aluminum based on image recognition according to claim 1, characterized in that: Regarding the image data preprocessing method in S3 above, the specific steps include: S311, Image Denoising: Gaussian filtering algorithm is used to remove Gaussian noise from the acquired image data, and then median filtering algorithm is used to remove salt-and-pepper noise from the image data; S312, Image Enhancement: The contrast in image data is improved by using a histogram equalization algorithm to highlight the difference between the defect area and the background, which facilitates the subsequent identification and extraction of defect features; S313, Image Segmentation: A threshold segmentation algorithm is used to separate the recycled aluminum material area from the background in the image data to obtain the region of interest; for images with complex backgrounds, an edge detection algorithm is combined to optimize the segmentation effect. S314, Image Normalization: Adjust the segmented image to a uniform size and normalize the pixel grayscale values ​​to the same range, so as to facilitate uniform processing by subsequent deep learning models.

3. The method for quality grading of recycled aluminum based on image recognition according to claim 1, characterized in that: The specific steps for training the deep learning module are as follows: S321, Model Selection: Select an improved model based on convolutional neural networks as the core recognition model; S322, Construct a dataset: Collect image samples of recycled aluminum materials of different quality grades and different defect types. The number of samples should be no less than 100,000. Label the relevant data such as the type, location, and size of the defect features in the samples. Then divide them into training set, validation set and test set in a ratio of 7:2:

1. S323, Model Training: Train the model using a deep learning framework, setting parameters such as learning rate, batch size, and number of iterations, and optimizing the model parameters using the cross-entropy loss function until the trained deep learning model achieves a recognition accuracy of ≥96% on the validation set. S324, Model Optimization: Data augmentation techniques are used to expand the dataset and improve the robustness of the final deep learning model.

4. The method for quality grading of recycled aluminum based on image recognition according to claim 1, characterized in that: The industrial camera in the image acquisition module is arranged on the lower top surface and the inner walls of both sides of the support frame of the feeding module. The support frame is also equipped with positioning sensors to locate the moving aluminum material to be tested. A protective cover is installed on the outside of the working end of the industrial camera. An observation port is set in the middle of the protective cover. A protective lens is installed inside the observation port. The protective cover is located on the outside of the working end of the industrial camera, and the outer surface of the protective lens is covered with a protective film.

5. The method for quality grading of recycled aluminum based on image recognition according to claim 4, characterized in that: An installation groove is provided on the inner wall of the protective cover at the location corresponding to the protective film, and the protective film and the surface of the protective lens are in sliding contact. A recycling roller is installed inside the installation groove, and the recycling roller is connected to the output end of the drive device provided on the inner wall of the installation groove.

6. The method for quality grading of recycled aluminum based on image recognition according to claim 5, characterized in that: The industrial camera is equipped with a cooling fan inside and a cooling groove on the outer surface of the camera housing. The gap between the inner wall of the protective cover and the acquisition end of the industrial camera is connected to the cooling groove. A connecting hole is provided in the middle of the protective lens, and the connecting hole is connected to the gap between the protective lens and the protective film.

7. The method for quality grading of recycled aluminum based on image recognition according to claim 6, characterized in that: The vertical cross-section of the mounting groove is L-shaped. After the protective film extends into the mounting groove along the vertical part, it passes around the guide roller set at the turning part of the mounting groove, and the movement trajectory changes to the horizontal part. It is then collected on the recovery roller along the horizontal part of the mounting groove. A guide groove is set on the inner wall of the mounting groove below the guide roller, and the opening of the guide groove points to the outer surface of the protective film in the observation port.

8. The method for quality grading of recycled aluminum based on image recognition according to claim 7, characterized in that: Guide blocks are evenly arranged on the lower top surface of the horizontal part of the installation groove. The side of the guide block near the recovery roller is a vertical surface, and the side away from the recovery roller is a horizontal surface.

9. A quality grading system for recycled aluminum based on image recognition, wherein the quality grading system for recycled aluminum is applicable to the quality grading method for recycled aluminum described in any one of claims 1-8, characterized in that, The recycled aluminum quality grading system includes a feeding module, a preprocessing module, an image acquisition module, an image processing and feature extraction module, and a grading module.

Images