A method and system for detecting defects in an aluminum foil tape

Abstract

The application discloses an aluminum foil strip defect detection method and system, which comprises the following steps: acquiring multimodal data collected by optical sensors and ultrasonic sensors in a high-speed production line; performing mirror reflection preprocessing on optical images, adopting an adaptive filter to eliminate environmental light interference, and obtaining a first feature set; weighting and fusing the first feature set and ultrasonic signals according to a preset weight matrix, and obtaining a second feature set; judging whether the second feature set contains defect features, and if the second feature set contains defect features, classifying the fused features through a convolutional neural network; adopting a support vector machine to perform generalization processing on the defect classification results, and acquiring a generalization model; connecting the generalization model with a production line control system, generating a control instruction according to the defect classification results, and adjusting production line speed parameters when defects are detected. The application improves the accuracy and robustness of defect detection, realizes intelligent defect detection and control of a high-speed production line, and effectively improves product quality and production efficiency.

Description

Technical Field

[0001] This invention relates to the field of aluminum foil strip inspection technology, and in particular discloses a method and system for detecting defects in aluminum foil strip. Background Technology

[0002] Aluminum foil defect detection is a crucial technology in modern manufacturing, especially in pharmaceutical packaging, food packaging, and new energy batteries, where surface quality directly impacts product safety and performance. High-quality aluminum foil must be free of micro-pinholes, scratches, or foreign matter, placing extremely high demands on detection technology. Traditional detection methods often rely on single optical imaging or manual sampling, which struggles to meet the real-time identification needs of complex surface features in high-speed production lines and are susceptible to interference from reflective materials, leading to frequent missed detections or false alarms. These limitations make existing solutions insufficient in terms of accuracy, efficiency, and adaptability to meet the intelligent manufacturing demands of Industry 4.0.

[0003] In the field of aluminum foil defect detection, the core challenge stems from the complexity of multimodal data acquisition and processing. In high-speed production lines, the specular reflection of the aluminum foil surface makes optical imaging susceptible to ambient light interference, making it difficult to reliably capture the characteristic information of minute defects. This imaging instability further hinders intelligent analysis algorithms from accurately distinguishing between novel or latent defects, such as microcracks or embedded foreign objects, during defect classification. The algorithm's insufficient generalization ability necessitates frequent updates to training data when dealing with aluminum foil of different materials or processes, increasing computational complexity and system maintenance costs. The interaction of these factors ultimately manifests as insufficient synergy between the detection system and existing production line equipment, particularly in high-speed dynamic environments, making it difficult to achieve low-latency closed-loop control.

[0004] Therefore, how to overcome specular reflection interference, improve algorithm generalization ability, and achieve real-time collaboration with production line equipment through multimodal data fusion and intelligent algorithm optimization in high-speed production lines has become a key issue for breakthroughs in aluminum foil strip defect detection technology. Summary of the Invention

[0005] This invention provides a method and system for detecting defects in aluminum foil strips, aiming to solve at least one of the defects in the prior art.

[0006] One aspect of the present invention relates to a method for detecting defects in aluminum foil strips, comprising the following steps:

[0007] Acquire multimodal data collected by optical and ultrasonic sensors in a high-speed production line and generate optical images;

[0008] The optical image is preprocessed by specular reflection, and an adaptive filter is used to eliminate ambient light interference to obtain the first feature set;

[0009] The first feature set and the ultrasonic signal are weighted and fused according to a preset weight matrix to obtain the second feature set;

[0010] Determine whether the second feature set contains defect features. If the second feature set contains defect features, classify the fused features using a convolutional neural network and output the defect classification result.

[0011] Support vector machines are used to generalize the defect classification results to obtain a generalization model;

[0012] The generalized model is integrated with the production line control system to generate control commands based on defect classification results. When a defect is detected, the production line speed parameters are adjusted.

[0013] Furthermore, the steps for acquiring multimodal data collected by optical and ultrasonic sensors in a high-speed production line and generating optical images include:

[0014] Optical data is acquired from an optical sensor, and ultrasonic data is acquired from an ultrasonic sensor. The data is synchronized with the ultrasonic data using timestamp alignment technology to obtain synchronized multimodal data.

[0015] For synchronized multimodal data, Gaussian filtering is used to denoise the synchronized multimodal data and generate optical images.

[0016] Further, the steps of performing specular reflection preprocessing on the optical image and using an adaptive filter to eliminate ambient light interference to obtain the first feature set include:

[0017] An optical image is acquired, and a reflection detection algorithm is used to determine whether there is a specular reflection region in the optical image. If it exists, the specular reflection region is corrected through geometric transformation to obtain the corrected first image.

[0018] For the first image, an adaptive filter is used to detect the pixel brightness of the first image. If the pixel brightness exceeds a preset threshold, the pixel value is adjusted by local mean filtering to obtain a second image that eliminates ambient light interference.

[0019] Based on the second image, obtain the grayscale histogram of the second image, and adjust the distribution of the grayscale histogram using histogram equalization technology to obtain a third image with enhanced contrast;

[0020] For the third image, a convolutional neural network algorithm is used to extract the spatial and texture features of the third image to generate the first feature set.

[0021] Furthermore, the step of weighted fusing the first feature set and the ultrasonic signal according to a preset weight matrix to obtain the second feature set includes:

[0022] The ultrasonic signal is acquired, and an adaptive filter is used to detect the noise level of the ultrasonic signal. If the noise level exceeds a preset threshold, the signal value is adjusted by local mean filtering to obtain the corrected first ultrasonic signal.

[0023] Based on the first ultrasonic signal, the frequency features of the first ultrasonic signal are extracted using Fourier transform to generate a first frequency feature set;

[0024] For the first frequency feature set and the first feature set, a preset weight matrix is ​​used to perform weighted fusion on the first frequency feature set and the first feature set to obtain the second feature set.

[0025] Further, the steps include determining whether the second feature set contains defect features. If the second feature set contains defect features, the steps also include classifying the fused features using a convolutional neural network and outputting the defect classification result:

[0026] A second feature set is obtained, and a preset threshold comparison method is used to determine whether the second feature set contains defect features, thereby obtaining the defect feature judgment result.

[0027] Determine whether the defect feature judgment result contains defect features. If the defect feature judgment result contains defect features, extract the fusion features from the second feature set to generate a fusion feature subset.

[0028] For the fused feature subset, a convolutional neural network is used to classify the fused feature subset to obtain the defect classification result;

[0029] Based on the defect classification results, the defect classification results are converted into structured data through preset mapping rules, and the final classification results are output.

[0030] Furthermore, the defect classification results are generalized using a support vector machine. The steps to obtain the generalized model include:

[0031] Obtain the defect classification results, use data preprocessing methods to normalize the defect classification results, extract the first feature subset from the defect classification results to obtain the normalized feature set;

[0032] Determine whether the normalized feature set meets the preset threshold judgment condition. If the normalized feature set meets the preset threshold judgment condition, then obtain the second feature subset from the normalized feature set through the feature extraction algorithm and determine the second feature subset.

[0033] For the second feature subset, a support vector machine is used to classify and train the second feature subset to generate a generalized classification model, thus obtaining the generalized model.

[0034] Furthermore, the generalized model is integrated with the production line control system to generate control commands based on defect classification results. The steps for adjusting production line speed parameters when a defect is detected include:

[0035] Real-time detection data is obtained from the production line control system. Data preprocessing methods are used to standardize the real-time detection data, and the first data subset containing defect features is extracted to obtain a standardized data set.

[0036] If the normalized dataset meets the preset threshold judgment condition, then the second data subset is obtained from the normalized dataset through the feature extraction algorithm, and the second data subset is determined.

[0037] For the second subset of data, a logistic regression algorithm is used to generate defect classification results, thus obtaining the defect classification results;

[0038] Based on the defect classification results, control commands are generated through a pre-established command mapping table to adjust the production line speed parameters and obtain the adjusted speed parameters.

[0039] Another aspect of the present invention relates to an aluminum foil strip defect detection system for implementing the above-described aluminum foil strip defect detection method. The aluminum foil strip defect detection system includes:

[0040] The first acquisition module is used to acquire multimodal data collected by optical sensors and ultrasonic sensors in the high-speed production line and generate optical images;

[0041] The second acquisition module is used to perform specular reflection preprocessing on the optical image, and to use an adaptive filter to eliminate ambient light interference to obtain the first feature set;

[0042] The third acquisition module is used to perform weighted fusion of the first feature set and the ultrasonic signal according to a preset weight matrix to obtain the second feature set.

[0043] The judgment module is used to determine whether the second feature set contains defect features. If the second feature set contains defect features, the fused features are classified through a convolutional neural network, and the defect classification result is output.

[0044] The fourth acquisition module is used to generalize the defect classification results using a support vector machine to obtain a generalized model.

[0045] The adjustment module is used to interface the generalized model with the production line control system, generate control commands based on the defect classification results, and adjust the production line speed parameters when a defect is detected.

[0046] Furthermore, the first acquisition module includes:

[0047] The first acquisition unit is used to acquire optical data from an optical sensor, acquire ultrasonic data from an ultrasonic sensor, and synchronize the data with the ultrasonic data using timestamp alignment technology to obtain synchronized multimodal data.

[0048] The first generation unit is used to perform noise reduction processing on the synchronous multimodal data using Gaussian filtering to generate an optical image.

[0049] Furthermore, the second acquisition module includes:

[0050] The second acquisition unit is used to acquire optical images and uses a reflection detection algorithm to determine whether there is a specular reflection area in the optical image. If it exists, the specular reflection area is corrected by geometric transformation to obtain the corrected first image.

[0051] The third acquisition unit is used to detect the pixel brightness of the first image using an adaptive filter. If the pixel brightness exceeds a preset threshold, the pixel value is adjusted by local mean filtering to obtain a second image that eliminates ambient light interference.

[0052] The fourth acquisition unit is used to acquire the grayscale histogram of the second image based on the second image, and adjust the distribution of the grayscale histogram through histogram equalization technology to obtain a third image with enhanced contrast.

[0053] The second generation unit is used to extract the spatial and texture features of the third image using a convolutional neural network algorithm to generate the first feature set.

[0054] The beneficial effects achieved by this invention are as follows:

[0055] This invention provides a method and system for detecting defects in aluminum foil strips. It acquires multimodal data from optical and ultrasonic sensors, performs specular reflection preprocessing and adaptive filtering on the optical images to eliminate ambient light interference, and then weightedly fuses the optical images with the ultrasonic signals to obtain a fused feature set. A convolutional neural network is used to classify the fused features containing defect characteristics, and a support vector machine is employed for generalization to obtain a generalization model. Finally, the generalization model is integrated with the production line control system to generate control commands based on the defect classification results, enabling automatic adjustment of the production line speed. This invention improves the accuracy and robustness of defect detection through multi-sensor data fusion and deep learning algorithms, achieving intelligent defect detection and control in high-speed production lines, effectively improving product quality and production efficiency. Attached Figure Description

[0056] Figure 1 This is a flowchart illustrating an embodiment of a defect detection method for aluminum foil strip according to the present invention. Detailed Implementation

[0057] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0058] like Figure 1 As shown, the first embodiment of the present invention proposes a method for detecting defects in aluminum foil strips, comprising the following steps:

[0059] Step S100: Acquire multimodal data collected by optical and ultrasonic sensors in the high-speed production line and generate an optical image.

[0060] Multimodal data refers to datasets acquired by various heterogeneous sensors or acquisition methods in a specific scenario, possessing complementarity or correlation. These datasets encompass different modalities, including text, images, time-series signals (such as mechanical parameters), and structured indicators (such as density and thickness). The core characteristic of multimodal data lies in achieving a comprehensive description and in-depth analysis of a target object or process through cross-modal fusion (such as the synergy between optical imaging and mechanical testing) and multidimensional representation (spatial distribution + time series). In this embodiment, the multimodal data integrates optical data detected by optical sensors and ultrasonic data detected by ultrasonic sensors.

[0061] Step S200: Perform specular reflection preprocessing on the optical image, and use an adaptive filter to eliminate ambient light interference to obtain the first feature set.

[0062] Specular reflection preprocessing refers to a technical process in optical inspection or computer graphics that optimizes specular reflection effects in advance by adjusting light source configuration, material properties, or algorithm parameters to improve subsequent analysis or rendering results. The core objectives of specular reflection preprocessing include suppressing interfering reflections (such as specular highlights masking real defects), enhancing effective signals (such as defect feature extraction), and optimizing computational efficiency (such as pre-computing reflection models).

[0063] An adaptive filter is an intelligent filtering system that dynamically adjusts its parameters based on the characteristics of the input signal and preset optimization criteria (such as minimizing the mean square error). Its core objective is to adapt to changes in the signal or environment by real-time correction of the filter coefficients, thereby achieving tasks such as noise suppression, signal enhancement, and system identification. Unlike traditional fixed-parameter filters, its core features lie in its self-learning ability and dynamic response mechanism, enabling it to achieve optimal filtering without relying on prior statistical knowledge.

[0064] The primary feature set refers to the set of basic features extracted from the raw data during the data modeling process, typically serving as the initial input for subsequent feature engineering or model training. Its core objective is to capture key patterns in the data that are directly related to the task objective and to provide underlying support for the construction of higher-order features (such as cross features and statistical aggregation features). The construction of this set needs to balance information content with computational efficiency, avoiding redundant noise that could interfere with the model's generalization performance.

[0065] Step S300: The first feature set and the ultrasonic signal are weighted and fused according to the preset weight matrix to obtain the second feature set.

[0066] A weight matrix is ​​a set of learnable parameters used in machine learning and deep learning models to represent the nonlinear mapping relationship between input features and output results. It is typically organized in matrix form, and its element values ​​are dynamically adjusted through backpropagation to minimize prediction error. Mathematically, it combines linear transformations (matrix multiplication) with nonlinear activation functions to achieve high-dimensional representation of data and extraction of complex patterns.

[0067] The second feature set refers to an enhanced feature set generated based on the first feature set through methods such as cross-modal fusion, higher-order combination, or domain knowledge injection. Its aim is to improve the model's ability to represent complex patterns and its task adaptability. Its core function is to compensate for the limitations of single modalities or basic features, strengthening the model's inference and generalization performance through multi-granularity information integration and semantic association.

[0068] Step S400: Determine whether the second feature set contains defect features. If the second feature set contains defect features, classify the fused features using a convolutional neural network and output the defect classification result.

[0069] Default features refer to key characterization parameters extracted from industrial products or image data, used to quantitatively describe appearance defects or abnormal areas. Their core objective is to transform the physical properties of defects (such as shape, texture, and grayscale distribution) into differentiated indicators that can be processed by mathematical models, providing data support for defect classification, location, and severity assessment.

[0070] Convolutional Neural Networks (CNNs) are deep learning architectures specifically designed for processing grid-like data (such as images, videos, and audio). Their core idea is to automatically learn spatial or temporal patterns in input data through local perception, weight sharing, and hierarchical feature extraction. Compared to fully connected neural networks, CNNs significantly reduce the number of parameters through convolutional operations while enhancing robustness to geometric changes such as translation and rotation, making them a cornerstone model in computer vision.

[0071] Defect classification results refer to the standardized defect category labels and their corresponding confidence scores output by extracting features and recognizing patterns from abnormal areas in industrial products or image data. These results must clearly define the defect type, severity, and spatial distribution to provide a quantitative basis for subsequent process optimization or quality decisions.

[0072] Step S500: Use a support vector machine to generalize the defect classification results and obtain a generalized model.

[0073] Support Vector Machine (SVM) is a supervised learning model based on statistical learning theory. Its core objective is to classify or regress data in high-dimensional space by constructing a maximum margin hyperplane. Its core idea is to minimize structural risk, that is, to maximize the distance between the decision boundary and the nearest sample point (support vector) while ensuring classification accuracy, thereby improving the model's generalization ability.

[0074] Generalization processing refers to the technical process of using algorithm design or data optimization to enable machine learning models to maintain stable predictive performance on unknown data outside the training set. Its core goal is to reduce the risk of overfitting the model to the training data and ensure that the learned features are universal, rather than just memorizing noise or irrelevant details of specific samples.

[0075] A generalizable model is a machine learning model that, after specific design or optimization, can maintain stable predictive performance on unseen or newly distributed data. Its core feature is to capture the essential patterns behind the data rather than noise by suppressing overfitting to the training data, thereby achieving reliable inference capabilities across scenarios and tasks.

[0076] Step S600: Connect the generalized model with the production line control system, generate control commands based on the defect classification results, and adjust the production line speed parameters when a defect is detected.

[0077] Production line speed parameters are core indicators for measuring production line operating efficiency. They refer to the number of products completed or the distance materials traveled per unit of time, usually expressed in units such as "pieces / hour," "meters / minute," or "kilograms / hour." Their core function is to balance efficiency and quality by quantifying production rhythm, and to provide data support for production planning, equipment scheduling, and process optimization.

[0078] Furthermore, in the aluminum foil strip defect detection method provided in this embodiment, step S100 includes:

[0079] Step S110: Acquire optical data from the optical sensor and ultrasonic data from the ultrasonic sensor. Synchronize the data with the ultrasonic data using timestamp alignment technology to obtain synchronized multimodal data.

[0080] In one possible implementation, the scenario for acquiring optical and ultrasonic data is set as defect detection of aluminum foil strips.

[0081] Optical sensors capture optical data on the surface of aluminum foil strips using a high-resolution camera, recording surface textures, scratches, or corrosion marks; ultrasonic sensors, on the other hand, acquire ultrasonic data from inside the aluminum foil strip by emitting and receiving ultrasonic waves, reflecting internal cracks or voids.

[0082] Timestamp alignment technology ensures synchronization between two types of data. For example, by using a unified clock signal, the timestamp error between optical data and ultrasonic data can be controlled within milliseconds, generating synchronized multimodal data.

[0083] This synchronization ensures spatiotemporal consistency in subsequent analysis, improving the accuracy of defect detection.

[0084] Step S120: For the synchronized multimodal data, Gaussian filtering is used to denoise the synchronized multimodal data to generate an optical image.

[0085] A Gaussian filter is a linear smoothing filter based on a Gaussian function, widely used in image processing, signal denoising, and other fields. Its core idea is to suppress high-frequency noise in data through weighted averaging while preserving as much of the original signal's structural characteristics as possible.

[0086] In this embodiment, the technical chain of precise synchronization → modal denoising → cross-modal fusion → deep optimization solves the problem of noise coupling and information loss in multimodal data. The generated optical image still has high fidelity in low light and high noise scenarios, which can meet the stringent visual requirements of industrial, medical, autonomous driving and other fields.

[0087] Furthermore, in the aluminum foil strip defect detection method provided in this embodiment, step S200 includes:

[0088] Step S210: Acquire an optical image and use a reflection detection algorithm to determine whether there is a specular reflection area in the optical image. If there is, correct the specular reflection area through geometric transformation to obtain the corrected first image.

[0089] For example, when processing optical images in a high-speed production line, a reflection detection algorithm is used to identify specular reflection areas.

[0090] Specular reflection is usually caused by the high gloss of the aluminum foil surface, resulting in localized overbrightness in the image and affecting subsequent analysis.

[0091] Reflection detection algorithms analyze pixel brightness gradients and reflection characteristics to determine the presence of bright areas. For example, when detecting aluminum foil strips, a brightness threshold of 200 is set; areas exceeding this value are marked as specular reflection areas. For these areas, geometric transformation correction adjusts the image viewing angle through rotation or perspective transformation to eliminate the influence of highlights, generating a corrected first image. This method ensures uniform image brightness distribution, providing a reliable foundation for subsequent processing.

[0092] Step S220: For the first image, an adaptive filter is used to detect the pixel brightness of the first image. If the pixel brightness exceeds a preset threshold, the pixel value is adjusted by local mean filtering to obtain a second image that eliminates ambient light interference.

[0093] In one possible implementation, an adaptive filter is used to detect the pixel brightness of the first image.

[0094] The adaptive filter dynamically adjusts its filtering parameters based on local pixel characteristics to adapt to different lighting conditions. For example, in aluminum foil surface detection, if the brightness of a pixel in a certain area exceeds a preset threshold of 180, the average brightness of a 3×3 neighborhood is calculated using local mean filtering. The excessively high pixel values ​​are then replaced with the mean value to generate a second image. This step effectively eliminates ambient light interference, ensures clear image details, and lays the foundation for subsequent feature extraction.

[0095] Step S230: Based on the second image, obtain the grayscale histogram of the second image, and adjust the distribution of the grayscale histogram using histogram equalization technology to obtain a third image with enhanced contrast.

[0096] It should be noted that histogram equalization is used to enhance the contrast of the second image.

[0097] A grayscale histogram reflects the distribution of pixel grayscale values. If the distribution is concentrated, the contrast is low.

[0098] Histogram equalization expands the dynamic range by redistributing grayscale values. For example, in an aluminum foil strip image, the original grayscale values ​​are concentrated in the 50-100 range. By equalization, the grayscale values ​​are expanded to 0-255, generating a third image. This process improves the visibility of image texture, facilitating subsequent feature extraction.

[0099] Step S240: For the third image, use a convolutional neural network algorithm to extract the spatial and texture features of the third image to generate the first feature set.

[0100] Specifically, convolutional neural network algorithms are used to extract spatial and texture features from a third image.

[0101] Convolutional neural networks (CNNs) capture information such as edges and textures in images through multiple layers of convolution and pooling operations. For example, in aluminum foil defect detection, the first layer of the network uses 32 3×3 convolutional kernels to extract low-level features, such as scratch edges; subsequent layers generate high-level features through pooling operations, forming the first feature set. This feature set contains spatial location and texture information, providing data support for defect classification.

[0102] In one embodiment, the combination of reflection detection and geometric transformation ensures the integrity of image preprocessing. For example, for the highlight areas on the surface of aluminum foil strips, the reflection detection algorithm first locates the highlight areas, and then adjusts the viewing angle through geometric transformation to generate a uniformly illuminated image. Adaptive filtering and histogram equalization further optimize image quality, while convolutional neural networks extract key features. This multi-step collaborative work, from image correction to feature extraction, forms a complete processing chain.

[0103] Preferably, the above-mentioned techniques are highly adaptable to practical applications. For example, under different lighting conditions, the adaptive filter can dynamically adjust its parameters to cope with changes in light; histogram equalization adapts to aluminum foil strips of different materials, ensuring the stability of feature extraction. These methods collectively improve the robustness of image processing and provide high-quality data for subsequent analysis.

[0104] Understandably, each step in the defect detection process for aluminum foil strips is closely interconnected. For example, reflection correction provides a clear input for filtering, the filtered image provides uniform brightness for equalization, and the equalized image provides high-contrast input for the neural network. This interlocking logic ensures that the processing flow from the original image to the feature set is efficient and reliable, and the final generated first feature set accurately reflects the spatial and textural characteristics of the aluminum foil strip surface.

[0105] Furthermore, in the aluminum foil strip defect detection method provided in this embodiment, step S300 includes:

[0106] Step S310: Acquire the ultrasound signal, use an adaptive filter to detect the noise level of the ultrasound signal, and if the noise level exceeds a preset threshold, adjust the signal value through local mean filtering to obtain the corrected first ultrasound signal.

[0107] For example, in the defect detection of aluminum foil strips, the acquisition of ultrasonic signals is a key step. Ultrasonic signals are emitted and received by an ultrasonic probe to detect defects inside or on the surface of the aluminum foil strip.

[0108] An ultrasonic probe emits pulses at a fixed frequency, such as 5 MHz. The received signal contains information about the material properties, but is often accompanied by noise interference. The signal acquisition equipment records the original ultrasonic waveform, generating an initial data stream, which provides the basis for subsequent processing.

[0109] In one possible implementation, an adaptive filter is used to detect the noise level of the ultrasound signal. The adaptive filter dynamically adjusts its parameters based on the local characteristics of the signal to identify the noise intensity. For example, a noise threshold is set to 20% of the signal amplitude; if the noise amplitude of a segment of the signal exceeds this value, the filter will initiate processing.

[0110] Local mean filtering calculates the average amplitude of a 5-point neighborhood and replaces excessively noisy signal points to generate a corrected first ultrasonic signal. This process preserves the original characteristics of the signal while reducing noise interference, providing a clear input for frequency analysis.

[0111] Step S320: Based on the first ultrasonic signal, use Fourier transform to extract the frequency features of the first ultrasonic signal and generate a first frequency feature set.

[0112] It should be noted that the Fourier transform is used to extract the frequency features of the first ultrasonic signal. The Fourier transform converts the time-domain signal into the frequency domain, revealing the signal's frequency distribution. For example, in aluminum foil strip detection, the Fourier transform analyzes the signal's spectrum, extracting the dominant frequency component (e.g., 3MHz) and the secondary frequency component (e.g., 6MHz), generating the first frequency feature set. This feature set reflects the acoustic characteristics of the aluminum foil strip's internal structure, facilitating subsequent feature fusion.

[0113] Step S330: For the first frequency feature set and the first feature set, a preset weight matrix is ​​used to perform weighted fusion on the first frequency feature set and the first feature set to obtain the second feature set.

[0114] Specifically, the preset weight matrix plays a key role in the weighted fusion of the first frequency feature set and the first feature set.

[0115] The weight matrix is ​​assigned a proportion based on the importance of the features. For example, the weight of frequency features is set to 0.6, and the weight of texture features is set to 0.4.

[0116] The fusion process integrates two sets of features through matrix operations to generate a second feature set. For example, in the detection of defects in aluminum foil strips, frequency features reflect the acoustic response of internal cracks, while texture features reflect the visual information of surface scratches. The second feature set generated after weighted fusion combines the advantages of both, improving the comprehensiveness of defect identification.

[0117] In one embodiment, the weighted fusion is implemented with consideration for adaptability to different detection scenarios. For example, in a noisy environment, the weight of the frequency features can be dynamically adjusted to 0.7 to enhance the detection capability of internal defects. This flexibility ensures the robustness of the fused feature set.

[0118] Preferably, each step of the above method is closely linked. The correction of the ultrasonic signal provides high-quality input for frequency feature extraction, while the fusion of frequency and texture features forms a more comprehensive feature description. This interconnected logical relationship ensures that the processing flow from signal acquisition to feature fusion is efficient and reliable.

[0119] Understandably, the combination of ultrasonic signals and optical image features is highly compatible for aluminum foil strip inspection. For example, frequency features can reveal internal micro-cracks, while texture features capture surface scratches; the resulting second feature set after fusion provides rich data support for subsequent defect classification. This multi-dimensional feature integration strategy significantly improves the accuracy and reliability of the inspection.

[0120] Furthermore, in the aluminum foil strip defect detection method provided in this embodiment, step S400 includes:

[0121] Step S410: Obtain the second feature set, and use a preset threshold comparison method to determine whether the second feature set contains defect features, and obtain the defect feature judgment result.

[0122] For example, in aluminum foil strip defect detection, after obtaining the second feature set, it is necessary to determine whether it contains defect features.

[0123] The preset threshold comparison method is based on the distribution characteristics of feature values. For example, if the second feature set includes frequency features and texture features, the threshold for frequency features is set to 3.5MHz, and the entropy threshold for texture features is set to 0.8. If a feature value exceeds the corresponding threshold, it is determined to contain a defect feature. This method ensures the objectivity of the judgment by quantifying the degree of anomaly of the features.

[0124] Step S420: Determine whether the defect feature judgment result contains defect features. If the defect feature judgment result contains defect features, extract the fusion features from the second feature set to generate a fusion feature subset.

[0125] In one possible implementation, a fusion feature subset is generated by extracting fusion features from the second feature set. This subset focuses on core defect-related information. For example, it might extract the amplitude of the 3MHz main frequency from the frequency features and the region with high entropy from the texture features to generate a subset. This subset reduces the complexity of subsequent processing by selecting highly relevant features, while retaining the key descriptions of the defects.

[0126] Step 430: For the fused feature subset, a convolutional neural network is used to classify the fused feature subset to obtain the defect classification result.

[0127] It should be noted that convolutional neural networks are used to classify fused feature subsets.

[0128] Convolutional neural networks extract feature patterns through multiple layers of convolution and pooling. For example, the network input is a fused feature subset, the first convolutional kernel is 3x3 to extract local features, the pooling layer compresses the data size, and the final fully connected layer outputs the classification result, such as "crack", "scratch", or "no defect". This layered processing can effectively capture the spatial and frequency domain characteristics of the feature subset.

[0129] Step S440: Based on the defect classification results, convert the defect classification results into structured data using preset mapping rules, and output the final classification results.

[0130] Specifically, the defect classification results need to be converted into structured data for output.

[0131] The pre-defined mapping rules associate the classification results with information such as defect type and location. For example, if the classification result is "crack", the mapping rules generate structured data containing defect type, center coordinates, and confidence level, such as "crack, (x:50, y:30), 0.95". This format facilitates subsequent system integration and decision support.

[0132] In one embodiment, the threshold comparison method can be adaptively adjusted according to the detection environment. For example, in a noisy environment, the frequency feature threshold can be dynamically increased to 4MHz to reduce false positives. This flexibility enhances the robustness of the method.

[0133] Preferably, the training process of the convolutional neural network takes into account the diversity of defects in the aluminum foil strip. For example, the training dataset contains various crack and scratch samples to ensure the network's ability to generalize to different defect patterns. The accuracy of the classification results is thus improved.

[0134] Understandably, the output design of structured data takes into account practical application needs. For example, the output coordinate information can be directly interfaced with automated equipment to guide precise repair. This interconnected process, from feature judgment to result output, ensures the efficiency and practicality of the detection.

[0135] Furthermore, in the aluminum foil strip defect detection method provided in this embodiment, step S500 includes:

[0136] Step S510: Obtain the defect classification results, use data preprocessing methods to normalize the defect classification results, extract the first feature subset from the defect classification results, and obtain the normalized feature set.

[0137] For example, in the scenario of aluminum foil strip defect detection, after obtaining the classification results, the classification results need to be normalized to ensure data consistency. For example, the classification results may include defect type and confidence level, such as "scratch, 0.85".

[0138] Data preprocessing methods can normalize the confidence level to the range of 0 to 1, removing outliers such as results with a confidence level below 0.1. This normalization standardizes the data format, facilitating subsequent analysis.

[0139] In one possible implementation, a first subset of features is extracted from the classification results to generate a normalized feature set. For example, the defect type, confidence level, and grayscale values ​​of the defect region are extracted from the classification results to form the first subset of features. The grayscale values ​​may range from 0 to 255, and are normalized to 0 to 1 through a linear transformation to generate the normalized feature set. This method reduces the impact of redundant data by focusing on key information.

[0140] Step S520: Determine whether the normalized feature set meets the preset threshold judgment condition. If the normalized feature set meets the preset threshold judgment condition, then obtain the second feature subset from the normalized feature set through the feature extraction algorithm and determine the second feature subset.

[0141] It should be noted that the normalized feature set must meet preset threshold conditions to determine whether further processing is required. For example, the confidence threshold is set to 0.7, and the grayscale feature mean threshold is set to 0.5.

[0142] If the normalized feature set meets certain conditions, such as a confidence level of 0.85 and a mean gray level of 0.6, then a second feature subset is obtained through a feature extraction algorithm. The feature extraction algorithm can employ principal component analysis to extract high-variance features, such as the edge intensity of the defect region, forming the second feature subset. This selection process highlights defect-related characteristics.

[0143] Step S530: For the second feature subset, use a support vector machine to classify and train the second feature subset to generate a generalized classification model, thus obtaining the generalized model.

[0144] Specifically, for the second feature subset, a support vector machine is used for classification training to generate a generalized classification model. For example, the second feature subset includes edge strength and region entropy values. The support vector machine constructs a hyperplane to classify the features into categories such as "crack" and "scratch".

[0145] During training, labeled samples, such as 1000 crack samples and 800 scratch samples, are used to ensure that the model learns diverse patterns. This classification training can generate a model with generalization ability.

[0146] In one embodiment, the generation of the generalized classification model takes into account the complexity of defects in aluminum foil strips. For example, by incorporating samples under different lighting conditions into the training data, the model can adapt to changes in lighting conditions in the production environment. This diverse training improves the robustness of the model.

[0147] Preferably, the feature extraction algorithm for the second feature subset can be dynamically adjusted according to the scene. For example, in a high-noise environment, texture features are extracted first rather than grayscale features to reduce noise interference. This adaptive adjustment enhances the targeting of feature extraction.

[0148] Understandably, the combination of normalization and feature extraction ensures logical consistency from classification results to the generalization model. For example, the normalized feature set is filtered through thresholds to reduce invalid data; the second feature subset focuses on highly relevant features, reducing training complexity; and the classification training using the vector machine ensures the model's accuracy. This interconnected process guarantees the efficiency and reliability of defect detection.

[0149] Furthermore, in the aluminum foil strip defect detection method provided in this embodiment, step S600 includes:

[0150] Step S610: Obtain real-time detection data from the production line control system, use data preprocessing methods to standardize the real-time detection data, extract the first data subset containing defect features, and obtain a standardized data set.

[0151] For example, in an aluminum foil strip defect detection scenario, acquiring real-time detection data from the production line control system involves raw data streams captured by sensors and cameras. For instance, a camera generates 100 frames per second, each containing pixel grayscale values ​​and defect location information. Data preprocessing methods must ensure the consistency of this data format.

[0152] In one possible implementation, preprocessing includes normalizing the grayscale values, linearly transforming the range of 0 to 255 to the interval of 0 to 1, while removing noisy data, such as abnormal frames with excessively large grayscale value fluctuations. This normalization process standardizes the data, facilitating subsequent feature extraction.

[0153] Specifically, when extracting the first subset of data containing defect features, key information can be filtered from the normalized data. For example, the texture features and area proportions of defective regions can be extracted from each frame of the image to form the first subset of data.

[0154] Texture features can be generated using a local binary pattern algorithm, and the area proportion is calculated as the ratio of defective pixels to the total number of pixels. Assuming the defect area proportion in a frame is 0.2 and the texture feature value is 0.75, these data constitute the first subset. This method focuses on the core attributes of the defect, reducing interference from irrelevant information.

[0155] Step S620: Determine whether the normalized dataset meets the preset threshold judgment condition. If the normalized dataset meets the preset threshold judgment condition, then obtain the second data subset from the normalized dataset through the feature extraction algorithm and determine the second data subset.

[0156] In one embodiment, the normalized dataset must meet preset threshold judgment conditions. For example, the area ratio threshold is set to 0.1, and the texture feature threshold is set to 0.5. If the area ratio of a certain group of data in the dataset is 0.2 and the texture feature value is 0.75, the condition is met, and then the next step of processing is performed.

[0157] It should be noted that the threshold setting needs to be adjusted according to the actual production line environment. For example, the threshold can be increased under high precision requirements to filter out more significant defect data.

[0158] Preferably, when obtaining the second data subset using a feature extraction algorithm, a dimensionality reduction method such as linear discriminant analysis can be used to extract highly discriminative features from the first data subset. For example, shape factor and contrast features of the defect region can be extracted to generate the second data subset. Assuming a shape factor of 0.8 and a contrast of 0.6, these features can better characterize the defect type. This method improves classification efficiency by focusing on highly relevant features.

[0159] Step S630: For the second data subset, use the logistic regression algorithm to generate defect classification results, and obtain the defect classification results.

[0160] For example, for the second subset of data, the logistic regression algorithm is used to generate defect classification results.

[0161] Logistic regression categorizes data into types such as "bubbles" and "wrinkles" by learning the mapping relationship between features and defect categories. During training, labeled samples are used, such as 500 bubble samples and 400 wrinkle samples, to ensure the model learns diverse features. The defect classification result might be "bubble, confidence 0.9," providing a basis for subsequent instruction generation.

[0162] Step S640: Based on the defect classification results, generate control commands through a pre-established command mapping table, adjust the production line speed parameters, and obtain the adjusted speed parameters.

[0163] In one possible implementation, control instructions are generated using a pre-established instruction mapping table based on the defect classification results. For example, the mapping table specifies that if a bubble defect is detected with a confidence level higher than 0.8, the production line speed should be reduced by 10%.

[0164] Assuming the classification result is "bubble, 0.9", the generation command will adjust the speed from 1 m / s to 0.9 m / s. This mapping mechanism can quickly respond to defect situations.

[0165] Understandably, the adjusted speed parameters need to be fed back to the production line control system in real time. For example, after the speed is adjusted, the system monitors the subsequent defect rate through sensors. If the defect rate decreases, the effectiveness of the instruction is verified. This closed-loop mechanism ensures the stability and efficiency of the production line operation.

[0166] This invention relates to an aluminum foil strip defect detection system for implementing the aforementioned aluminum foil strip defect detection method. The aluminum foil strip defect detection system includes a first acquisition module, a second acquisition module, a third acquisition module, a judgment module, a fourth acquisition module, and an adjustment module. The first acquisition module acquires multimodal data collected by optical and ultrasonic sensors in a high-speed production line to generate an optical image. The second acquisition module performs specular reflection preprocessing on the optical image and uses an adaptive filter to eliminate ambient light interference, obtaining a first feature set. The third acquisition module performs weighted fusion of the first feature set and the ultrasonic signal according to a preset weight matrix to obtain a second feature set. The judgment module determines whether the second feature set contains defect features. If the second feature set contains defect features, a convolutional neural network is used to classify the fused features and output a defect classification result. The fourth acquisition module uses a support vector machine to generalize the defect classification result to obtain a generalization model. The adjustment module interfaces the generalization model with the production line control system, generates control commands based on the defect classification result, and adjusts the production line speed parameters when a defect is detected.

[0167] Furthermore, the aluminum foil strip defect detection system provided in this embodiment includes a first acquisition module comprising a first acquisition unit and a first generation unit. The first acquisition unit is used to acquire optical data from an optical sensor and ultrasonic data from an ultrasonic sensor, and to synchronize the data with the ultrasonic data using timestamp alignment technology to obtain synchronized multimodal data. The first generation unit is used to perform noise reduction processing on the synchronized multimodal data using Gaussian filtering to generate an optical image.

[0168] Preferably, the aluminum foil strip defect detection system provided in this embodiment includes a second acquisition module comprising a second acquisition unit, a third acquisition unit, a fourth acquisition unit, and a second generation unit. The second acquisition unit acquires an optical image and uses a reflection detection algorithm to determine whether a specular reflection region exists in the optical image. If it exists, the specular reflection region is corrected through geometric transformation to obtain a corrected first image. The third acquisition unit uses an adaptive filter to detect the pixel brightness of the first image. If the pixel brightness exceeds a preset threshold, the pixel value is adjusted through local mean filtering to obtain a second image free from ambient light interference. The fourth acquisition unit acquires the grayscale histogram of the second image and adjusts the distribution of the grayscale histogram through histogram equalization technology to obtain a third image with enhanced contrast. The second generation unit uses a convolutional neural network algorithm to extract the spatial and texture features of the third image to generate a first feature set.

[0169] This embodiment provides a method and system for detecting defects in aluminum foil strips. Compared with existing technologies, it acquires multimodal data from optical and ultrasonic sensors, performs specular reflection preprocessing and adaptive filtering on the optical image to eliminate ambient light interference, and then weightedly fuses the optical image with the ultrasonic signal to obtain a fused feature set. A convolutional neural network is used to classify the fused features containing defect characteristics, and a support vector machine is employed for generalization to obtain a generalization model. Finally, the generalization model is integrated with the production line control system to generate control commands based on the defect classification results, enabling automatic adjustment of the production line speed. This embodiment improves the accuracy and robustness of defect detection through multi-sensor data fusion and deep learning algorithms, realizing intelligent defect detection and control in high-speed production lines, effectively improving product quality and production efficiency.

[0170] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A method for detecting defects in aluminum foil strips, characterized in that, Includes the following steps: Acquire multimodal data collected by optical and ultrasonic sensors in a high-speed production line and generate optical images, specifically including: Optical data is acquired from an optical sensor, and ultrasonic data is acquired from an ultrasonic sensor. The data is synchronized with the ultrasonic data using timestamp alignment technology to obtain synchronized multimodal data. For synchronized multimodal data, Gaussian filtering is used to denoise the synchronized multimodal data and generate optical images; The optical image is preprocessed by specular reflection, and an adaptive filter is used to eliminate ambient light interference to obtain the first feature set; The first feature set and the ultrasonic signal are weighted and fused according to a preset weight matrix to obtain the second feature set, which specifically includes: An ultrasound signal is acquired, and an adaptive filter is used to detect the noise level of the ultrasound signal. If the noise level exceeds a preset threshold, the signal value is adjusted by local mean filtering to obtain a corrected first ultrasound signal. Based on the first ultrasonic signal, Fourier transform is used to extract the frequency features of the first ultrasonic signal to generate a first frequency feature set; For the first frequency feature set and the first feature set, a preset weight matrix is ​​used to perform weighted fusion on the first frequency feature set and the first feature set to obtain the second feature set; Determine whether the second feature set contains defect features. If the second feature set contains defect features, then classify the fused features using a convolutional neural network and output the defect classification result, specifically including: The second feature set is obtained, and a preset threshold comparison method is used to determine whether the second feature set contains defect features, thereby obtaining the defect feature judgment result. The second feature set contains frequency features and texture features. The threshold for frequency features is set to 3.5MHz, and the entropy threshold for texture features is set to 0.

8. If one of the feature values ​​of frequency features and texture features exceeds the corresponding threshold, it is determined that the feature contains defect features. Determine whether the defect feature determination result contains defect features. If the defect feature determination result contains defect features, then extract the fusion features from the second feature set to generate a fusion feature subset. For the fused feature subset, a convolutional neural network is used to classify the fused feature subset to obtain the defect classification result; Based on the defect classification results, the defect classification results are converted into structured data using preset mapping rules, and the final classification results are output. The defect classification results are generalized using a support vector machine to obtain a generalization model; The generalized model is integrated with the production line control system. Control commands are generated based on the defect classification results. When a defect is detected, the production line speed parameters are adjusted. Specifically, this includes: Real-time detection data is obtained from the production line control system. The real-time detection data is normalized using a data preprocessing method. A first data subset containing defect features is extracted to obtain a normalized data set. If the normalized dataset meets a preset threshold condition, then a second data subset is obtained from the normalized dataset using a feature extraction algorithm. For the second subset of data, a logistic regression algorithm is used to generate defect classification results, thus obtaining the defect classification results; Based on the defect classification results, control commands are generated through a pre-established command mapping table to adjust the production line speed parameters and obtain the adjusted speed parameters.

2. The method for detecting defects in aluminum foil strips as described in claim 1, characterized in that, The steps of performing specular reflection preprocessing on the optical image and using an adaptive filter to eliminate ambient light interference to obtain the first feature set include: An optical image is acquired, and a reflection detection algorithm is used to determine whether there is a specular reflection region in the optical image. If there is, the specular reflection region is corrected by geometric transformation to obtain a corrected first image. For the first image, an adaptive filter is used to detect the pixel brightness of the first image. If the pixel brightness exceeds a preset threshold, the pixel value is adjusted by local mean filtering to obtain a second image that eliminates ambient light interference. Based on the second image, obtain the grayscale histogram of the second image, and adjust the distribution of the grayscale histogram using histogram equalization technology to obtain a third image with enhanced contrast; For the third image, a convolutional neural network algorithm is used to extract the spatial and texture features of the third image to generate a first feature set.

3. The method for detecting defects in aluminum foil strips as described in claim 1, characterized in that, The step of generalizing the defect classification results using a support vector machine to obtain a generalized model includes: The defect classification results are obtained, and the defect classification results are normalized using a data preprocessing method. A first feature subset is extracted from the defect classification results to obtain a normalized feature set. Determine whether the normalized feature set meets a preset threshold judgment condition. If the normalized feature set meets the preset threshold judgment condition, then obtain a second feature subset from the normalized feature set through a feature extraction algorithm and determine the second feature subset. For the second feature subset, a support vector machine is used to classify and train the second feature subset to generate a generalized classification model, thus obtaining the generalized model.

4. A defect detection system for aluminum foil strips, used to implement the defect detection method for aluminum foil strips as described in any one of claims 1 to 3, characterized in that, The aluminum foil strip defect detection system includes: The first acquisition module is used to acquire multimodal data collected by optical sensors and ultrasonic sensors in the high-speed production line and generate optical images; The second acquisition module is used to perform specular reflection preprocessing on the optical image, and to use an adaptive filter to eliminate ambient light interference to obtain a first feature set; The third acquisition module is used to perform weighted fusion of the first feature set and the ultrasonic signal according to a preset weight matrix to obtain the second feature set. The judgment module is used to determine whether the second feature set contains defect features. If the second feature set contains defect features, the fused features are classified by a convolutional neural network, and the defect classification result is output. The fourth acquisition module is used to generalize the defect classification results using a support vector machine to obtain a generalization model; The adjustment module is used to interface the generalized model with the production line control system, generate control commands based on the defect classification results, and adjust the production line speed parameters when a defect is detected.

5. The aluminum foil strip defect detection system as described in claim 4, characterized in that, The first acquisition module includes: The first acquisition unit is used to acquire optical data from an optical sensor, acquire ultrasonic data from an ultrasonic sensor, and synchronize the data with the ultrasonic data using timestamp alignment technology to obtain synchronized multimodal data. The first generation unit is used to perform noise reduction processing on the synchronous multimodal data using Gaussian filtering to generate an optical image.

6. The aluminum foil strip defect detection system as described in claim 4, characterized in that, The second acquisition module includes: The second acquisition unit is used to acquire an optical image and use a reflection detection algorithm to determine whether there is a specular reflection area in the optical image. If there is, the specular reflection area is corrected by geometric transformation to obtain a corrected first image. The third acquisition unit is used to detect the pixel brightness of the first image using an adaptive filter. If the pixel brightness exceeds a preset threshold, the pixel value is adjusted by local mean filtering to obtain a second image that eliminates ambient light interference. The fourth acquisition unit is used to acquire the grayscale histogram of the second image based on the second image, and adjust the distribution of the grayscale histogram through histogram equalization technology to obtain a third image with enhanced contrast. The second generation unit is used to extract the spatial and texture features of the third image using a convolutional neural network algorithm to generate a first feature set.

Images