Automatic product detection system based on machine vision
By introducing multi-spectral imaging and neural network processing in machine vision systems, combined with real-time feedback to adjust lighting conditions, the accuracy of detecting complex environments and diversified products in the prior art is solved, and more efficient and reliable product detection is achieved.
Patent Information
- Application Number
- CN202411064792.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-05
- Publication Date
- 2024-11-19
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing machine vision systems have technical bottlenecks in dealing with complex environments and diversified product characteristics, making it difficult to provide accurate and stable detection results, especially in the case of light changes and complex materials, it is impossible to effectively identify different materials or internal structures. Products affect the reliability and accuracy of testing.
Using an automated product detection system based on machine vision, combining multi-spectral imaging technology and neural network processing, it captures light at different wavelengths through multi-spectral light sources, performs image data processing and analysis, and uses convolutional neural networks and deep neural networks to simulate the biological brain. Visual information, extract image features, and provide optimal imaging conditions through real-time feedback and adaptive adjustment of lighting conditions.
Significantly improve the comprehensiveness and accuracy of detection, can capture subtle differences and hidden defects in multiple spectral ranges, improve the system's adaptability and detection performance, reduce human intervention and misjudgment, and improve the degree of automation of the production line. and efficiency.
Smart Images

Figure CN118976708A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of product detection, and in particular, relates to an automatic product detection system based on machine vision. Background Art
[0002] In modern industrial production, product quality inspection is an important link to ensure that products meet standards, reduce defective rates, and improve production efficiency. Traditional quality inspection methods mainly rely on manual inspection and single spectral imaging. These methods have problems such as low efficiency, high misjudgment rate, and poor adaptability, and are difficult to meet the diverse and complex production needs. With the development of artificial intelligence and multispectral imaging technology, automated inspection systems based on machine vision have gradually become an important direction for industrial inspection. However, existing machine vision systems still have technical bottlenecks in dealing with complex environments, diverse product characteristics, and real-time dynamic adjustments.
[0003] Traditional single-spectrum imaging methods face many limitations in industrial inspection. Single-spectrum imaging can usually only capture the surface features of products under specific lighting conditions, and cannot perform comprehensive inspections of subtle differences and defects that may exist under different spectra. This method is prone to missing details and potential defects when the lighting changes and the product materials are complex. For example, when inspecting transparent or translucent materials or products with highly reflective surfaces, single-spectrum imaging can hardly provide clear and accurate image data, resulting in inaccurate and unstable inspection results. In addition, the limitations of single-spectrum imaging make it impossible for the system to effectively identify and distinguish products with similar appearances but different materials or internal structures, affecting the reliability and accuracy of inspections.
[0004] In view of this, the present invention is proposed. Summary of the invention
[0005] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide an automated product inspection system based on machine vision, thereby solving the problems raised in the above-mentioned background technology.
[0006] In order to solve the above technical problems, the basic concept of the technical solution adopted by the present invention is:
[0007] An automated product inspection system based on machine vision includes: an image data acquisition unit, which acquires image data of a product and uploads the image data to an image processing unit;
[0008] An image processing unit, the image processing unit is used to simulate the visual information of a biological brain to process the image data obtained by the image processing unit and identify the product;
[0009] The multi-spectral analysis unit uses a multi-spectral light source to sequentially emit light of different wavelengths to illuminate the product to be inspected, uses a multi-spectral imaging device to synchronously capture the image information of the product, and analyzes the image features under different spectra;
[0010] An analysis unit, which is used to identify the corresponding product based on the acquired image data and automatically adjust parameters according to the characteristics of different products, and provide optimal imaging conditions through real-time feedback and adaptive adjustment of light intensity and angle;
[0011] The data processing and control unit performs comprehensive analysis on the collected multispectral images and neuromorphic calculation results, and provides real-time detection results and control signals;
[0012] The output unit displays the test results and outputs control signals to reject unqualified products.
[0013] Optionally, the steps of the image processing unit for simulating the visual information of a biological brain to process the image data acquired by the image processing unit are:
[0014] receiving image data captured by a multispectral or hyperspectral imaging unit and performing preprocessing operations on the image data;
[0015] Through convolutional neural networks, CNN simulates the primary visual cortex of the biological brain and performs multi-level feature extraction on the preprocessed image to extract basic features such as edges, textures, and colors in the image;
[0016] Using deeper neural network layers to simulate the advanced visual cortex of the biological brain, further combine and analyze primary features, identify complex shapes and patterns in images, and extract more advanced feature information;
[0017] According to the characteristic information in the image and the characteristics of the product, the parameters of the image processing unit are automatically adjusted through the feedback mechanism to optimize the processing effect and improve the detection ability of different products and defects;
[0018] The extracted high-level features are input into the classifier to comprehensively evaluate and classify the objects in the image to determine whether there are defects or abnormalities.
[0019] Optionally, the step of performing product recognition using a pre-trained deep neural network model based on the initially extracted features to determine the corresponding product category in the image is:
[0020] Integrate the initially extracted image features into a high-dimensional feature vector and input the feature vector into the pre-trained deep neural network model;
[0021] Automatically adjust the parameters of the analysis unit according to the identified product characteristics;
[0022] The analysis unit continuously accumulates and analyzes detection data, optimizes parameter adjustment strategies and algorithms, and improves the system's adaptive capabilities and detection performance;
[0023] The pre-trained deep neural network model performs forward propagation calculations, passes the input feature vector through each layer of the network, and gradually extracts higher-level abstract features, which is expressed as: H (l) =f(W (l) H (l-1) +b (l) ), where H (l) : The activation value of the lth layer, W (l) is the weight matrix of the lth layer, b (l) is the bias vector of the lth layer, f is the activation function ReLU, H (0) =X is the input high-dimensional feature vector;
[0024] In the last layer of the deep neural network model, the probability distribution of each possible product category is output and according to the probability distribution output by the classification layer, the category with the highest probability is selected as the corresponding product category in the image to complete product recognition. The expression is: P = softmax (W (L) H (L-1) +b (L) ), where P is the probability distribution of product categories and L is the number of the last layer of the deep neural network model.
[0025] Optionally, after completing product identification, you also need to pass the formula Evaluate the confidence of the recognition result. If the confidence is lower than a preset threshold, further processing or manual intervention is required. y The confidence of the recognition result, P y The probability of the identified product category y, H(P) is the entropy of the probability distribution P, which represents the uncertainty of the probability distribution, log(k) is the logarithm of the number of categories k in the probability distribution, which is used to standardize the entropy value, and S is the overall performance score of the model.
[0026] Optionally, the analysis unit is used to identify the corresponding product according to the acquired image data and automatically adjust the parameters according to the characteristics of different products, and provide the optimal imaging conditions through real-time feedback and adaptive adjustment of the illumination intensity and angle, as follows:
[0027] Obtaining the identified product type and sending the image data of the corresponding product to the analysis unit;
[0028] The analysis unit extracts the image quality, brightness distribution, and reflection under the current lighting conditions from the image and analyzes the image according to the identified product type T p , the current lighting condition information vector Q and the real-time feedback information F, using the lighting adjustment parameter calculation function Calculate the lighting adjustment parameter θ L ;
[0029] The analysis unit adjusts the brightness and angle of the light source according to the analysis results, and its expression is: L new is the new lighting condition after adjustment, L is the current lighting condition, and A is the light source adjustment function.
[0030] Obtain the adjusted images and perform quality verification to ensure that the lighting adjustment has achieved the expected effect;
[0031] If the lighting effect does not meet your expectations, return to the previous step for further adjustments.
[0032] Optionally, the steps of using a multi-spectral light source to sequentially emit light of different wavelengths to illuminate the product to be inspected, using a multi-spectral imaging device to synchronously capture image information of the product, and analyzing image features under different spectra are as follows:
[0033] Obtain the identified product and select a group of light with different wavelengths according to the preset detection requirements, and use a multi-spectral imaging device to synchronously capture the spectral image information of the product under the irradiation of each specific wavelength of light;
[0034] Acquire spectral image information and perform feature extraction on the image at each wavelength to obtain image features under different spectra;
[0035] Match the extracted multi-dimensional feature vector with the preset standard feature library to identify the characteristics and potential defects of the product;
[0036] The analysis results are output and verified for accuracy by comparing them with preset standards and historical data. The parameters of the multispectral light source and imaging equipment are adjusted in real time based on the analysis results and verification status.
[0037] Optionally, the data processing and control unit performs a comprehensive analysis on the collected multispectral images and neuromorphic calculation results to provide real-time detection results and control signals in the following steps:
[0038] receiving preliminary image features and analysis results processed by the neuromorphic computing architecture and multispectral image data captured from a multispectral imaging device;
[0039] The acquired multispectral image features are integrated with the neuromorphic calculation results to generate a comprehensive feature vector and assign corresponding weights to each feature in the comprehensive feature vector according to different products and detection requirements;
[0040] The analysis results are classified through classification algorithms to determine the quality status of the product and generate corresponding test results. For any abnormal conditions detected, an alarm signal is immediately generated and detailed abnormal information is recorded.
[0041] Optionally, the output unit displays the test results and outputs a control signal to remove unqualified products in the following steps:
[0042] The output unit receives the real-time test results and classification decisions from the data processing and control unit and displays the test results in real time through the display device, including the product's identification information, quality assessment, defect type and location;
[0043] Generate corresponding control signals based on the test results. For unqualified products, generate rejection signals and transmit the control signals to the actuators;
[0044] After receiving the control signal, the actuator immediately takes the rejection action to remove the unqualified products from the production line and confirms whether the unqualified products have been successfully rejected through sensors or detection devices.
[0045] After adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art. Of course, any product implementing the present invention does not necessarily need to achieve all the advantages described below at the same time:
[0046] By emitting light of different wavelengths in sequence through a multispectral imaging device, the image information of the product under multispectral conditions is captured, and combined with the preliminary image features processed by the neuromorphic computing architecture, a comprehensive feature vector is generated, which improves the comprehensiveness of image features and the accuracy of detection. This multispectral imaging technology can capture image data of products in multiple spectral ranges, revealing subtle differences and hidden defects that are not detectable under single spectral conditions. Each wavelength of light can highlight the reflection, absorption, and transmission characteristics of different materials and surface properties, allowing the system to comprehensively analyze the characteristics of the product from multiple dimensions. By combining multispectral image data and neuromorphic computing results, the system generates a comprehensive feature vector that covers detailed feature information of the product under various spectral conditions, significantly improving the comprehensiveness and accuracy of detection.
[0047] The specific implementation modes of the present invention are further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] The drawings described below are only some embodiments. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.
[0049] In the figure:
[0050] Figure 1 Block diagram of the automated product inspection system.
[0051] It should be noted that these drawings and textual descriptions are not intended to limit the conceptual scope of the present invention in any way, but are intended to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION
[0052] The present invention will now be described in further detail with reference to the accompanying drawings.
[0053] See also Figure 1 As shown, in this embodiment, an automated product inspection system based on machine vision is provided, including an image data acquisition unit, which acquires image data of the product and uploads the image data to an image processing unit; it should be noted that the image data acquisition unit effectively acquires image data of the product through a high-precision camera and sensor, and uploads the data to the image processing unit. This process ensures the real-time and accuracy of the image data, and provides a solid foundation for subsequent processing and analysis. In this way, the system can capture the appearance information of the product in real time, and promptly discover and record the appearance characteristics and potential defects of the product. This efficient data acquisition method not only improves the detection speed of the production line, but also reduces the possibility of human intervention and errors, thereby improving the reliability and automation of the entire system.
[0054] An image processing unit, the image processing unit is used to simulate the visual information of a biological brain to process the image data obtained by the image processing unit and identify the product;
[0055] It should be noted that the bio-simulation processing method of the image processing unit makes it far superior to traditional image processing technology in processing speed and accuracy. It can quickly adapt to products of different types and specifications, and improve the flexibility and adaptability of the production line.
[0056] The multi-spectral analysis unit uses a multi-spectral light source to sequentially emit light of different wavelengths to illuminate the product to be inspected, uses a multi-spectral imaging device to synchronously capture the image information of the product, and analyzes the image features under different spectra;
[0057] It should be noted that the material composition and internal structure information of the product are obtained through spectral analysis. Multi-spectral analysis technology enables the system to cope with various complex testing needs, especially in a production environment with diversified and complex materials, and provide reliable test results. Through multi-level testing methods, the quality and consistency of the product are ensured and the accuracy of the test is improved.
[0058] An analysis unit, which is used to identify the corresponding product based on the acquired image data and automatically adjust parameters according to the characteristics of different products, and provide optimal imaging conditions through real-time feedback and adaptive adjustment of light intensity and angle;
[0059] It should be noted that the optimal imaging conditions are provided through real-time feedback and adaptive adjustment of light intensity and angle. This design ensures that each product is inspected under the best lighting conditions, greatly improving the quality of image data and the accuracy of inspection results. Real-time feedback and adaptive adjustment functions enable the system to respond quickly to environmental changes and product differences, ensuring the stability and continuity of the inspection process. The intelligent adjustment mechanism of the analysis unit not only improves the automation level of the system, but also reduces the need for human intervention, reduces the complexity of operation and the probability of errors.
[0060] The data processing and control unit performs comprehensive analysis on the collected multispectral images and neuromorphic calculation results, and provides real-time detection results and control signals;
[0061] It should be noted that: by comprehensively analyzing various data sources and generating high-precision test results, accurate product testing and quality control are ensured. Real-time detection and control signal generation enable the system to respond and adjust quickly, maintaining the efficiency and stability of the production process.
[0062] The output unit displays the test results and outputs control signals to reject unqualified products.
[0063] It should be noted that by displaying the test results in real time, operators can timely understand the product quality status and production line operation status, and make necessary adjustments and decisions. The control signal generated by the output unit can automatically remove unqualified products and prevent unqualified products from entering the next production link, thus ensuring the quality of the final product. This design reduces the need for manual intervention and judgment, and improves the automation and efficiency of the production line.
[0064] In this embodiment, the steps of the image processing unit for simulating the visual information of the biological brain to process the image data acquired by the image processing unit are as follows:
[0065] receiving image data captured by a multispectral or hyperspectral imaging unit and performing preprocessing operations on the image data;
[0066] Through convolutional neural networks, CNN simulates the primary visual cortex of the biological brain and performs multi-level feature extraction on the preprocessed image to extract basic features such as edges, textures, and colors in the image;
[0067] It should be noted that CNN can capture local features and global patterns in images through multi-level convolution operations, significantly improving the accuracy and detail of feature extraction. These basic features such as edges, textures, and colors are the basic components of image recognition and analysis, and can help the system better understand and parse image content. By simulating the primary visual cortex of the biological brain, the system can analyze images in a way that is closer to human visual processing, improving the naturalness and accuracy of feature extraction. This approach not only improves detection efficiency, but also enables the system to adapt to various complex product features and image changes.
[0068] Using deeper neural network layers to simulate the advanced visual cortex of the biological brain, further combine and analyze primary features, identify complex shapes and patterns in images, and extract more advanced feature information;
[0069] It should be noted that using deeper neural network layers can deeply understand and analyze the complex structures and relationships in the image. This advanced feature analysis can combine primary features into more abstract and high-level feature representations through multiple layers of nonlinear transformations in deep neural networks, capturing complex shapes and patterns in the image. By simulating the advanced visual cortex of the biological brain, the system can identify subtle differences and features in the image that are difficult to detect, improving the ability to detect complex products and small defects.
[0070] According to the characteristic information in the image and the characteristics of the product, the parameters of the image processing unit are automatically adjusted through the feedback mechanism to optimize the processing effect and improve the detection ability of different products and defects;
[0071] It should be noted that the feedback mechanism automatically adjusts processing parameters such as filter settings, detection thresholds, and lighting conditions by analyzing image features and detection results in real time to ensure that each product is detected under optimal conditions. This adaptive adjustment capability not only improves the flexibility and accuracy of detection, but also reduces the complexity of human intervention and parameter settings, enabling the system to efficiently respond to emergencies and product changes on various production lines. Moreover, by continuously optimizing the processing effect, the system can continuously improve its detection capabilities for various products and defects, ensuring the efficient operation of the production line and the stability of product quality.
[0072] The extracted high-level features are input into the classifier to comprehensively evaluate and classify the objects in the image to determine whether there are defects or abnormalities.
[0073] It should be noted that the classifier can accurately identify different types and states of products, including potential defects and abnormalities, through comprehensive analysis of high-level features. Through this comprehensive evaluation, the system can make judgments quickly and accurately to ensure that each product meets quality standards. For detected abnormalities, the system can immediately generate an alarm signal and record detailed abnormal information, providing strong support for subsequent quality control and problem tracing.
[0074] In this embodiment, the steps of performing product recognition using a pre-trained deep neural network model based on the initially extracted features and determining the corresponding product category in the image are as follows:
[0075] Integrate the initially extracted image features into a high-dimensional feature vector and input the feature vector into the pre-trained deep neural network model;
[0076] Automatically adjust the parameters of the analysis unit according to the identified product characteristics;
[0077] The analysis unit continuously accumulates and analyzes detection data, optimizes parameter adjustment strategies and algorithms, and improves the system's adaptive capabilities and detection performance;
[0078] The pre-trained deep neural network model performs forward propagation calculations, passes the input feature vector through each layer of the network, and gradually extracts higher-level abstract features, which is expressed as: H (l) =f(W (l) H (l-1) +b (l) ), where H (l) : The activation value of the lth layer, W (l) is the weight matrix of the lth layer, b (l) is the bias vector of the lth layer, f is the activation function ReLU, H (0) =X is the input high-dimensional feature vector;
[0079] It should be noted that by gradually extracting higher-level abstract features, the system can gradually extract higher-level abstract features at different levels, improving the recognition ability of complex image structures and patterns. The forward propagation calculation of each layer can capture different levels and details in the image, making the final extracted features richer and more accurate. This multi-level feature extraction method greatly improves the system's image understanding ability and detection accuracy, especially when dealing with complex products and small defects.
[0080] In the last layer of the deep neural network model, the probability distribution of each possible product category is output and according to the probability distribution output by the classification layer, the category with the highest probability is selected as the corresponding product category in the image to complete product recognition. The expression is: P = softmax (W (L) H(L-1) +b (L) ), where P is the probability distribution of product categories and L is the number of the last layer of the deep neural network model.
[0081] It should be noted that the softmax function converts the output of the network into a probability distribution, so that the sum of the probabilities of each category is 1. This allows us to intuitively compare the possibilities of different categories and select the most likely category as the final recognition result. This probability distribution output method not only improves the accuracy of classification, but also provides an uncertainty assessment of the classification results, which helps to make more reliable decisions in practical applications.
[0082] In this embodiment, after completing product identification, it is also necessary to use the formula Evaluate the confidence of the recognition result. If the confidence is lower than a preset threshold, further processing or manual intervention is required. y The confidence of the recognition result, P y The probability of the identified product category y, H(P) is the entropy of the probability distribution P, which represents the uncertainty of the probability distribution, log(k) is the logarithm of the number of categories k in the probability distribution, which is used to standardize the entropy value, and S is the overall performance score of the model. This application provides a comprehensive and accurate measure of confidence by combining the probability of the recognition result, the uncertainty of the probability distribution, and the overall performance of the model. By introducing entropy and normalization factors, the formula can effectively measure the uncertainty of the recognition results and ensure the scientificity and rationality of the confidence assessment. At the same time, the introduction of the overall performance score of the model makes the confidence assessment not only rely on the current recognition results, but also combines the overall performance of the model during training and testing, further improving the reliability of the assessment.
[0083] In this embodiment, the analysis unit is used to identify the corresponding product according to the acquired image data and automatically adjust the parameters according to the characteristics of different products, and provide the optimal imaging conditions through real-time feedback and adaptive adjustment of the light intensity and angle. The steps are:
[0084] Acquire the identified product type and send the corresponding product image data to the analysis unit; the application first identifies the product type, and the system can perform customized analysis based on the requirements and characteristics of the specific product, which improves the accuracy and relevance of the detection. Send the corresponding image data to the analysis unit so that the analysis process is based on the latest and relevant visual information, ensuring the real-time and effectiveness of the data. This method not only improves the detection efficiency of the system, but also can promptly identify and handle defects and problems in the product, reducing the defective rate on the production line.
[0085] The analysis unit extracts the image quality, brightness distribution, and reflection under the current lighting conditions from the image and analyzes the image according to the identified product type T p, the current lighting condition information vector Q and the real-time feedback information F, using the lighting adjustment parameter calculation function Calculate the lighting adjustment parameter θ L By extracting the current lighting condition information, the system can monitor and analyze the changes in ambient lighting in real time to ensure that each test is performed under the best lighting environment. According to the product type and real-time feedback information, the lighting adjustment parameters are calculated to ensure the personalized and precise adjustment of lighting conditions. This dynamic adjustment mechanism not only improves the precision and accuracy of detection, but also reduces misjudgments and omissions caused by uneven lighting.
[0086] The analysis unit adjusts the brightness and angle of the light source according to the analysis results, and its expression is: L new is the new lighting condition after adjustment, L is the current lighting condition, and A is the light source adjustment function;
[0087] Obtain the adjusted images and perform quality verification to ensure that the lighting adjustment has achieved the expected effect;
[0088] If the lighting effect does not meet your expectations, return to the previous step for further adjustments.
[0089] In this embodiment, the steps of using a multi-spectral light source to sequentially emit light of different wavelengths to illuminate the product to be inspected, using a multi-spectral imaging device to synchronously capture image information of the product, and analyzing image features under different spectra are as follows:
[0090] Obtain the identified product and select a set of light of different wavelengths according to the preset detection requirements. Under the illumination of each specific wavelength, use the multi-spectral imaging device to synchronously capture the spectral image information of the product; by identifying the product type and detection requirements, the system can select the most effective wavelength combination to highlight the key features and potential defects of the product. The synchronous capture function of the multi-spectral imaging device ensures the consistency and synchronization of image data at different wavelengths, and improves the reliability and comparability of the data. This targeted and flexible spectral selection method can significantly improve the accuracy and efficiency of detection.
[0091] Acquire spectral image information and perform feature extraction on the image at each wavelength to obtain image features under different spectra;
[0092] The extracted multi-dimensional feature vectors are matched with the preset standard feature library to identify the characteristics and potential defects of the product; through comparative analysis, accurate product identification and defect detection are achieved. The preset standard feature library contains the feature data of different products as a reference standard. By matching the feature vectors extracted in real time with the standard library, the system can quickly and accurately identify the specific characteristics and potential defects of the product. This matching process can effectively detect minor anomalies and quality issues, ensuring that each product meets quality standards. The accuracy and efficiency of feature matching enable the system to maintain a high level of quality control in a large-scale production environment, reducing defective rates and production losses.
[0093] The analysis results are output, and the accuracy of the analysis results is verified by comparing them with the preset standards and historical data. According to the analysis results and verification, the parameters of the multispectral light source and imaging equipment are adjusted in real time. By comparing the analysis results with the preset standards and historical data, the system can verify the accuracy and reliability of the test results. Any deviations or anomalies can be discovered and corrected in a timely manner to ensure the high accuracy and stability of the detection process. The parameters of the multispectral light source and imaging equipment are adjusted in real time, so that the system can dynamically adapt to changes in the environment and products and maintain the optimal detection conditions. This feedback and adjustment mechanism not only improves the intelligence level of the system, but also reduces the need for manual intervention, reduces the complexity of operation and the probability of errors.
[0094] In this embodiment, the data processing and control unit performs a comprehensive analysis on the collected multispectral images and neuromorphic calculation results, and provides real-time detection results and control signals in the following steps:
[0095] Receive the preliminary image features and analysis results processed by the neuromorphic computing architecture and the multispectral image data captured from the multispectral imaging device; the multispectral imaging device provides image data under different spectra, which can reveal the characteristics and potential defects of the product under different lighting conditions. By receiving these two types of data at the same time, the system can obtain more comprehensive image information, which helps to improve the depth and breadth of detection and ensure that the characteristics of each product are fully captured and analyzed. This multi-source data fusion method can significantly improve the detection performance and reliability of the system.
[0096] The acquired multispectral image features are fused with the neuromorphic calculation results to generate a comprehensive feature vector and assign corresponding weights to each feature in the comprehensive feature vector according to different products and detection requirements; the multispectral image features and neuromorphic calculation results each contain information of different dimensions. By fusion, a richer and more comprehensive comprehensive feature vector is generated, and the system can more accurately describe and identify product characteristics. Assigning weights to each feature in the comprehensive feature vector according to different products and detection requirements can further improve the feature differentiation ability and detection effect.
[0097] The analysis results are classified through classification algorithms to determine the quality status of the product and generate corresponding test results. For any abnormal conditions detected, an alarm signal is immediately generated and detailed abnormal information is recorded.
[0098] In this embodiment, the steps of the output unit displaying the test results and outputting a control signal to remove unqualified products are as follows:
[0099] The output unit receives the real-time test results and classification decisions from the data processing and control unit and displays the test results in real time through the display device, including the product's identification information, quality assessment, defect type and location;
[0100] Generate corresponding control signals based on the test results. For unqualified products, generate rejection signals and transmit the control signals to the actuators;
[0101] After receiving the control signal, the actuator immediately takes the rejection action to remove the unqualified products from the production line and confirms whether the unqualified products have been successfully rejected through sensors or detection devices.
[0102] The present invention is not limited to the above-mentioned embodiments. Anyone should be aware that any structural changes made under the enlightenment of the present invention, and any technical solutions that are the same or similar to the present invention, fall within the protection scope of the present invention. The technology, shape, and structural parts not described in detail in the present invention are all well-known technologies.
Claims
1. An automated product inspection system based on machine vision, characterized in that: include: An image data acquisition unit, which acquires image data of a product and uploads the image data to an image processing unit; An image processing unit, the image processing unit is used to simulate the visual information of a biological brain to process the image data obtained by the image processing unit and identify the product; Multispectral analysis unit: The multispectral analysis unit uses a multispectral light source to sequentially emit light of different wavelengths to illuminate the product to be inspected, uses a multispectral imaging device to synchronously capture the image information of the product, and analyzes the image features under different spectra; An analysis unit, which is used to identify the corresponding product based on the acquired image data and automatically adjust parameters according to the characteristics of different products, and provide optimal imaging conditions through real-time feedback and adaptive adjustment of light intensity and angle; Data processing and control unit, which is used to comprehensively analyze the collected multispectral images and neuromorphic calculation results, and provide real-time detection results and control signals; The output unit displays the test results and outputs control signals to reject unqualified products.
2. The automatic product inspection system based on machine vision according to claim 1, characterized in that: The image processing unit is used to simulate the visual information of the biological brain. The steps of processing the image data obtained by the image processing unit are as follows: receiving image data captured by a multispectral or hyperspectral imaging unit and performing preprocessing operations on the image data; Through convolutional neural networks, CNN simulates the primary visual cortex of the biological brain and performs multi-level feature extraction on the preprocessed image to extract basic features such as edges, textures, and colors in the image; Using deeper neural network layers to simulate the advanced visual cortex of the biological brain, further combine and analyze primary features, identify complex shapes and patterns in images, and extract more advanced feature information; According to the characteristic information in the image and the characteristics of the product, the parameters of the image processing unit are automatically adjusted through the feedback mechanism to optimize the processing effect and improve the detection ability of different products and defects; The extracted high-level features are input into the classifier to comprehensively evaluate and classify the objects in the image to determine whether there are defects or abnormalities.
3. The automatic product inspection system based on machine vision according to claim 1, characterized in that: Based on the initially extracted features, the pre-trained deep neural network model is used to identify the product and determine the corresponding product category in the image. Integrate the initially extracted image features into a high-dimensional feature vector and input the feature vector into the pre-trained deep neural network model; Automatically adjust the parameters of the analysis unit according to the identified product characteristics; The analysis unit continuously accumulates and analyzes detection data, optimizes parameter adjustment strategies and algorithms, and improves the system's adaptive capabilities and detection performance; The pre-trained deep neural network model performs forward propagation calculations, passes the input feature vector through each layer of the network, and gradually extracts higher-level abstract features, which is expressed as: H (l) =f(W (l) H (l-1) +b (l) ), where H (l) : The activation value of the lth layer, W (l) is the weight matrix of the lth layer, b (l) is the bias vector of the lth layer, f is the activation function ReLU, H (0) =X is the input high-dimensional feature vector; In the last layer of the deep neural network model, the probability distribution of each possible product category is output and according to the probability distribution output by the classification layer, the category with the highest probability is selected as the corresponding product category in the image to complete product recognition. The expression is: P = softmax (W (L) H (L-1) +b (L) ), where P is the probability distribution of product categories and L is the number of the last layer of the deep neural network model.
4. The automatic product inspection system based on machine vision according to claim 1, characterized in that: After completing product identification, you also need to use the formula Evaluate the confidence of the recognition result. If the confidence is lower than a preset threshold, further processing or manual intervention is required. y The confidence of the recognition result, P y The probability of the identified product category y, H(P) is the entropy of the probability distribution P, which represents the uncertainty of the probability distribution, log(k) is the logarithm of the number of categories k in the probability distribution, which is used to standardize the entropy value, and S is the overall performance score of the model.
5. The automatic product inspection system based on machine vision according to claim 1, characterized in that: The analysis unit is used to identify the corresponding product based on the acquired image data and automatically adjust the parameters according to the characteristics of different products. The steps of providing the optimal imaging conditions through real-time feedback and adaptive adjustment of the light intensity and angle are as follows: Obtaining the identified product type and sending the image data of the corresponding product to the analysis unit; The analysis unit extracts the image quality, brightness distribution, and reflection under the current lighting conditions from the image and analyzes the image according to the identified product type T p , the current lighting condition information vector Q and the real-time feedback information F, using the lighting adjustment parameter calculation function Calculate the lighting adjustment parameter θ L ; The analysis unit adjusts the brightness and angle of the light source according to the analysis results, and its expression is: L new is the new lighting condition after adjustment, L is the current lighting condition, and A is the light source adjustment function. Obtain the adjusted images and perform quality verification to ensure that the lighting adjustment has achieved the expected effect; If the lighting effect does not meet your expectations, return to the previous step for further adjustments.
6. The automatic product inspection system based on machine vision according to claim 1, characterized in that: The steps of using a multi-spectral light source to sequentially emit light of different wavelengths to illuminate the product to be inspected, using a multi-spectral imaging device to synchronously capture the image information of the product, and analyzing the image features under different spectra are as follows: Obtain the identified product and select a set of light with different wavelengths according to the preset detection requirements, and use a multi-spectral imaging device to synchronously capture the spectral image information of the product under the irradiation of each specific wavelength of light; Acquire spectral image information and perform feature extraction on the image at each wavelength to obtain image features under different spectra; Match the extracted multi-dimensional feature vector with the preset standard feature library to identify the characteristics and potential defects of the product; The analysis results are output and verified for accuracy by comparing them with preset standards and historical data. The parameters of the multispectral light source and imaging equipment are adjusted in real time based on the analysis results and verification status.
7. The automatic product inspection system based on machine vision according to claim 1, characterized in that: The steps of the data processing and control unit to comprehensively analyze the collected multispectral images and neuromorphic calculation results and provide real-time detection results and control signals are as follows: receiving preliminary image features and analysis results processed by the neuromorphic computing architecture and multispectral image data captured from a multispectral imaging device; The acquired multispectral image features are integrated with the neuromorphic calculation results to generate a comprehensive feature vector and assign corresponding weights to each feature in the comprehensive feature vector according to different products and detection requirements; The analysis results are classified through classification algorithms to determine the quality status of the product and generate corresponding test results. For any abnormal conditions detected, an alarm signal is immediately generated and detailed abnormal information is recorded.
8. The automatic product inspection system based on machine vision according to claim 1, characterized in that: The steps of the output unit, displaying the test results, and outputting control signals to remove unqualified products are as follows: The output unit receives the real-time test results and classification decisions from the data processing and control unit and displays the test results in real time through the display device, including the product's identification information, quality assessment, defect type and location; Generate corresponding control signals based on the test results. For unqualified products, generate rejection signals and transmit the control signals to the actuators; After receiving the control signal, the actuator immediately takes the rejection action to remove the unqualified products from the production line and confirms whether the unqualified products have been successfully rejected through sensors or detection devices.
Citation Information
Cited By
EVA high-concentration color master batch detection method based on multispectral imaging
CN119223817A
Product detection method and system based on machine vision
CN119323659A
A product detection method and system based on machine vision
CN119323659B
Portable precise laser detector
CN119666332A
Biological recognition device based on multi-wavelength light source and GOI image sensor
CN120783400A
