Rapid defect marking, training and image screening method and system for stone curtain wall millimeter wave image

Abstract

The invention provides a stone curtain wall millimeter wave image rapid defect marking, training and image screening method and system, and the system comprises an image display module which is used for displaying a collected stone curtain wall millimeter wave image; the automatic labeling module is used for carrying out preliminary defect labeling on the input millimeter wave image; the man-machine interaction module provides a user interface to allow a user to check, audit and correct the labeling result; the database result output module is used for outputting the final labeling result in a standard format and supporting multi-dimensional screening and exporting functions; the defect identification model is used for training and identifying defects in the stone curtain wall millimeter wave image based on a deep learning model; and the image screening module is used for evaluating the quality of the image and screening out a high-quality image according to a set threshold value based on a screening algorithm of image quality. Through automatic and intelligent technical means, the stone curtain wall detection efficiency and accuracy are improved, and the safety and durability of the building curtain wall are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of building safety detection, and in particular to a method and system for rapid defect marking, training and image screening of millimeter wave images of stone curtain walls. Background Art

[0002] With the widespread application of building curtain walls, stone curtain walls, as an important component of them, are increasingly valued for their safety and durability; traditional stone curtain wall inspection methods mainly include manual visual inspection, tapping and listening methods, etc.

[0003] Deficiencies of existing technology: 1. High dependence on manual labor: The methods of manual detection, manual classification, and manual screening are inefficient and prone to errors.

[0004] 2. Low labeling efficiency: There is a lack of automated defect labeling tools, which requires a lot of manual labeling, which is time-consuming and labor-intensive.

[0005] 3. Inaccurate image screening: There is a lack of effective image screening algorithms, which makes it difficult to automatically mark defective images, or to batch select problematic curtain wall images for screening, viewing and training.

[0006] 4. Low data utilization: In existing methods, annotated image data is often not effectively managed and utilized, and a closed loop of continuous improvement cannot be formed.

[0007] Therefore, the prior art has deficiencies and needs further improvement. Summary of the invention

[0008] In view of the problems existing in the prior art, the present invention provides a method and system for rapid defect marking, training and image screening of millimeter wave images of stone curtain walls.

[0009] To achieve the above object, the specific scheme of the present invention is as follows: The present invention provides a rapid defect marking, training and image screening system for millimeter wave images of stone curtain walls, the system comprising: Image display module, used to display the collected millimeter wave images of stone curtain walls, providing users with intuitive visual feedback, making it easier for users to observe and analyze defects in the images; The automatic annotation module performs preliminary defect annotation on the input millimeter wave image based on the deep learning algorithm, identifies various types of defects in the image, including cracks, detachments, and holes, and annotates them. The preliminary results will be stored in the database; the automatic annotation module includes a deep learning model based on a convolutional neural network, which is trained with a pre-trained data set to identify and annotate defect areas in the image; The human-computer interaction module provides a user interface to allow users to view, review and modify the annotation results; Database result output module, used to output the final annotation results in a standard format, and supports multi-dimensional screening and export functions; Defect recognition model, based on deep learning model, including input layer, convolution layer, pooling layer, fully connected layer and output layer, used to train and identify defects in millimeter wave images of stone curtain walls; The image screening module is based on the image quality screening algorithm, which is used to evaluate the quality of the image and screen out high-quality images according to the set threshold.

[0010] Furthermore, the automatic labeling module adopts transfer learning technology and is initialized using a pre-trained model to speed up training and improve model performance.

[0011] Furthermore, the human-computer interaction module includes a manual annotation tool, allowing the user to modify the automatic annotation results.

[0012] Furthermore, the database result output module supports multi-dimensional screening, including screening by time, defect type, and scan type.

[0013] Furthermore, the image screening module includes an image quality assessment submodule for extracting key quality parameters of the image including clarity and noise level, and scoring based on these parameters.

[0014] The present invention also provides a method for rapid defect annotation, training and image screening of millimeter wave images of stone curtain walls. Based on the above system, the method comprises the following steps: S1, Image acquisition: Use high-resolution millimeter wave imaging equipment, set the working frequency suitable for stone curtain wall detection, ensure the clarity and penetration of the image, fix the imaging equipment on a stable bracket, perform a comprehensive scan of the stone curtain wall along the preset path, obtain its millimeter wave image data, and record the environmental conditions during the scan; S2, defect annotation: construct a millimeter wave image dataset containing various defect types, and manually annotate these images to form an annotated dataset; S3, model training: divide the labeled millimeter wave image dataset into a training set and a test set, and use the training set to train the defect detection model; set the corresponding hyperparameters, including batch size, learning rate, and number of iterations, to ensure the stability and efficiency of the training process; S4, image screening: screening based on image quality, excluding low-quality images; S5, Result Output: Output the defect annotation results and the filtered image data in a unified format. The annotation results include the location, type and size of the defects, presented in text or graphic form; the filtered image data is saved in the form of image files for subsequent analysis and processing; generate a detailed inspection report, summarizing the key information, problems found and recommended repair measures during the inspection process; S6, Field inspection and database management: Conduct field inspection on the stone curtain wall corresponding to the millimeter wave image, peel the curtain wall from the building, and check the defects of the hanging parts behind the curtain wall; if the actual on-site situation is consistent with the identification of the defect recognition model, add this image to the system database; perform database screening, check all images added to the database, and filter by time, defect type, and scan type, and export the selected database data in batches for further analysis and training.

[0015] Furthermore, in step S2, in the defect labeling step, transfer learning technology is used to initialize using a pre-trained model to speed up training and improve model performance; and a manual labeling tool is provided to allow users to correct automatic labeling results.

[0016] Furthermore, in step S3, data enhancement technology is used in the training process to increase the diversity of training samples and further improve the generalization ability of the model.

[0017] Furthermore, in step S3, based on the annotated training set, a deep convolutional neural network model is trained as a defect annotation model. The model adopts a residual network as the basic architecture and introduces an attention mechanism to enhance the model's attention to defect areas. The cross entropy loss function is used as the optimization target, and the Adam optimizer is used for parameter update. After multiple rounds of iterative training, the model can automatically identify and annotate defect areas in newly collected millimeter wave images of stone curtain walls. At the same time, a manual annotation function is provided to facilitate users to annotate or correct images according to actual on-site conditions.

[0018] Further, in step S4, the image quality screening method includes: extracting quality parameters of the image, the quality parameters including clarity and noise level, and designing a scoring mechanism based on these quality parameters to score the image; setting a threshold according to the scoring result, and screening out high-quality images above the threshold.

[0019] The technical solution of the present invention has the following beneficial effects: 1. Improve defect annotation efficiency: Combination of automatic labeling and human-computer interaction: The combination of the automatic labeling module and the human-computer interaction module greatly reduces the time and workload of manual labeling; the automatic labeling module can preliminarily identify and label defects, and users only need to perform a small amount of review and correction, which significantly improves labeling efficiency.

[0020] Manual annotation tool: A manual annotation tool is provided to allow users to correct automatic annotation results to ensure the accuracy of annotation.

[0021] 2. Improve model recognition accuracy: Deep learning algorithm: The deep learning model based on convolutional neural network (CNN) can effectively identify and annotate various defects in millimeter wave images.

[0022] Transfer learning technology: Using pre-trained models for initialization accelerates the model training process and improves the performance and generalization ability of the model.

[0023] Data enhancement technology: Through data enhancement technology (such as random rotation, scaling, cropping, and adding noise, etc.), the diversity of training samples is increased, further improving the recognition accuracy of the model.

[0024] 3. Achieve efficient image screening: Image quality assessment: A screening algorithm based on image quality is designed, which can evaluate key quality parameters such as image clarity and noise level, and set appropriate thresholds based on the scoring results to automatically screen out high-quality images.

[0025] Multi-dimensional screening function: The database result output module supports multi-dimensional screening by time, defect type, scan type and other conditions, making it easier for users to manage and use annotation data.

[0026] 4. Reduce testing costs: Reduce manual intervention: Through automated and intelligent technical means, the reliance on manual labor is reduced and the labor cost of detection is reduced.

[0027] Improve detection efficiency: The automated process greatly shortens the detection time and improves the overall detection efficiency, thereby reducing the overall cost of detection.

[0028] 5. Improve the reliability and accuracy of test results: On-site inspection and verification: On-site inspection of the stone curtain walls corresponding to some millimeter-wave images was carried out to ensure that the recognition results of the model were consistent with the actual situation, thereby improving the reliability of the detection results.

[0029] Detailed test report: Generate a detailed test report that summarizes key information, problems found, and recommended repairs during the test to help users better understand and use the test results.

[0030] 6. Data management and reuse: Database management: A database management system has been established to view, review and correct annotation results, and supports multi-dimensional screening and batch export to facilitate data management and reuse.

[0031] Continuous improvement: By using the filtered high-quality image data for further model training, a closed loop of continuous improvement is formed to continuously improve the performance of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is a logic diagram of the present invention; Figure 2 It is the annotation interface of the millimeter wave image of the stone curtain wall; Figure 3 A database screening interface for millimeter wave images of stone curtain walls; Figure 4 It is a principle block diagram of the present invention; Figure 5 It is the overall flow chart of the present invention. DETAILED DESCRIPTION

[0033] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments; it can be understood that the specific embodiments described herein are only used to explain the present invention rather than to limit the present invention; it should also be noted that, for ease of description, only the parts related to the present invention rather than all are shown in the accompanying drawings.

[0034] Combination Figure 1-Figure 5 As shown, the present invention provides a method and system for rapid defect annotation, training and image screening of millimeter wave images of stone curtain walls, the system comprising: Image display module, used to display the collected millimeter wave images of stone curtain walls, providing users with intuitive visual feedback, making it easier for users to observe and analyze defects in the images; The automatic annotation module performs preliminary defect annotation on the input millimeter wave image based on the deep learning algorithm, identifies various types of defects in the image, including cracks, detachments, and holes, and annotates them. The preliminary results will be stored in the database; the automatic annotation module includes a deep learning model based on a convolutional neural network, which is trained with a pre-trained data set to identify and annotate defect areas in the image; The human-computer interaction module provides a user interface to allow users to view, review and modify the annotation results; Database result output module, used to output the final annotation results in a standard format, and supports multi-dimensional screening and export functions; Defect recognition model, based on deep learning model, including input layer, convolution layer, pooling layer, fully connected layer and output layer, used to train and identify defects in millimeter wave images of stone curtain walls; The image screening module is based on the image quality screening algorithm, which is used to evaluate the quality of the image and screen out high-quality images according to the set threshold.

[0035] The automatic labeling module adopts transfer learning technology and is initialized using a pre-trained model to speed up training and improve model performance.

[0036] The human-computer interaction module includes a manual annotation tool, allowing the user to modify the automatic annotation results.

[0037] The database result output module supports multi-dimensional screening, including screening by time, defect type, and scan type.

[0038] The image screening module includes an image quality assessment submodule for extracting key quality parameters of the image including clarity and noise level, and scoring based on these parameters.

[0039] The present invention also provides a method for rapid defect annotation, training and image screening of millimeter wave images of stone curtain walls. Based on the above system, the method comprises the following steps: S1, Image acquisition: Use high-resolution millimeter wave imaging equipment, set the working frequency suitable for stone curtain wall detection, ensure the clarity and penetration of the image, fix the imaging equipment on a stable bracket, perform a comprehensive scan of the stone curtain wall along the preset path, obtain its millimeter wave image data, and record the environmental conditions during the scan; S2, defect annotation: construct a millimeter wave image dataset containing various defect types, and manually annotate these images to form an annotated dataset; S3, model training: divide the labeled millimeter wave image dataset into a training set and a test set, and use the training set to train the defect detection model; set the corresponding hyperparameters, including batch size, learning rate, and number of iterations, to ensure the stability and efficiency of the training process; S4, image screening: screening based on image quality, excluding low-quality images; S5, Result Output: Output the defect annotation results and the filtered image data in a unified format. The annotation results include the location, type and size of the defects, presented in text or graphic form; the filtered image data is saved in the form of image files for subsequent analysis and processing; generate a detailed inspection report, summarizing the key information, problems found and recommended repair measures during the inspection process; S6, Field inspection and database management: Conduct field inspection on the stone curtain wall corresponding to the millimeter wave image, peel the curtain wall from the building, and check the defects of the hanging parts behind the curtain wall; if the actual on-site situation is consistent with the identification of the defect recognition model, add this image to the system database; perform database screening, check all images added to the database, and filter by time, defect type, and scan type, and export the selected database data in batches for further analysis and training.

[0040] In step S2, the defect labeling step, transfer learning technology is used to initialize the pre-trained model to speed up the training and improve the model performance; and a manual labeling tool is provided to allow the user to correct the automatic labeling results.

[0041] In step S3, data enhancement technology is used in the training process to increase the diversity of training samples and further improve the generalization ability of the model.

[0042] In step S3, based on the annotated training set, a deep convolutional neural network model is trained as a defect annotation model. The model uses a residual network as the basic architecture and introduces an attention mechanism to enhance the model's attention to defect areas. The cross entropy loss function is used as the optimization target, and the Adam optimizer is used for parameter update. After multiple rounds of iterative training, the model can automatically identify and annotate defect areas in newly collected millimeter wave images of stone curtain walls. At the same time, a manual annotation function is provided to facilitate users to annotate or correct images according to actual on-site conditions.

[0043] In step S4, the image quality screening method includes: extracting quality parameters of the image, the quality parameters include clarity and noise level, and designing a scoring mechanism based on these quality parameters to score the image; setting a threshold according to the scoring result, and screening out high-quality images above the threshold.

[0044] Working principle of the present invention: The present invention provides a method and system for rapid defect annotation, training and image screening of millimeter wave images of stone curtain walls, and its working principle mainly includes the following key steps: Image acquisition: Equipment selection: Use high-resolution millimeter wave imaging equipment and set the working frequency suitable for stone curtain wall detection to ensure image clarity and penetration.

[0045] Device fixation: Fix the imaging device on a stable bracket to ensure stability during scanning.

[0046] Full scan: Perform a full scan of the stone curtain wall along the preset path to obtain its millimeter wave image data.

[0047] Environmental Record: Record the environmental conditions during scanning, such as temperature, humidity, etc., to ensure consistency in image quality.

[0048] Defect annotation: Dataset construction: Build a millimeter-wave image dataset containing various defect types, and manually annotate these images to form an annotated dataset.

[0049] Model training: Based on the labeled dataset, a deep convolutional neural network (CNN) model is trained as the defect labeling model. The model uses the residual network (ResNet) as the basic architecture and introduces the attention mechanism to enhance the model's attention to the defect area.

[0050] Automatic annotation: Use the trained CNN model to perform preliminary defect annotation on newly acquired millimeter wave images.

[0051] Human-computer interaction: Through the human-computer interaction module, a user interface is provided, allowing users to view, review and correct the automatic annotation results to ensure the accuracy of the annotation.

[0052] Manual annotation: Provides manual annotation tools, allowing users to annotate or correct images based on actual site conditions.

[0053] Model training: Dataset division: The labeled millimeter-wave image dataset is divided into a training set and a test set, usually in a ratio of 8:2.

[0054] Data enhancement: Use data enhancement techniques (such as random rotation, scaling, cropping, and adding noise) to increase the diversity of training samples and improve the generalization ability of the model.

[0055] Parameter setting: Set appropriate hyperparameters such as batch size, learning rate, and number of iterations to ensure the stability and efficiency of the training process.

[0056] Model optimization: Optimize model parameters through the back-propagation algorithm to improve the recognition accuracy of the model.

[0057] Image Filter: Quality assessment: Design a screening method based on image quality to extract key quality parameters such as clarity and noise level of each mmWave image.

[0058] Scoring mechanism: Based on these parameters, a scoring mechanism is designed to score the images. During the scoring process, the weight of each parameter can be adjusted according to the actual situation.

[0059] Threshold setting: Set an appropriate threshold based on the scoring results to screen the images. Images below the threshold are considered to be of poor quality and will be excluded from subsequent analysis.

[0060] High-quality images: The filtered images will reflect the actual situation of the stone curtain wall more clearly and accurately.

[0061] Result output: Marking results: Output the defect marking results in a unified format. The marking results include information such as the location, type and size of the defect, and can be presented in text or graphic form.

[0062] Image file: The screened image data is saved in the form of an image file for subsequent analysis and processing.

[0063] Test Report: Generate a detailed test report that summarizes key information, problems found, and recommended repairs during the test process so that users can better understand and use the test results.

[0064] On-site inspection and database management: On-site inspection: Conduct on-site inspection on the stone curtain wall corresponding to some millimeter wave images, peel the curtain wall from the building, and check the defects of the hanging parts behind the curtain wall.

[0065] Database update: If the actual on-site situation is consistent with the identification of the defect recognition model, the image will be added to the system database.

[0066] Database filtering: Provides database filtering function, which can view all images added to the database and provide filtering methods such as time, defect type, scan type, etc.

[0067] Data export: Export selected database data in batches for further analysis and training.

[0068] Workflow Diagram Image acquisition → 2. Defect annotation → 3. Model training → 4. Image screening → 5. Result output → 6. On-site inspection and database management Through the above process, the present invention realizes the full process automation from image acquisition to defect annotation, model training, image screening, and then to result output and field verification, which significantly improves the efficiency and accuracy of stone curtain wall detection.

[0069] Embodiment 1: Provide a rapid defect annotation, training and image screening system for millimeter wave images of stone curtain walls: 1. Rapid defect annotation system: System architecture: includes image display module, automatic annotation module, human-computer interaction module and database result output module.

[0070] Automatic labeling module: uses deep learning algorithms to perform preliminary defect labeling on the input millimeter wave images and add them to the database.

[0071] Human-computer interaction module: provides a user interface that allows users to view, review and modify the database.

[0072] Database result output module: The final database annotation results are output in a standard format, which can perform multi-dimensional screening and export functions for subsequent model training.

[0073] 2. Defect identification model: Model structure: CNN-based deep learning model, including input layer, convolution layer, pooling layer, fully connected layer and output layer.

[0074] Training process: Use the labeled millimeter-wave image dataset for training and optimize the model parameters through the back-propagation algorithm.

[0075] Transfer learning: Use pre-trained models for initialization to speed up the training process and improve model performance.

[0076] 3. Image screening: Algorithm process: Receive the output results of the defect recognition model and filter the images according to the set filtering threshold.

[0077] Screening threshold setting: Set appropriate thresholds according to actual needs to ensure the accuracy of screening results.

[0078] Image quality assessment: Perform quality assessment on the screened images and eliminate low-quality images.

[0079] First, after collecting the millimeter wave image of the stone curtain wall, the defect recognition model is used to identify the curtain wall millimeter wave image to determine the defect type and location. Then, an on-site inspection is carried out on the stone curtain wall corresponding to some millimeter wave images. The curtain wall is peeled off from the building and the defects of the hanging parts behind the curtain wall are checked. If the actual on-site situation is consistent with the identification of the defect recognition model, the image is added to the system database.

[0080] The system provides a database screening function, which can view all images added to the database, and provide methods such as time, defect type, scan type, etc. for screening, and can evaluate image quality through image screening algorithms. The selected database data can be exported in batches.

[0081] The exported database data is input into the defect recognition model for training and migration.

[0082] Based on the above technical solution, compared with the existing method, it has the following advantages: 1. Improve defect marking efficiency: By combining automatic marking with human-computer interaction, the defect marking time can be greatly shortened.

[0083] 2. Improve model recognition accuracy: Use deep learning algorithms and transfer learning technology to improve the recognition accuracy and generalization ability of defect recognition models.

[0084] 3. Achieve efficient image screening: By designing an image screening algorithm and introducing the concept of a database, rapid screening and classification of millimeter-wave images can be achieved.

[0085] 4. Reduce testing costs: reduce manual intervention and testing time, and reduce the overall cost of stone curtain wall testing.

[0086] The steps include: 1. Image Acquisition During the specific implementation, a high-resolution millimeter-wave imaging device is first selected, and its operating frequency is set to a frequency band suitable for stone curtain wall detection to ensure the clarity and penetration of the image; the imaging device is fixed on a stable bracket to ensure stability during the scanning process; during the scanning process, the device conducts a comprehensive scan of the stone curtain wall along a preset path to obtain its millimeter-wave image data; at the same time, the environmental conditions during the scan, such as temperature, humidity, etc., are recorded, because these factors may affect the image quality.

[0087] 2. Defect annotation In the defect annotation stage, we first build a millimeter wave image dataset containing various defect types (such as cracks, detachment, holes, etc.), and manually annotate these images to form an annotated dataset. Then, based on the annotated dataset, we train a deep convolutional neural network (CNN) model as a defect annotation model. The model uses the residual network (ResNet) as the basic architecture and introduces an attention mechanism to enhance the model's attention to defect areas. During the training process, the cross entropy loss function is used as the optimization target, and the Adam optimizer is used for parameter update. After multiple rounds of iterative training, the model can automatically identify and annotate defect areas in newly collected millimeter wave images of stone curtain walls. A manual annotation function is provided, and images can be annotated as database files or misjudgment marks according to the actual situation on site.

[0088] 3. Model Training During the model training phase, the labeled millimeter wave image dataset is divided into a training set and a test set, usually in a ratio of 8:2. The training set is used to train the defect detection model, while the test set is used to evaluate the performance of the model. During the training process, data enhancement techniques such as random rotation, scaling, cropping, and adding noise are used to increase the diversity of training samples and improve the generalization ability of the model. At the same time, appropriate hyperparameters such as batch size, learning rate, and number of iterations are set to ensure the stability and efficiency of the training process. After training, the model can achieve a high detection accuracy on the test set.

[0089] 4. Image screening In the image screening stage, a screening method based on image quality is designed, and a database screening interface is provided. First, the key quality parameters such as clarity and noise level of each millimeter wave image are extracted. Then, a scoring mechanism is designed based on these parameters to score each image. During the scoring process, the weight of each parameter can be adjusted according to the actual situation to reflect the importance of different parameters to image quality. Finally, a suitable threshold is set according to the scoring results to screen the images. Images below the threshold are considered to be of poor quality and will be excluded from subsequent analysis. The screened images will reflect the actual situation of the stone curtain wall more clearly and accurately.

[0090] 5. Result Output Finally, the defect annotation results and the filtered image data are output in a unified format. The annotation results include information such as the location, type and size of the defect, which can be presented in text or graphics. The filtered image data is saved in the form of an image file for subsequent analysis and processing. At the same time, a detailed inspection report can also be generated to summarize the key information, problems found and recommended repair measures during the inspection process, so that users can better understand and use the inspection results.

[0091] Embodiment 2: Image acquisition: Equipment selection: Choose a high-resolution millimeter-wave imaging device and set the operating frequency to 24 GHz to ensure image clarity and penetration.

[0092] Device fixation: Fix the imaging device on a stable bracket to ensure stability during scanning.

[0093] Full scan: Perform a full scan of the stone curtain wall along the preset path to obtain its millimeter wave image data.

[0094] Environmental records: Record the environmental conditions during scanning, such as temperature 25°C and humidity 50%, to ensure consistency in image quality.

[0095] Initial defect annotation: Dataset construction: Build a millimeter-wave image dataset containing various defect types, including cracks, detachment, holes, etc., and manually annotate these images to form an annotated dataset.

[0096] Model training: Based on the labeled dataset, a deep convolutional neural network (CNN) model is trained. The model uses the residual network (ResNet) as the basic architecture and introduces an attention mechanism to enhance the model's attention to the defect area.

[0097] Automatic annotation: Use the trained CNN model to perform preliminary defect annotation on newly acquired millimeter wave images.

[0098] Human-computer interaction: The human-computer interaction module provides a user interface that allows inspectors to view, review and correct automatic labeling results to ensure the accuracy of the labeling.

[0099] Manual annotation: For some complex defects, inspectors use manual annotation tools to make corrections.

[0100] Through the above steps, the millimeter wave image of the stone curtain wall was successfully collected and the initial defect annotation was carried out. The inspection personnel reviewed and corrected the annotation results to ensure the accuracy of the annotation.

[0101] Dataset division: Data preparation: The labeled millimeter wave image dataset is divided into a training set and a test set in a ratio of 8:2.

[0102] Data enhancement: Use data enhancement techniques (such as random rotation, scaling, cropping, and adding noise) to increase the diversity of training samples and improve the generalization ability of the model.

[0103] Model training: Parameter setting: Set an appropriate batch size (such as 32), learning rate (such as 0.001), and number of iterations (such as 1000) to ensure the stability and efficiency of the training process.

[0104] Model optimization: Optimize model parameters through the back-propagation algorithm to improve the recognition accuracy of the model.

[0105] Image Filter: Quality assessment: Design a screening method based on image quality to extract key quality parameters such as clarity and noise level of each mmWave image.

[0106] Scoring mechanism: Based on these parameters, a scoring mechanism is designed to score the images. During the scoring process, the weight of each parameter can be adjusted according to the actual situation.

[0107] Threshold setting: Set an appropriate threshold based on the scoring results to screen the images. Images below the threshold are considered to be of poor quality and will be excluded from subsequent analysis.

[0108] High-quality images: The filtered images will reflect the actual situation of the stone curtain wall more clearly and accurately.

[0109] Through the above steps, a high-performance defect recognition model was successfully trained and the images were effectively screened. The high-quality images after screening were used for further analysis and training, which improved the performance and generalization ability of the model.

[0110] Result output: Marking results: Output the defect marking results in a unified format. The marking results include information such as the location, type and size of the defect, and can be presented in text or graphic form.

[0111] Image file: The screened image data is saved in the form of an image file for subsequent analysis and processing.

[0112] Test Report: Generate a detailed test report that summarizes key information, problems found, and recommended repairs during the test process so that users can better understand and use the test results.

[0113] On-site inspection: Select areas: Based on the marked results in the test report, select some areas with suspected defects for on-site inspection.

[0114] On-site inspection: Peel off the selected stone curtain wall from the building and check for defects in the hanging parts behind the curtain wall.

[0115] Database update: If the actual on-site situation is consistent with the identification of the defect recognition model, the image will be added to the system database.

[0116] Database filtering: Provides database filtering function, which can view all images added to the database and provide filtering methods such as time, defect type, scan type, etc.

[0117] Data export: Export selected database data in batches for further analysis and training.

[0118] Through the above steps, a detailed inspection report was successfully generated, and some suspected defect areas were inspected on site. The actual on-site situation was basically consistent with the identification of the defect recognition model, which verified the accuracy and reliability of the model. At the same time, the database was updated to provide more data support for future inspections.

[0119] The present invention demonstrates the full process automation from image acquisition, defect annotation, model training, image screening to result output and field inspection. These examples not only verify the effectiveness and practicality of the present invention, but also provide specific guidance and reference for practical application.

[0120] The above description is only a preferred embodiment of the present invention, and does not limit the scope of the present invention. All equivalent structural changes made by using the contents of the present invention specification and drawings under the inventive concept of the present invention, or directly / indirectly applied in other related technical fields are included in the protection scope of the present invention.

Claims

1. A rapid defect annotation, training and image screening system for millimeter wave images of stone curtain walls, characterized in that: The system includes: Image display module, used to display the collected millimeter wave images of stone curtain walls, providing users with intuitive visual feedback, making it easier for users to observe and analyze defects in the images; The automatic annotation module performs preliminary defect annotation on the input millimeter wave image based on the deep learning algorithm, identifies various types of defects in the image, including cracks, detachments, and holes, and annotates them. The preliminary results are stored in the database. The automatic annotation module includes a deep learning model based on a convolutional neural network, which is trained with a pre-trained data set to identify and annotate defect areas in the image. The human-computer interaction module provides a user interface to allow users to view, review and modify the annotation results; Database result output module, used to output the final annotation results in a standard format, and supports multi-dimensional screening and export functions; Defect recognition model, based on deep learning model, including input layer, convolution layer, pooling layer, fully connected layer and output layer, used to train and identify defects in millimeter wave images of stone curtain walls; The image screening module is based on the image quality screening algorithm, which is used to evaluate the quality of the image and screen out high-quality images according to the set threshold.

2. The system according to claim 1, characterized in that The automatic labeling module adopts transfer learning technology and is initialized using a pre-trained model to speed up training and improve model performance.

3. The system according to claim 1, characterized in that The human-computer interaction module includes a manual annotation tool, allowing the user to modify the automatic annotation results.

4. The system according to claim 1, characterized in that The database result output module supports multi-dimensional screening, including screening by time, defect type, and scan type.

5. The system according to claim 1, characterized in that The image screening module includes an image quality assessment submodule for extracting key quality parameters of the image including clarity and noise level, and scoring based on these parameters.

6. A method for rapid defect annotation, training and image screening of millimeter wave images of stone curtain walls, based on the system according to any one of claims 1 to 5, characterized in that: The method comprises the following steps: S1, Image acquisition: Use high-resolution millimeter wave imaging equipment, set the working frequency suitable for stone curtain wall detection, ensure the clarity and penetration of the image, fix the imaging equipment on a stable bracket, perform a comprehensive scan of the stone curtain wall along the preset path, obtain its millimeter wave image data, and record the environmental conditions during the scan; S2, defect annotation: construct a millimeter wave image dataset containing various defect types, and manually annotate these images to form an annotated dataset; S3, model training: divide the labeled millimeter wave image dataset into a training set and a test set, and use the training set to train the defect detection model; set the corresponding hyperparameters, including batch size, learning rate, and number of iterations, to ensure the stability and efficiency of the training process; S4, image screening: screening based on image quality, excluding low-quality images; S5, Result Output: Output the defect annotation results and the filtered image data in a unified format. The annotation results include the location, type and size of the defects, presented in text or graphic form; the filtered image data is saved in the form of image files for subsequent analysis and processing; generate a detailed inspection report, summarizing the key information, problems found and recommended repair measures during the inspection process; S6, Field inspection and database management: Conduct field inspection on the stone curtain wall corresponding to the millimeter wave image, peel the curtain wall from the building, and check the defects of the hanging parts behind the curtain wall; if the actual on-site situation is consistent with the identification of the defect recognition model, add this image to the system database; perform database screening, check all images added to the database, and filter by time, defect type, and scan type, and export the selected database data in batches for further analysis and training.

7. The method according to claim 6, characterized in that In step S2, the defect labeling step, transfer learning technology is used to initialize the pre-trained model to speed up the training and improve the model performance; and a manual labeling tool is provided to allow the user to correct the automatic labeling results.

8. The method according to claim 6, further characterized in that: In step S3, data enhancement technology is used in the training process to increase the diversity of training samples and further improve the generalization ability of the model.

9. The method according to claim 6, further characterized in that: In step S3, based on the annotated training set, a deep convolutional neural network model is trained as a defect annotation model. The model uses a residual network as the basic architecture and introduces an attention mechanism to enhance the model's attention to defect areas. The cross entropy loss function is used as the optimization target, and the Adam optimizer is used for parameter update. After multiple rounds of iterative training, the model can automatically identify and annotate defect areas in newly collected millimeter wave images of stone curtain walls. At the same time, a manual annotation function is provided to facilitate users to annotate or correct images according to actual on-site conditions.

10. The method according to claim 6, further characterized in that: In step S4, the image quality screening method includes: extracting quality parameters of the image, the quality parameters include clarity and noise level, and designing a scoring mechanism based on these quality parameters to score the image; setting a threshold according to the scoring result, and screening out high-quality images above the threshold.

Images