3D deep learning defect detection method based on machine vision

Abstract

The invention discloses a 3D deep learning defect detection method based on machine vision, and relates to the technical field of visual detection, and the method comprises the following steps: S1, building a learning network: collecting and sorting an image data set, guaranteeing the diversity and representativeness of data, and obtaining the learning network; inputting the collected and sorted image data set for training, then testing the deep learning model architecture, then testing the deep learning model architecture after training, extracting features in the image data set and storing data, performing model training after extraction and storage, and testing the model after training to obtain a deep learning model architecture; after the test is completed, learning network construction can be completed; and S2, image acquisition: carrying out comprehensive scanning on the surface of the product by using a high-resolution 3D camera to obtain three-dimensional image data of the product, so that deep learning can be carried out during detection, manual importing is not needed, and the detection efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of visual inspection technology, and specifically to a 3D deep learning defect detection method based on machine vision. Background Art

[0002] Since the 1960s, the speed of computers has been significantly improved. With the advent of CCD technology, defect detection technology based on machine vision has been widely used in industrial production lines. For example, the machinery, electronics, printing, and textile industries have improved product quality and production efficiency with the help of advanced detection technology. As a non-destructive testing method, machine vision detection uses a linear or area array camera to obtain a clear image of the object to be tested, and the computer performs image processing to complete the real-time detection of target defects; Existing machine vision inspection requires manual import of defect photos into the data center before inspection, and then inspection is performed based on this data. This method is rather cumbersome. For this reason, we propose a 3D deep learning defect detection method based on machine vision. Summary of the invention

[0003] The technical problem to be solved by the present invention is to overcome the existing defects and provide a 3D deep learning defect detection method based on machine vision, which can perform deep learning during detection without the need for manual introduction, thereby improving the efficiency of detection and effectively solving the problems in the background technology.

[0004] To achieve the above object, the present invention provides the following technical solution: a 3D deep learning defect detection method based on machine vision, comprising the following steps: S1 builds a learning network: collects and organizes image data sets to ensure data diversity and representativeness, then inputs the collected and organized image data sets into the training to test the deep learning model architecture, then extracts features from the image data set and saves the data, then trains the model after extraction and preservation, and tests it after training. After the test is completed, the learning network construction is completed; S2 Image Acquisition: Use a high-resolution 3D camera to fully scan the product surface and obtain the three-dimensional image data of the product: S3 data preprocessing: preprocess the collected images, including denoising, enhancement, correction, etc., to improve the accuracy of subsequent processing; S4 Feature Extraction: Extract high-level features from images using deep learning networks; S5 defect detection: The extracted features are input into the learning network for comparison with the deep learning model architecture within the learning network to perform defect detection and identification. Unsaved defect data will be automatically uploaded to the deep learning network for storage. S6 Data Analysis and Reporting: Perform statistical analysis on the test results, generate a test report, and then display it through visualization technology.

[0005] Furthermore, in step 1, after the image data set is collected and organized, it needs to be preprocessed, such as resizing, format conversion, etc., to adapt to the input requirements of the network.

[0006] Furthermore, in step 1, after collecting and organizing the image data set, it is necessary to perform data enhancement on the collected and organized image data set, such as performing random cropping and rotation operations on the images, to increase the scale and diversity of the data set.

[0007] Furthermore, in the step 1, the deep learning model architecture is tested after training, and a convolutional neural network (CNN), a recurrent neural network (RNN), or a combination of the two can be selected. The convolution layer in the deep learning model architecture is tested after training to extract local features of the image in the image data set, and the pooling layer reduces the dimension of the features and retains important information. By stacking multiple convolutional layers and pooling layers, the network can gradually extract high-level features of the image.

[0008] Furthermore, during model training in step 1, the prepared data set is divided into a training set and a test set, the model is trained using the training set, and the weights and biases of the model are updated through a back propagation algorithm. During the training process, a cross entropy loss function can be used to measure the error of the model, and optimizers such as Adam can be used to accelerate the training process.

[0009] Furthermore, in the step 1, the model is tested using a test set to test the trained model, evaluate the performance of the model, and analyze the advantages and disadvantages of the model by calculating indicators such as the accuracy and recall rate of the model on the test set.

[0010] Furthermore, the denoising in step 3 needs to adopt methods such as mean filtering, median filtering or Gaussian filtering to remove random noise in the image and improve the signal-to-noise ratio of the image.

[0011] Furthermore, the enhancement in step 3 needs to be performed by adjusting brightness, contrast or using histogram equalization to enhance the visual effect of the image, make the image features more obvious, and improve the robustness of the recognition system.

[0012] Furthermore, the correction in step 3 needs to correct the image distortion caused by shooting angle or distance through geometric transformation (such as rotation, scaling, cropping, etc.) to ensure that the image can still be correctly recognized at different angles and sizes.

[0013] Furthermore, the visualization technology in step 6 may be t-SNE technology.

[0014] Compared with the prior art, the beneficial effects of the present invention are: the 3D deep learning defect detection method based on machine vision has the following advantages: 1. The present invention uploads the defect detection data detected during the detection process to the learning network so that the learning network automatically increases the database without manual uploading, which not only improves efficiency but also improves the accuracy of inspection during a long detection process.

[0015] 2. The present invention can reduce or eliminate noise in the image, enhance the contrast and brightness of the image, and correct the image distortion caused by shooting angle or distance by denoising, enhancing and correcting the image taken by the 3D camera, so that the image is clearer and more accurate, thereby facilitating subsequent detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is the work flow chart of the present invention. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] See also Figure 1 , this embodiment provides a technical solution: a 3D deep learning defect detection method based on machine vision, comprising the following steps: S1 builds a learning network: collects and organizes image data sets to ensure data diversity and representativeness, then inputs the collected and organized image data sets into the training to test the deep learning model architecture, then extracts features from the image data set and saves the data, then trains the model after extraction and preservation, and tests it after training. After the test is completed, the learning network construction is completed; After collecting and organizing the image data set, it needs to be preprocessed, such as resizing and format conversion, to meet the input requirements of the network; After collecting and organizing the image data set, it is necessary to perform data enhancement on the collected and organized image data set, such as performing random cropping and rotation operations on the image to increase the scale and diversity of the data set; Testing the deep learning model architecture after training You can choose convolutional neural network (CNN), recurrent neural network (RNN) or a combination of the two. The convolution layer in the deep learning model architecture after training extracts local image features in the image dataset, and the pooling layer reduces the dimension of the features and retains important information. By stacking multiple convolutional layers and pooling layers, the network can gradually extract high-level features of the image; During model training, the prepared data set is divided into a training set and a test set. The model is trained using the training set, and the weights and biases of the model are updated through the back-propagation algorithm. During the training process, the cross entropy loss function can be used to measure the model error, and optimizers such as Adam can be used to accelerate the training process. When testing the model, use the test set to test the trained model, evaluate the performance of the model, and analyze the advantages and disadvantages of the model by calculating the accuracy, recall and other indicators of the model on the test set; S2 Image Acquisition: Use a high-resolution 3D camera to fully scan the product surface and obtain the three-dimensional image data of the product: S3 data preprocessing: preprocess the collected images, including denoising, enhancement, correction, etc., to improve the accuracy of subsequent processing; Denoising requires the use of methods such as mean filtering, median filtering or Gaussian filtering to remove random noise in the image and improve the signal-to-noise ratio of the image; Enhancement requires adjusting brightness, contrast or using histogram equalization to enhance the visual effect of the image, make the image features more obvious, and improve the robustness of the recognition system; Correction requires correction of image distortion caused by shooting angle or distance through geometric transformation (such as rotation, scaling, cropping, etc.) to ensure that the image can still be correctly recognized at different angles and sizes; S4 Feature Extraction: Extract high-level features from images using deep learning networks; S5 defect detection: The extracted features are input into the learning network for comparison with the deep learning model architecture within the learning network to perform defect detection and identification. Unsaved defect data will be automatically uploaded to the deep learning network for storage. S6 Data Analysis and Reporting: Perform statistical analysis on the test results, generate a test report, and then display it through visualization technology. The visualization technology can be t-SNE technology.

[0019] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A 3D deep learning defect detection method based on machine vision, characterized by: The following steps are involved: S1 builds a learning network: collects and organizes image data sets to ensure data diversity and representativeness, then inputs the collected and organized image data sets into the training to test the deep learning model architecture, then extracts features from the image data set and saves the data, then trains the model after extraction and preservation, and tests it after training. After the test is completed, the learning network construction is completed; S2 Image Acquisition: Use a high-resolution 3D camera to fully scan the product surface and obtain the three-dimensional image data of the product: S3 data preprocessing: preprocess the collected images, including denoising, enhancement, correction, etc., to improve the accuracy of subsequent processing; S4 Feature Extraction: Extract high-level features from images using deep learning networks; S5 defect detection: The extracted features are input into the learning network for comparison with the deep learning model architecture within the learning network to perform defect detection and identification. Unsaved defect data will be automatically uploaded to the deep learning network for storage. S6 Data Analysis and Reporting: Perform statistical analysis on the test results, generate a test report, and then display it through visualization technology.

2. The 3D deep learning defect detection method based on machine vision according to claim 1, characterized in that: In step 1, after the image data set is collected and organized, it needs to be preprocessed, such as resizing, format conversion, etc., to meet the input requirements of the network.

3. The 3D deep learning defect detection method based on machine vision according to claim 1, characterized in that: In step 1, after collecting and organizing the image data set, it is necessary to perform data enhancement on the collected and organized image data set, such as performing random cropping and rotation operations on the images, to increase the scale and diversity of the data set.

4. The 3D deep learning defect detection method based on machine vision according to claim 1, characterized in that: In the step 1, the deep learning model architecture is tested after training, and a convolutional neural network (CNN), a recurrent neural network (RNN), or a combination of the two can be selected. The convolution layer in the deep learning model architecture is tested after training to extract local features of the image in the image data set, and the pooling layer reduces the dimension of the features and retains important information. By stacking multiple convolutional layers and pooling layers, the network can gradually extract high-level features of the image.

5. The 3D deep learning defect detection method based on machine vision according to claim 1, characterized in that: In step 1, the prepared data set is divided into a training set and a test set during model training. The model is trained using the training set, and the weights and biases of the model are updated using a back propagation algorithm. During the training process, a cross entropy loss function can be used to measure the error of the model, and optimizers such as Adam can be used to accelerate the training process.

6. The 3D deep learning defect detection method based on machine vision according to claim 1, characterized in that: In step 1, the model is tested using a test set to test the trained model, evaluate the performance of the model, and analyze the advantages and disadvantages of the model by calculating indicators such as the accuracy and recall rate of the model on the test set.

7. The 3D deep learning defect detection method based on machine vision according to claim 1, characterized in that: In step 3, the denoising needs to adopt methods such as mean filtering, median filtering or Gaussian filtering to remove random noise in the image and improve the signal-to-noise ratio of the image.

8. The 3D deep learning defect detection method based on machine vision according to claim 1, characterized in that: The enhancement in step 3 needs to be performed by adjusting the brightness, contrast or using histogram equalization to enhance the visual effect of the image, make the image features more obvious, and improve the robustness of the recognition system.

9. The 3D deep learning defect detection method based on machine vision according to claim 1, characterized in that: The correction in step 3 needs to be corrected for image distortion caused by shooting angle or distance through geometric transformation (such as rotation, scaling, cropping, etc.) to ensure that the image can still be correctly recognized at different angles and sizes.

10. The 3D deep learning defect detection method based on machine vision according to claim 1, characterized in that: The visualization technology in step 6 may be selected as t-SNE technology.

Images