A waste lithium battery classification processing method and system based on image segmentation recognition

Abstract

The application discloses a kind of based on image segmentation identification waste lithium battery classification processing method and system, it is related to waste lithium battery recycling technical field, including, configure high-resolution camera, lighting equipment and ambient light sensor, real-time collection waste lithium battery multiple view images, obtain two-dimensional image dataset, two-dimensional image dataset is input in SIFT algorithm and is detected and matched to feature point, based on matched feature point, three-dimensional model is generated by MVS technology, and the geometric information of battery surface is obtained using triangulation method;The application generates high-quality three-dimensional model by using high-resolution multiple view image acquisition combined with SIFT feature point matching and MVS technology, and the quality of model is optimized by denoising, curvature enhancement and dynamic parameter adjustment, to ensure the accurate modeling of complex geometry. The optimized three-dimensional model is analyzed using deep convolutional neural network, which greatly improves the robustness and accuracy of defect detection.

Description

Technical Field

[0001] This invention relates to the field of waste lithium battery recycling technology, and in particular to a waste lithium battery classification and processing method and system based on image segmentation and recognition. Background Technology

[0002] With the development of technology, the demand for energy recycling is gradually increasing, and the recycling of spent lithium batteries has become a hot research area. Traditional recycling relies on manual inspection, which is not only inefficient but also suffers from insufficient detection accuracy, easily leading to misjudgments. In recent years, more and more research has focused on improving the classification accuracy of spent lithium batteries. By combining multi-view imaging and 3D reconstruction technology, detailed battery surface information can be provided without damaging the battery surface structure, laying the foundation for defect analysis and subsequent classification processing.

[0003] However, existing methods for classifying and processing waste lithium batteries still have some shortcomings. On the one hand, they have limited ability to accurately model complex geometries, resulting in low quality of 3D models. On the other hand, existing defect detection algorithms lack sufficient robustness and accuracy when dealing with different types of defects, which may lead to the omission or misjudgment of key defects, affecting the final classification results. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a waste lithium battery classification and processing method based on image segmentation and recognition to solve the problems of poor 3D model quality and insufficient accuracy of defect detection.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a waste lithium battery classification and processing method based on image segmentation and recognition, which includes configuring a high-resolution camera, lighting equipment and an ambient light sensor to acquire multi-view images of waste lithium batteries in real time and obtain a two-dimensional image dataset.

[0008] The two-dimensional image dataset is input into the SIFT algorithm to detect and match feature points. Based on the matched feature points, a three-dimensional model is generated using the MVS technique, and the geometric information of the battery surface is obtained using the triangulation method.

[0009] The 3D model is denoised and its curvature is enhanced, and the parameter settings are dynamically adjusted to obtain the optimized 3D model.

[0010] The optimized 3D model is analyzed using a deep convolutional neural network to identify and locate defects on the battery, and to obtain the location, type, and severity score of the defects.

[0011] By integrating the location, type, and severity scores of defects with the geometric information of the battery surface, the battery is evaluated to obtain battery category information and real-time status.

[0012] The optimized 3D model parameters, battery category information, and real-time status are integrated to generate a classification report, and the batteries are processed according to category based on the classification report.

[0013] As a preferred embodiment of the image segmentation and recognition-based waste lithium battery classification and processing method of the present invention, the step of inputting a two-dimensional image dataset into the SIFT algorithm to detect and match feature points, and generating a three-dimensional model based on the matched feature points using MVS technology, specifically includes the following steps.

[0014] The two-dimensional image dataset is input into the SIFT algorithm to extract multiple scale-invariant feature points. The interior point pairs are then filtered out by the FLANN matcher to establish cross-view geometric correspondence and obtain the matching feature points.

[0015] Based on the matched feature points, the depth information of each pixel is calculated using COLMAP in MVS to generate a sparse depth map. The point cloud density is then optimized by point cloud filtering, and finally a 3D model is constructed using the Marching Cubes algorithm.

[0016] As a preferred embodiment of the waste lithium battery classification and processing method based on image segmentation and recognition described in this invention, the geometric information of the battery surface is obtained by reconstructing a three-dimensional point cloud from multi-view images based on triangulation, filtering and optimizing the density of the three-dimensional point cloud, and then performing normal vector and curvature calculation and surface deformation feature analysis.

[0017] As a preferred embodiment of the image segmentation and recognition-based waste lithium battery classification and processing method of the present invention, the step of dynamically adjusting parameter settings to obtain the optimized three-dimensional model specifically includes the following steps.

[0018] Initial parameters are set based on the initial point cloud density and curvature distribution of the 3D model, and the effect of the 3D model under the initial parameters is displayed in real time through the voxel rasterization algorithm.

[0019] Based on the effect of the 3D model, the initial parameters are dynamically fine-tuned by reducing the denoising intensity and increasing the curvature sensitivity. The parameters are further corrected by combining the effect of the 3D model after parameter adjustment. At the same time, a regional differentiation strategy is adopted to automatically assign different parameter weights to high curvature regions and low curvature regions.

[0020] Repeat the parameter adjustment process until the 3D model meets the accuracy index, then lock the final parameters and save the optimized 3D model.

[0021] The accuracy metric is defined based on the geometric difference between the 3D model and the actual battery surface.

[0022] As a preferred embodiment of the image segmentation and recognition-based waste lithium battery classification and processing method of the present invention, the step of using a deep convolutional neural network to analyze the optimized 3D model, identify and locate defects on the battery, and obtain the location, type, and severity score of the defects specifically includes the following steps.

[0023] The optimized 3D model is converted into 2D projection data through multi-view projection, and then input into a deep convolutional neural network to extract local curvature features and global shape features. The defect area is then located by combining the RPN.

[0024] Based on the defect region, the Faster R-CNN framework's multi-task learning branch is used to fuse local curvature features with global shape features and jointly optimize classification and regression tasks to obtain the defect's bounding box coordinates, type, and severity score.

[0025] As a preferred embodiment of the image segmentation and recognition-based waste lithium battery classification and processing method of the present invention, the step of integrating the location, type, and severity scores of defects with the geometric information of the battery surface to evaluate the battery and obtain the battery's category information and real-time status specifically includes the following steps:

[0026] The location, type, and severity score of defects are mapped back to three-dimensional space through coordinate transformation with the geometric information of the battery surface. The point cloud spherical hash mapping is used to associate them into the same data structure, and combined with KD-Tree spatial indexing to generate integrated information.

[0027] Based on the integrated information, the surface defect patterns of the battery are evaluated using a convolutional neural network to obtain the battery category information. The dynamic charging and discharging behavior of the battery is evaluated using a lightweight temporal autoencoder to obtain the real-time status of the battery.

[0028] As a preferred embodiment of the waste lithium battery classification and processing method based on image segmentation and recognition described in this invention, the classification report is generated by multi-dimensional data aggregation and document automation tools based on integrated three-dimensional model parameters, category information and real-time status. Based on the risk level labels in the classification report, an automated decision tree is used to trigger the classification and disposal process of waste lithium batteries.

[0029] Secondly, the present invention provides a waste lithium battery classification and processing system based on image segmentation and recognition, including an image acquisition module, a three-dimensional reconstruction module, a three-dimensional model optimization module, a defect detection and localization module, a comprehensive evaluation module, and a classification processing module.

[0030] The image acquisition module is used to configure a high-resolution camera, lighting equipment, and ambient light sensor to acquire multi-view images of waste lithium batteries in real time and obtain a two-dimensional image dataset.

[0031] The 3D reconstruction module is used to input the 2D image dataset into the SIFT algorithm to detect and match feature points. Based on the matched feature points, a 3D model is generated through MVS technology, and the geometric information of the battery surface is obtained using triangulation.

[0032] The 3D model optimization module is used to perform noise reduction and curvature enhancement on 3D models, and dynamically adjust parameter settings to obtain optimized 3D models.

[0033] The defect detection and localization module is used to analyze the optimized 3D model using a deep convolutional neural network, identify and locate defects on the battery, and obtain the location, type and severity score of the defects.

[0034] The comprehensive evaluation module integrates the location, type, and severity scores of defects with the geometric information of the battery surface to evaluate the battery and obtain battery category information and real-time status.

[0035] The classification processing module integrates the optimized 3D model parameters, battery category information, and real-time status to generate a classification report, and processes the batteries according to their categories based on the classification report.

[0036] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the waste lithium battery classification and processing method based on image segmentation and recognition as described in the first aspect of the present invention.

[0037] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the waste lithium battery classification and processing method based on image segmentation and recognition as described in the first aspect of the present invention.

[0038] The beneficial effects of this invention are as follows: High-resolution multi-view image acquisition combined with SIFT feature point matching and MVS technology generates high-quality 3D models. The model quality is optimized through denoising, curvature enhancement, and dynamic parameter adjustment, ensuring accurate modeling of complex geometric shapes. Secondly, a deep convolutional neural network is used to analyze the optimized 3D model. The RPN and Faster R-CNN frameworks are combined to achieve the fusion of local curvature features and global shape features, as well as the joint optimization of classification and regression tasks, greatly improving the robustness and accuracy of defect detection. It can not only accurately locate and classify various types of defects but also accurately assess their severity, thus providing strong support for the accurate classification and efficient recycling of waste lithium batteries. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart of the waste lithium battery classification and processing method based on image segmentation and recognition in Example 1.

[0041] Figure 2 This is a schematic diagram of the waste lithium battery sorting and processing system based on image segmentation and recognition in Example 1.

[0042] Figure 3 This is a flowchart of the waste lithium battery sorting and processing method in Example 1.

[0043] Figure 4 This is a flowchart of the waste lithium battery sorting and processing system structure in Example 1. Detailed Implementation

[0044] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0045] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0046] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0047] Example 1, referring to Figures 1-4 This is the first embodiment of the present invention, which provides a method for classifying and processing waste lithium batteries based on image segmentation and recognition, including the following steps:

[0048] S1. Equipped with a high-resolution camera, lighting equipment, and ambient light sensor, it acquires multi-view images of used lithium batteries in real time to obtain a two-dimensional image dataset.

[0049] Specifically, the operations include the following:

[0050] To achieve multi-view image acquisition of waste lithium batteries, detailed hardware configuration and calibration are required first. High-resolution industrial cameras are placed in different positions and angles to ensure that all surfaces of the battery can be covered. Each camera is equipped with independently controlled LED lighting equipment, which ensures that clear, shadow-free images can be obtained even under changing ambient light conditions. At the same time, the ambient light sensor monitors the ambient light conditions in real time and automatically adjusts the brightness and color temperature of the lighting equipment according to the ambient light information, thereby maintaining the consistency of image quality.

[0051] Next, precise camera calibration and synchronized data acquisition are performed. Each camera is calibrated using a standard checkerboard pattern to determine its focal length, principal point, and position and orientation relative to the battery. This step is crucial because it directly affects the accuracy of subsequent 3D reconstruction. All cameras must work synchronously, using global shutter technology to capture images and avoid blurring caused by motion. Synchronization can be achieved through GPIO signals and Network Time Protocol (NTP) to ensure that all cameras capture images at the same time and guarantee the consistency of image data. During each shot, preliminary preprocessing such as noise reduction and contrast enhancement is performed through the camera's built-in ISP to improve image quality.

[0052] After completing multi-view image acquisition, the acquired two-dimensional image data is transmitted to the central processing unit in real time via a high-speed network interface. To ensure the integrity and security of the two-dimensional image data, redundant array disk technology is used for two-dimensional image data storage to prevent data loss due to hardware failure. In addition, HDFS can be used to efficiently manage large-scale two-dimensional image data, and a dedicated data management framework has been developed to support image data indexing, retrieval, and version control, making it convenient for technicians to view and manage the acquired two-dimensional image dataset.

[0053] S2. Input the two-dimensional image dataset into the SIFT algorithm to detect and match feature points. Based on the matched feature points, generate a three-dimensional model using MVS technology and obtain the geometric information of the battery surface using triangulation.

[0054] Specifically, the operations include the following:

[0055] In the initial stage of waste lithium battery sorting and processing, it is necessary to process the acquired multi-view two-dimensional images. The core of this process is to use the SIFT (Scale Invariant Feature Transform) algorithm to construct Gaussian pyramids at different scales and to find local extrema as feature points through difference calculation. First, the original two-dimensional image is continuously scaled and Gaussian filters are applied to generate a series of smoothed images to form a Gaussian pyramid. The difference calculation between adjacent layers forms a difference pyramid, which is used to detect local maxima and minima as candidate feature points. This process can not only capture the tiny details on the battery surface, but also effectively avoid mismatch problems caused by changes in lighting or texture differences.

[0056] To ensure the stability and uniqueness of feature points, further removal of low-contrast points and edge response filtering are crucial. Taylor expansion is used to approximate the precise location and response value of each candidate feature point, eliminating low-contrast points. Simultaneously, the principal curvature ratio is calculated using the Hessian matrix to remove unstable points located on edges. Next, one or more principal directions are assigned to each feature point, and the peak value is selected as the principal direction by calculating the gradient direction histogram to ensure rotation invariance. This step ensures that feature points remain consistent even when the image is rotated. Finally, the FLANN (Fast Library for Approximate Nearest Neighbors) matcher is used to match these feature points, filtering out inlier pairs and establishing geometric correspondences across views, thereby obtaining matched feature points. The FLANN algorithm significantly improves matching efficiency and accuracy through fast library approximate nearest neighbor search.

[0057] Based on the obtained matching feature points, the COLMAP software in the MVS (Multi-View Stereo) technology is used to calculate the depth information of each pixel and generate a sparse depth map. COLMAP uses bundle adjustment technology to jointly optimize the camera pose and the position of 3D points, further improving the reconstruction accuracy. This process greatly improves the accuracy of 3D reconstruction, especially in capturing details of complex shapes such as battery surfaces. Bundle adjustment technology ensures geometric consistency by globally optimizing all camera parameters and the position of 3D points.

[0058] After generating a sparse depth map, it needs to be filtered and density optimized. Common filtering methods include statistical filtering and voxel meshing, which are used to remove noise points and increase point cloud density. Statistical filtering removes isolated noise points by analyzing the number of neighboring points around each point. Voxel meshing divides the point cloud into uniform small cubic units and interpolates the points in each unit to fill in missing points, thereby improving the overall density of the point cloud. Subsequently, the Marching Cubes algorithm is used to transform the optimized point cloud into a triangular mesh. By traversing the voxel space to identify cubic units containing surfaces and generating corresponding triangular meshes, a complete 3D model is finally constructed. The Marching Cubes algorithm can not only accurately represent complex surface morphologies, but also generate smooth and coherent 3D models, making the 3D models both beautiful and practical.

[0059] In addition, Laplacian smoothing is applied to reduce the number of polygons while maintaining visual quality. Laplacian smoothing reduces sharp edges in the 3D model by weighted averaging of the position of each vertex, enhancing visual detail. This is crucial for subsequent analysis and application. The entire 3D model generation process not only improves the reliability and accuracy of the processing results, but also provides strong technical support for subsequent defect detection, classification tasks, and the recycling of waste lithium batteries.

[0060] After the 3D model is constructed, the geometric information of the battery surface is further extracted. First, the 3D point cloud is reconstructed from multi-view images using the triangulation method. Then, the ICP (Iterative Closest Point) algorithm is used to register the point clouds from different viewpoints to ensure seamless connection between the point clouds. The ICP algorithm optimizes the alignment accuracy of the point clouds by minimizing the distance error between them. At the same time, the 3D point cloud is filtered to remove noise. Voxel meshing and Poisson reconstruction techniques are used to optimize the point cloud density and smoothness, making the final 3D model more accurate and consistent.

[0061] Next, the normal vector and curvature of each point in the 3D point cloud are calculated. These attributes are crucial for understanding the shape and structure of the battery surface. The normal vector provides directional information about the surface, while the curvature reflects the degree of bending of the surface. By combining the normal vector and curvature information, surface deformation feature analysis is performed to identify possible uneven or deformed areas. For example, a curvature abrupt change threshold is defined based on the significance of curvature changes in surface geometry. By analyzing the joint features of the curvature abrupt change threshold and surface reflection intensity, defect types such as corrosion and cracks can be accurately located. This in-depth surface deformation feature analysis, combined with convolutional neural networks, can efficiently classify and identify surface anomalies, providing a scientific basis for the safety assessment and recycling of waste lithium batteries.

[0062] A precise understanding of surface morphology can significantly improve the efficiency and reliability of automated processing. For example, in robotic arm operations, accurate 3D models and surface information can help the robotic arm more accurately grasp and process waste batteries, avoiding damage and omissions. This highly efficient automated processing not only improves work efficiency but also reduces the need for human intervention and lowers costs.

[0063] S3. Perform noise reduction and curvature enhancement operations on the 3D model, and dynamically adjust the parameter settings to obtain the optimized 3D model.

[0064] Specifically, the operations include the following:

[0065] Initial parameters are set based on the initial point cloud density and curvature distribution of the 3D model. These initial parameters include denoising intensity and curvature sensitivity, which directly affect the quality of the 3D model. In this stage, the original point cloud is converted into a voxel representation using a voxel rasterization algorithm. Voxel rasterization is a method of dividing 3D space into uniform voxels. The average value or interpolation is calculated in each voxel to generate a visual mesh model. This method can not only quickly show the general shape of the 3D model, but also help evaluate the effectiveness of the initial parameters. For example, for high-density point cloud areas, smaller voxel sizes can be used to capture details, while for low-density areas, larger voxel sizes can be used to reduce computational complexity. In this way, a relatively clear 3D model view can be obtained in the initial stage, and the effect of the 3D model under the initial parameters can be displayed in real time, providing an intuitive basis for further parameter adjustments.

[0066] Based on the real-time display effect, the parameters are dynamically fine-tuned. In specific operations, the changes in the 3D model are observed by gradually reducing the denoising intensity and increasing the curvature sensitivity. The parameters are then further corrected based on the model effect after parameter adjustment. This iterative adjustment process aims to find the optimal parameter combination so that the model can effectively remove noise and accurately capture the detailed features of the battery surface. In this process, the use of a regional differentiation strategy is particularly important. Since the surface of waste lithium batteries has various different geometric characteristics, such as edges, sharp parts, and flat areas, different parameter weights need to be assigned to different regions. For example, in high curvature regions, such as the root of the battery tabs or indentations, curvature sensitivity is increased to better capture subtle changes; while in low curvature regions, such as large flat surfaces, the denoising intensity is appropriately reduced to retain more detailed information. This ensures that the 3D model can achieve the best performance in different regions and improve the overall quality. In addition, adaptive algorithms can be used to automatically identify high and low curvature regions and dynamically adjust parameters according to actual needs to further improve the robustness and adaptability of the 3D model.

[0067] After multiple parameter adjustments, the 3D model gradually approaches its optimal state. The next step is to verify whether it meets the predefined accuracy indicators. These accuracy indicators are defined based on the geometric differences between the 3D model and the actual battery surface, and typically include quantitative indicators such as surface error and shape deviation. To accurately evaluate the accuracy of the 3D model, various methods can be used for verification. For example, a laser scanner or other high-precision measuring equipment can be used to acquire the actual geometric data of the battery surface. Then, the geometric data can be compared and analyzed with the 3D model. Common evaluation indicators include root mean square error and maximum absolute error. By calculating and analyzing these indicators, the degree of difference between the 3D model and the actual battery surface can be fully understood. If all the indicators of the 3D model meet the requirements, the current parameter settings are locked, and the optimized 3D model is saved. This process not only ensures the high quality of the 3D model but also provides a solid foundation for subsequent defect detection and analysis.

[0068] Through the above operations, a complete process from initial parameter setting to final 3D model optimization was achieved. This not only improved the reliability and accuracy of the processing results, but also provided high-quality data support for subsequent deep convolutional neural network analysis, significantly improving the efficiency and effectiveness of waste lithium battery recycling and reuse.

[0069] S4. Analyze the optimized 3D model using a deep convolutional neural network to identify and locate defects on the battery, and obtain the location, type, and severity score of the defects.

[0070] Specifically, the operations include the following:

[0071] Before starting the analysis using a deep convolutional neural network, the optimized 3D model needs to be converted into 2D projection data through multi-view projection. In the specific operation, multiple viewpoints are selected, such as top view, side view, etc., and the 3D model is rendered from these viewpoints to generate corresponding 2D images. The 2D image under each viewpoint not only contains the basic information of the battery surface, but also retains the key geometric features of the original 3D model. In order to further enhance the quality of the 2D image, some preprocessing techniques can be applied, such as contrast adjustment and edge enhancement, to ensure that the data input to the deep convolutional neural network has high clarity and detail.

[0072] Two-dimensional projection data is input into a convolutional neural network model for feature extraction. Pre-trained convolutional neural network architectures, such as ResNet and VGG, are typically used here. These convolutional neural network models are fully trained on large-scale image datasets and can effectively capture complex features in images. Through forward propagation, the convolutional neural network model automatically extracts multi-level feature representations from the two-dimensional image, including low-level edge and texture information and high-level semantic features. Then, the RPN (Region Proposal Network) is used to initially locate potential defective regions. The RPN scans features through a sliding window mechanism to generate a series of candidate regions, thereby identifying defective regions. The design of the RPN reduces false positives while maintaining high recall, providing a reliable basis for subsequent accurate battery classification.

[0073] Using the Faster R-CNN framework, a multi-task learning branch is set up in the convolutional neural network model, mainly including the fusion of local curvature features and global shape features, as well as the joint optimization of classification and regression tasks. First, for each defect region generated by RPN, its internal local features are further extracted. Specifically, RoI pooling layers can be used to unify defect regions of different sizes to a fixed size for easier subsequent processing. On this basis, local curvature features and global shape features are combined to form a richer feature representation. Local curvature features can be calculated from the 3D model and reflect the curvature of the surface in a specific defect region, while global shape features describe the general shape of the entire battery surface. The fusion of local curvature features and global shape features helps to improve the ability to identify complex defects.

[0074] Next, in the joint optimization phase of the classification and regression tasks, the Faster R-CNN framework performs two tasks simultaneously: first, it classifies each defect region to determine whether it belongs to a certain type of defect; second, it regresses the bounding boxes to accurately adjust the position and size of the candidate boxes to better match the actual defect location. The classification task typically uses the Softmax function to output the probability distribution of each defect category, while the regression task optimizes the parameters by minimizing the distance error between the predicted box and the ground truth box. This multi-task learning approach not only improves detection accuracy but also effectively reduces complexity and computational burden. Finally, the trained and optimized convolutional neural network model based on the Faster R-CNN framework can output the bounding box coordinates, type, and severity score of each defect region, providing an accurate basis for subsequent battery evaluation.

[0075] Through the above operations, it is ensured that the optimized 3D model can accurately identify and locate various defects on the battery and obtain detailed defect information.

[0076] S5. Integrate the location, type, and severity scores of defects with the geometric information of the battery surface to evaluate the battery and obtain battery category information and real-time status.

[0077] Specifically, the operations include the following:

[0078] The defect location, type, and severity score detected in the 2D projection are converted back to their corresponding locations in 3D space. In practice, using the camera parameters and viewpoint information recorded during the previous multi-view projection, each defect location is mapped from the 2D image coordinate system back to the 3D model coordinate system through inverse perspective transformation. This process ensures that all defect information can be accurately reflected on the original 3D model. Next, point cloud spherical hash mapping technology is used to associate this defect information with the geometric information of the battery surface. Point cloud spherical hash mapping is an efficient point cloud indexing method. It achieves fast lookup and association by dividing the space into multiple spherical regions and assigning a unique hash value to each spherical region. It can efficiently integrate defect information and point cloud data of the battery surface into the same data structure.

[0079] To further improve data query efficiency and processing speed, KD-Tree (K-Dimensional Tree) spatial indexing technology is adopted. KD-Tree is a tree structure used to organize multidimensional data, which is particularly suitable for nearest neighbor search in high-dimensional space. By constructing KD-Tree, the point cloud data closest to a certain defect location can be retrieved quickly, thereby more accurately associating the defect with the specific location of the battery surface. When generating integrated defect location information, the data after coordinate transformation and point cloud spherical hash mapping is input into KD-Tree to form a unified data structure. This data structure not only contains the location, type and severity score of all defects, but also combines the geometric features of the battery surface, such as curvature and normal vector, providing a comprehensive information foundation for subsequent evaluation.

[0080] Based on the integrated defect location information, the surface defect pattern of the battery is evaluated by a convolutional neural network. In the specific operation, the integrated information is input into the convolutional neural network model, which automatically extracts multi-level feature representations and uses these features to perform classification tasks, determine the category corresponding to each defect area, such as corrosion, cracks, etc. Finally, the convolutional neural network model outputs the category information of the battery, that is, it determines which type of problem the battery belongs to based on the surface defect pattern, such as whether it needs to be repaired or scrapped.

[0081] In addition to evaluating surface defect patterns, it is also necessary to analyze the dynamic charge-discharge behavior of the battery to obtain its real-time status. A lightweight time-series autoencoder (LSTM) is used for evaluation. LSTM is a recurrent neural network specifically designed for processing sequential data, making it ideal for capturing long-term dependencies in time series. Specifically, data such as voltage, current, and temperature of the battery during different charge-discharge cycles are collected to form a time series dataset. This time series data is then input into the LSTM autoencoder for training. The LSTM autoencoder consists of an encoder and a decoder. The encoder compresses the input time series into a low-dimensional representation, while the decoder attempts to reconstruct the original time series from this low-dimensional representation. By minimizing the reconstruction error, the LSTM model can learn effective feature representations of the battery's charge-discharge behavior. Finally, the trained LSTM autoencoder is used to predict new time series data, assess the battery's current health status, and provide a health score. The mathematical expression is:

[0082] ;

[0083] in, Indicates health score, Indicates the number of time series segments. Indices representing time series segments Indicates the first The weights of each time series segment Indicates the first Reconstruction error of a time series segment This represents the mean of all reconstruction errors. Indicates the number of battery features. An index representing battery characteristics. Indicates the first State score of each battery feature;

[0084] Health ratings help identify potential functional problems in batteries, such as capacity degradation and increased internal resistance.

[0085] Through the above operations, defect information is efficiently integrated with battery surface geometric data, and a comprehensive evaluation is performed using convolutional neural networks and lightweight temporal autoencoders, which significantly improves the accuracy and efficiency of waste lithium battery classification and state assessment.

[0086] S6. Integrate the optimized 3D model parameters, battery category information, and real-time status to generate a classification report, and process the batteries according to their categories based on the classification report.

[0087] Specifically, the operations include the following:

[0088] To achieve efficient classification and processing of waste lithium batteries, the first step is to integrate optimized 3D model parameters, battery category information, and real-time status to generate a detailed classification report. This process begins with multidimensional data aggregation, using ETL tools to extract relevant data from various data sources, including 3D model parameters, defect location and type, severity scores, and battery surface geometry. After cleaning and transformation, this data is loaded into a comprehensive database to ensure seamless integration of data from different sources. Next, document automation tools are used to automatically generate classification reports based on predefined templates. These templates typically contain standard-formatted fields to display various assessment results, such as defect location, health score, and risk level. This not only improves the efficiency of report generation but also ensures consistency and accuracy in format.

[0089] In the classification report, each battery is assigned a risk level label based on its defect type, health score, and real-time status assessment. To determine this risk level label, a scoring system is designed, assigning weights to each factor and calculating the total score to determine the final risk level. For example, the risk level can be determined by comprehensively considering factors such as the severity and location of the defect and the battery's health status. Once the risk level is determined, an automated decision tree can be used to trigger the corresponding processing flow. An automated decision tree is a rule-based decision support strategy that can automatically select the appropriate operation path based on the input conditions. For example, for batteries with a high risk level, the decision tree may suggest direct scrapping, while for batteries with a low risk level, it may recommend further inspection or simple repair.

[0090] Based on the risk level labels in the classification report, the automated decision tree will initiate the corresponding waste lithium battery classification and disposal process. First, a series of rules are configured in the decision tree strategy. These rules define the treatment measures corresponding to different risk levels. For example, the rule is set as "If the risk level is high, trigger the scrapping process; if the risk level is medium, enter the repair process; if the risk level is low, mark it as usable". After the classification report is generated and the risk level is marked, the automated decision tree will automatically trigger the corresponding processing process according to the preset rules. This can be achieved through ERP to ensure that the processing instructions can be quickly transmitted to relevant departments and equipment. During the battery classification and processing, monitoring equipment is configured to track the processing progress of each battery in real time and record the processing results. In addition, a feedback mechanism can be set up to collect the actual processing effect in order to continuously optimize the rules and processes in the decision tree.

[0091] The above operations generate detailed classification reports, and based on risk level labels, an automated decision tree triggers corresponding processing procedures, significantly improving the efficiency and accuracy of waste lithium battery classification and processing. This not only reduces the possibility of human error but also enhances the effectiveness of resource recycling, providing strong support for achieving efficient management and environmentally friendly treatment.

[0092] This embodiment also provides a waste lithium battery classification and processing system based on image segmentation and recognition, including: an image acquisition module, a three-dimensional reconstruction module, a three-dimensional model optimization module, a defect detection and localization module, a comprehensive evaluation module, and a classification and processing module;

[0093] The image acquisition module is used to configure a high-resolution camera, lighting equipment, and ambient light sensor to acquire multi-view images of waste lithium batteries in real time and obtain a two-dimensional image dataset.

[0094] The 3D reconstruction module is used to input the 2D image dataset into the SIFT algorithm to detect and match feature points. Based on the matched feature points, a 3D model is generated through MVS technology, and the geometric information of the battery surface is obtained using triangulation.

[0095] The 3D model optimization module is used to perform noise reduction and curvature enhancement on 3D models, and dynamically adjust parameter settings to obtain optimized 3D models.

[0096] The defect detection and localization module is used to analyze the optimized 3D model using a deep convolutional neural network, identify and locate defects on the battery, and obtain the location, type and severity score of the defects.

[0097] The comprehensive evaluation module integrates the location, type, and severity scores of defects with the geometric information of the battery surface to evaluate the battery and obtain battery category information and real-time status.

[0098] The classification processing module integrates the optimized 3D model parameters, battery category information, and real-time status to generate a classification report, and processes the batteries according to their categories based on the classification report.

[0099] This embodiment also provides a computer device applicable to the waste lithium battery classification and processing method based on image segmentation and recognition, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the waste lithium battery classification and processing method based on image segmentation and recognition proposed in the above embodiment.

[0100] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0101] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the image segmentation and recognition-based waste lithium battery classification and processing method proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0102] In summary, this invention generates high-quality 3D models by employing high-resolution multi-view image acquisition combined with SIFT feature point matching and MVS technology. The model quality is further optimized through denoising, curvature enhancement, and dynamic parameter adjustment, ensuring accurate modeling of complex geometries. Secondly, a deep convolutional neural network is used to analyze the optimized 3D model. The RPN and Faster R-CNN frameworks are combined to achieve the fusion of local curvature features and global shape features, as well as joint optimization of classification and regression tasks, significantly improving the robustness and accuracy of defect detection. This not only accurately locates and classifies various types of defects but also accurately assesses their severity, thus providing strong support for the accurate classification and efficient recycling of waste lithium batteries.

[0103] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for classifying and processing waste lithium batteries based on image segmentation and recognition, characterized in that: include, Equipped with a high-resolution camera, lighting equipment, and ambient light sensor, it acquires multi-view images of waste lithium batteries in real time and obtains a two-dimensional image dataset. The two-dimensional image dataset is input into the SIFT algorithm to detect and match feature points. Based on the matched feature points, a three-dimensional model is generated using the MVS technique, and the geometric information of the battery surface is obtained using triangulation. The 3D model is denoised and its curvature is enhanced, and the parameter settings are dynamically adjusted to obtain the optimized 3D model. The optimized 3D model is analyzed using a deep convolutional neural network to identify and locate defects on the battery, and to obtain the location, type, and severity scores of the defects. The specific steps include the following: The optimized 3D model is converted into 2D projection data through multi-view projection, and then input into a deep convolutional neural network to extract local curvature features and global shape features. The defect area is then located by combining the RPN. Based on the defect region, the Faster R-CNN framework is used to fuse local curvature features and global shape features, and jointly optimize classification and regression tasks to obtain the bounding box coordinates, type and severity score of the defect. By integrating the location, type, and severity scores of defects with the geometric information of the battery surface, the battery is evaluated to obtain battery category information and real-time status. The optimized 3D model parameters, battery category information, and real-time status are integrated to generate a classification report, and the batteries are processed according to their categories based on the classification report.

2. The waste lithium battery classification and processing method based on image segmentation and recognition as described in claim 1, characterized in that: The process of inputting a two-dimensional image dataset into the SIFT algorithm for feature point detection and matching, and then generating a three-dimensional model based on the matched feature points using MVS technology, specifically includes the following steps. The two-dimensional image dataset is input into the SIFT algorithm to extract multiple scale-invariant feature points. The interior point pairs are then filtered out by the FLANN matcher to establish cross-view geometric correspondence and obtain the matching feature points. Based on the matched feature points, the depth information of each pixel is calculated using COLMAP in MVS to generate a sparse depth map. The point cloud density is then optimized by point cloud filtering, and finally a 3D model is constructed using the Marching Cubes algorithm.

3. The waste lithium battery classification and processing method based on image segmentation and recognition as described in claim 1, characterized in that: The geometric information of the battery surface is obtained by reconstructing a three-dimensional point cloud from multi-view images based on triangulation, filtering and optimizing the density of the three-dimensional point cloud, and then performing normal vector and curvature calculations and surface deformation feature analysis.

4. The waste lithium battery classification and processing method based on image segmentation and recognition as described in claim 3, characterized in that: The process of dynamically adjusting parameter settings to obtain the optimized 3D model includes the following steps. Initial parameters are set based on the initial point cloud density and curvature distribution of the 3D model, and the effect of the 3D model under the initial parameters is displayed in real time through the voxel rasterization algorithm. Based on the effect of the 3D model, the initial parameters are dynamically fine-tuned by reducing the denoising intensity and increasing the curvature sensitivity. The parameters are further corrected by combining the effect of the 3D model after parameter adjustment. At the same time, a regional differentiation strategy is adopted to automatically assign different parameter weights to high curvature regions and low curvature regions. Repeat the parameter adjustment process until the 3D model meets the accuracy index, then lock the final parameters and save the optimized 3D model. The accuracy metric is defined based on the geometric difference between the 3D model and the actual battery surface.

5. The waste lithium battery classification and processing method based on image segmentation and recognition as described in claim 4, characterized in that: The process of integrating defect location, type, and severity scores with battery surface geometry to evaluate the battery and obtain its category information and real-time status includes the following steps: The location, type, and severity score of defects are mapped back to three-dimensional space through coordinate transformation with the geometric information of the battery surface. The point cloud spherical hash mapping is used to associate them into the same data structure, and combined with KD-Tree spatial indexing to generate integrated information. Based on the integrated information, the surface defect patterns of the battery are evaluated using a convolutional neural network to obtain the battery category information. The dynamic charging and discharging behavior of the battery is evaluated using a lightweight temporal autoencoder to obtain the real-time status of the battery.

6. The waste lithium battery classification and processing method based on image segmentation and recognition as described in claim 5, characterized in that: The process of classifying batteries according to categories based on the classification report includes generating a classification report through multi-dimensional data aggregation and document automation tools based on the integrated 3D model parameters, category information and real-time status, and triggering the classification and disposal process of waste lithium batteries using an automated decision tree based on the risk level labels in the classification report.

7. A waste lithium battery sorting and processing system based on image segmentation and recognition, based on the waste lithium battery sorting and processing method based on image segmentation and recognition according to any one of claims 1 to 6, characterized in that: It includes an image acquisition module, a 3D reconstruction module, a 3D model optimization module, a defect detection and localization module, a comprehensive evaluation module, and a classification processing module; The image acquisition module is used to configure a high-resolution camera, lighting equipment, and ambient light sensor to acquire multi-view images of waste lithium batteries in real time and obtain a two-dimensional image dataset. The 3D reconstruction module is used to input the 2D image dataset into the SIFT algorithm to detect and match feature points. Based on the matched feature points, a 3D model is generated through MVS technology, and the geometric information of the battery surface is obtained using triangulation. The 3D model optimization module is used to perform noise reduction and curvature enhancement on 3D models, and dynamically adjust parameter settings to obtain optimized 3D models. The defect detection and localization module is used to analyze the optimized 3D model using a deep convolutional neural network, identify and locate defects on the battery, and obtain the location, type and severity score of the defects. The comprehensive evaluation module integrates the location, type, and severity scores of defects with the geometric information of the battery surface to evaluate the battery and obtain battery category information and real-time status. The classification processing module integrates the optimized 3D model parameters, battery category information, and real-time status to generate a classification report, and processes the batteries according to their categories based on the classification report.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the waste lithium battery classification and processing method based on image segmentation and recognition as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the waste lithium battery classification and processing method based on image segmentation and recognition as described in any one of claims 1 to 6.

Images