Measurement method, host computer, measurement system and storage medium

Abstract

This application discloses a measurement method, a host computer, a measurement system, and a storage medium, relating to the field of industrial measurement technology. The method includes: in response to a point-taking operation triggered when measuring a material to be measured, acquiring a material image of the material to be measured and obtaining the first coordinates of the point-taking operation in the material image; inferring the material image using a pre-trained annotation model on an associated cloud server to obtain at least one candidate feature point and its second coordinates in the material image; determining a target feature point from the candidate feature points based on the spatial relationship between the first coordinates and the second coordinates of the candidate feature points; and determining the second coordinates of the target feature point as annotation data for the material image and displaying the annotation data. This application can improve the efficiency of industrial measurement.

Description

Technical Field

[0001] This application relates to the field of industrial measurement technology, and in particular to measurement methods, host computers, measurement systems and storage media. Background Technology

[0002] In the field of intelligent manufacturing, AI-based visual dimensional measurement technology has gradually replaced traditional manual measurement, achieving preliminary automated analysis. However, the overall efficiency of existing technologies still heavily relies on model training methods that use manually labeled data. Currently, the measurement process and the data labeling process on images are separate: measurement engineers must manually label a large number of material images before model training can begin. This means that during actual measurement operations, measurement and data labeling cannot be performed simultaneously. This reliance on manual labeling, resulting in situations where measurement is not performed while labeling is being done, or vice versa, hinders the efficiency of industrial measurement. Summary of the Invention

[0003] The main purpose of this application is to provide a measurement method, host computer, measurement system and storage medium, which aims to solve the technical problem of low efficiency in industrial measurement due to reliance on manual annotation in existing dimensional measurement.

[0004] To achieve the above objectives, this application proposes a measurement method, the measurement method comprising:

[0005] In response to a sampling operation triggered when measuring the material to be measured, a material image of the material to be measured is acquired, and the first coordinates of the sampling operation in the material image are acquired.

[0006] The material image is inferred by a pre-trained annotation model on the associated cloud server to obtain at least one candidate feature point and its second coordinates in the material image;

[0007] Based on the spatial relationship between the first coordinate and the second coordinate of the candidate feature point, the target feature point is determined from the candidate feature points;

[0008] The second coordinates of the target feature point are determined as the annotation data of the material image, and the annotation data is displayed.

[0009] In one embodiment, prior to the step of inferring the material image through a pre-trained annotation model on an associated cloud server to obtain at least one candidate feature point and its second coordinates in the material image, the method further includes:

[0010] Obtain a set of labeled material images, wherein the set of labeled material images includes multiple material images under different imaging conditions, different material surfaces and / or different feature types, and each material image is labeled with feature points and the coordinates of the feature points;

[0011] The initial annotation model is trained using the labeled material image set until the prediction accuracy of the initial annotation model for feature points and their coordinates in the material images reaches a preset first accuracy threshold, thus obtaining the annotation model.

[0012] In one embodiment, the step of determining the second coordinates of the target feature point as the annotation data of the material image includes:

[0013] The second coordinates of the target feature point are validated, wherein the validation includes whether the second coordinate is a non-negative integer, and / or whether the second coordinate is within the pixel coordinate system of the material image;

[0014] The annotation data of the material image is obtained based on the target feature points that have passed the validity verification and the second coordinates of the target feature points.

[0015] In one embodiment, the step of acquiring a material image of the material to be tested and acquiring the first coordinates of the sampling operation in the material image includes:

[0016] If the acquisition of the material image fails, the step of acquiring the material image of the material to be tested is repeated, and the exception information corresponding to the failure to acquire the material image is recorded; and / or,

[0017] If the first coordinate acquisition fails, the step of acquiring the first coordinate in the material image is re-executed, and the abnormal information corresponding to the failure of the first coordinate acquisition is recorded.

[0018] In one embodiment, the step of inferring the material image using a pre-trained annotation model on an associated cloud server includes:

[0019] If the annotation model fails to infer, a backup annotation model is used for inference or manual annotation is triggered, and the abnormal information corresponding to the failure of the annotation model is recorded.

[0020] In one embodiment, after determining the second coordinates of the target feature point as the annotation data of the material image and displaying the annotation data, the method further includes:

[0021] The labeled data is verified to obtain verified labeled data;

[0022] The initial measurement model is trained using the calibration data and its corresponding material image until the accuracy of the size of the material to be measured output by the initial measurement model reaches a preset second accuracy threshold, thus obtaining the measurement model. The measurement model is used to measure the size of the material to be measured based on the calibration data.

[0023] In one embodiment, after the step of acquiring the material image of the material to be tested, the method further includes:

[0024] In the case where there are special feature points in the material image, the special feature points in the material image are preprocessed to obtain actual feature points, wherein the special feature points are feature points that cannot be identified in the material image;

[0025] Based on a geometric measurement strategy, the dimensions of the material to be measured are determined according to the coordinates of actual feature points.

[0026] In addition, to achieve the above objectives, this application also proposes a host computer, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the measurement method as described above.

[0027] In addition, to achieve the above objectives, this application also provides a measurement system, which includes a host computer and a cloud server, wherein the computer program of the host computer, when executed by a processor, implements the steps of the measurement method as described above.

[0028] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and which, when executed by a processor, implements the steps of the measurement method described above.

[0029] One or more technical solutions proposed in this application have at least the following technical effects:

[0030] Existing technologies rely on manual annotation, which sometimes results in annotation not occurring during measurement and vice versa, hindering the efficiency of industrial measurement. This application addresses this by acquiring a material image and the first coordinates of the point acquisition operation within the material image in response to a point-taking operation during measurement. This embeds the image acquisition action into the actual measurement process, providing a foundation for subsequent parallel annotation. Subsequently, a cloud-pretrained annotation model is used to infer the acquired material image, outputting at least one candidate feature point and its second coordinates within the material image, replacing the original manual identification and annotation process. By comparing the spatial relationship between the first coordinates of the point acquisition and the second coordinates of the candidate feature points, target feature points matching the current measurement intent are selected, enabling parallel processing of measurement operations and feature annotation. Finally, annotated data is generated and displayed based on the second coordinates of the target feature points.

[0031] Through the above steps, this application integrates the traditionally serial measurement and annotation processes into a parallel processing flow, enabling automatic feature annotation while continuously performing measurement operations. This fundamentally solves the problem of measurement interruption caused by manual annotation, resulting in low measurement efficiency and significantly improving the efficiency of industrial measurement. Attached Figure Description

[0032] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0033] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a flowchart illustrating an embodiment of the measurement method of this application.

[0035] Figure 2 A schematic diagram illustrating the training and deployment process of the labeled model provided for the measurement method of this application;

[0036] Figure 3 A flowchart illustrating another embodiment of the measurement method of this application;

[0037] Figure 4 A flowchart illustrating the anomaly handling process provided for the measurement method of this application;

[0038] Figure 5 A schematic diagram illustrating the training and deployment process of the measurement model provided for the measurement method of this application;

[0039] Figure 6 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the measurement method of this application.

[0040] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0041] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of this application and are not intended to limit this application.

[0042] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0043] In the field of intelligent manufacturing, AI-based visual dimensional measurement technologies (such as deep learning-based object detection and semantic segmentation algorithms) have gradually replaced traditional manual measurement, achieving a technological upgrade from manual visual inspection to automatic algorithm analysis. However, existing measurement models heavily rely on manual annotation during training, and manual annotation has the following drawbacks:

[0044] 1. The workload of annotation is huge and the annotation cost accounts for a high proportion (industry research shows that data annotation costs account for 80% of the overall deployment cost of AI measurement models, and the annotation cost will increase exponentially with product iteration).

[0045] 2. Manual annotation involves several steps, including image acquisition, annotation by measurement engineers, quality inspection and review of annotation data, and import of annotation data into the measurement model. There are instances where annotation is not performed during measurement or measurement is not performed during annotation. Furthermore, the average time for annotation of a single image is 20 seconds, resulting in a one-day cycle from the time the new measurement model is proposed to its online implementation. This makes it difficult to meet the rapid changeover requirements of small batches and multiple varieties in industrial production.

[0046] 3. Differences in the understanding of feature point definitions among different measurement engineers can lead to deviations in annotation results, directly affecting the accuracy of measurement model training and requiring an additional 10% to 20% of time to calibrate the annotation results. In addition, prolonged repetitive annotation can easily lead to visual fatigue. After working continuously for 4 hours, the annotation error rate can increase from 5% to 10%, especially for small features (such as micropores with a diameter of <1mm), where the annotation error increases significantly.

[0047] As mentioned above, current industrial measurement methods suffer from low efficiency due to the need for manual annotation of sample data to obtain AI measurement models. This application proposes a measurement method from the perspective of sample data. The measurement system responds to the point-taking operation triggered by the measurement engineer when measuring the material under test. The system acquires a material image of the material and the first coordinate of the point-taking operation within the material image. This step provides an image basis for annotation during measurement. Then, the acquired material image is uploaded to a pre-trained annotation model on a cloud server for inference, obtaining at least one candidate feature point in the material image and its second coordinate. This step replaces the existing manual identification and annotation of feature points with the computational inference capabilities of a large model, reducing the workload and labor costs of manual annotation and increasing annotation speed. Finally, based on the spatial relationship between the first coordinate and the second coordinate of the candidate feature point, the target feature point is determined from the candidate feature points, reducing errors caused by misunderstandings and visual fatigue during manual annotation based on objective data. Finally, annotation data is generated based on the second coordinates of the target feature points and displayed on the host computer, realizing the visualization of the annotation data. This allows measurement engineers to verify the accuracy of the annotation results, further improving the reliability of the annotation data. In summary, the measurement method of this application achieves parallel processing of manual measurement operations and intelligent annotation generation. During parallel processing, the annotation model replaces the traditional feature point identification and annotation process that relies on manual experience, overcoming the inherent defects of high labor costs, low consistency, slow speed, and low efficiency. The annotation data is visualized in real time, providing a data foundation for manual verification and annotation model optimization. By improving the accuracy and speed of the initial annotation data, the efficiency of industrial measurement is improved.

[0048] It should be noted that the execution entity in this embodiment can be a measurement system, including a host computer and a cloud server. The host computer can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone. The following description uses a measurement system as an example to illustrate this embodiment and the subsequent embodiments.

[0049] Based on this, the embodiments of this application provide a measurement method, referring to... Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the measurement method of this application.

[0050] In this embodiment, the measurement method includes steps S10 to S40:

[0051] Step S10: In response to the sampling operation triggered when measuring the material to be measured, acquire the material image of the material to be measured, and acquire the first coordinate of the sampling operation in the material image;

[0052] It should be noted that when the measurement engineer performs manual measurement operations on the material to be measured, a point acquisition operation will be triggered. In response to this point acquisition operation, the measurement system will automatically capture the material image on the current screen and record the first coordinate of the mouse click position in the pixel coordinate system of the material image.

[0053] Optionally, the material to be tested refers to the parts or products to be tested on the industrial production line, such as metal-machined gears, plastic-molded housings, or precision ceramic components.

[0054] Alternatively, the point selection operation can be performed by a measurement engineer using the measurement software to click on the displayed material image with a mouse or touchscreen to specify a feature point whose dimension needs to be measured, such as the center of a circle, the endpoint of a straight line, or the vertex of a corner.

[0055] Alternatively, the material image may be a high-resolution image acquired by an optical measuring machine (OMM).

[0056] In one embodiment, the measurement system captures material images (which can be 2D or 3D images) in real time through the image acquisition module of the host computer, and performs the acquisition operation synchronously, thereby ensuring that annotation can be performed during measurement. This enables measurement and annotation to be performed simultaneously, saving time and improving industrial measurement efficiency.

[0057] In one specific implementation, when the measurement engineer begins to manually measure the material to be measured, the host computer in the measurement system will automatically start a parallel processing flow of measurement and annotation. The host computer in the measurement system is equipped with a high-precision image acquisition module. This image acquisition module will take a screenshot of the material to be measured while the mouse is clicking on points, so as to ensure that the acquired material image can accurately reflect the characteristics and size information of the material to be measured.

[0058] Step S20: Infer the material image through the pre-trained annotation model on the associated cloud server to obtain at least one candidate feature point and its second coordinates in the material image;

[0059] It should be noted that the acquired material image is sent to the associated cloud server, where a pre-trained annotation model has been deployed. The pre-trained annotation model will perform inference on the received material image, analyze and find all possible feature points in the material image, that is, obtain at least one candidate feature point and the second coordinate of the candidate feature point in the material image. The candidate feature point can be one or more.

[0060] Optionally, the material images acquired in the measurement system are transmitted to the cloud server via a host computer. This transmission can be done through a dedicated communication channel between the host computer and the cloud server, or via a local area network, industrial IoT, or other data links. Encryption can also be performed during transmission to ensure data security and integrity. The specific transmission and encryption methods are not limited. The purpose is to efficiently and reliably deliver the material images to the cloud server to provide data input for the inference of the annotation model.

[0061] In one embodiment, the annotation model can be a large model based on a transformer architecture (a codec architecture), trained on massive amounts of industrial image data, capable of automatically identifying feature points in material images and their secondary coordinates within the image. It is understood that cloud servers provide powerful computing capabilities, utilizing the visual perception capabilities of artificial intelligence to perform comprehensive and automatic feature annotation on material images. This can replace manual annotation, overcoming the existing technology where manual annotation of a single image takes an average of 20 seconds, thus improving annotation response time and efficiency.

[0062] In one specific implementation, after taking a screenshot, the host computer in the measurement system transmits the material image and the first coordinates of the point-taking operation in the material image to the trained annotation model for inference. After the annotation model completes the inference, it obtains all feature points on the material image, but retains only one target feature point and its second coordinates.

[0063] In one feasible implementation, steps S21-S22 are included before step S20:

[0064] Step S21: Obtain a set of labeled material images, wherein the set of labeled material images includes multiple material images under different imaging conditions, different material surfaces and / or different feature types, and each material image is labeled with feature points and the coordinates of the feature points;

[0065] It should be noted that before the annotation model can be put into practical use, it is trained and deployed on a cloud server. First, a set of labeled material images is acquired, which covers most industrial scenarios, including material images acquired under different conditions.

[0066] Optionally, 100,000 labeled material images collected by an optical measuring machine during historical periods can be acquired. The optical measuring machine is equipped with a micron-level precision lens and can acquire 2D images of the material to be measured, which can be transmitted to a cloud server via a local area network.

[0067] Optionally, these conditions include, but are not limited to, different imaging conditions such as different light intensities, shooting angles, and depths of field; diverse material surfaces, such as reflective metal surfaces, diffuse plastic surfaces, and textured coating surfaces; and comprehensive feature types, such as common geometric features on industrial parts, such as round holes, threaded holes, keyways, curved surface contours, chamfers, and assembly reference surfaces. It can be understood that multiple material images under different imaging conditions, different material surfaces, and / or different feature types include multiple material images under different imaging conditions, multiple material images under different material surfaces, or multiple material images under different feature types; multiple material images under different imaging conditions and different material surfaces; and all possible combinations.

[0068] Optionally, each image is precisely annotated by measurement engineers. Annotation involves marking feature points (such as the center of a hole) on the material image and recording the precise location of those feature points in the pixel coordinate system of the material image, i.e., the coordinates of the feature points. This process ensures the diversity and accuracy of the annotated material image set, enabling the annotation model to adapt to the complexity of real-world industrial scenarios.

[0069] Step S22: Train the initial annotation model using the labeled material image set until the initial annotation model's prediction accuracy for feature points and their coordinates in the material images reaches a preset first accuracy threshold, thus obtaining the annotation model.

[0070] It should be noted that the initial annotation model is trained using a set of labeled material images. The training process involves the initial annotation model automatically learning and summarizing patterns from the labeled image data. The measurement system continuously inputs material images from the labeled image set into the initial annotation model, allowing it to predict feature points and their coordinates. The initial annotation model's predictions are then compared with the correct answers provided by professional measurement engineers within the labeled image set. Based on the differences, the initial annotation model automatically adjusts its internal parameters, gradually improving its predictive ability. This iterative learning process continues until the initial annotation model's prediction accuracy for new material images—that is, the accuracy with which it identifies feature points and their coordinates—reaches a preset first accuracy threshold. When the initial annotation model stably reaches or exceeds this first accuracy threshold, the training process ends, and the resulting annotation model is a practically applicable annotation model with feature recognition capabilities.

[0071] Alternatively, the initial labeled model can be a deep learning network that has not yet learned any specific knowledge, such as a visual model based on a transformer architecture.

[0072] Optionally, the first accuracy threshold is a performance standard set in advance according to the actual application requirements, such as requiring the initial annotation model to achieve an accuracy of more than 85% on the test set (divided from the set of labeled material images).

[0073] In this embodiment, a set of labeled material images covering multi-dimensional industrial scenarios is constructed. Based on this, an initial labeling model based on the transformer architecture is trained. This enables the final labeling model to learn and internalize the visual patterns corresponding to various complex imaging conditions, diverse material surfaces, and numerous feature types. This results in a high-precision and robust ability to identify feature points in material images. Consequently, when the final labeling model is applied in practice, it ensures that the candidate feature points and coordinates output during inference have high accuracy and reliability, reducing the risk of measurement process failure or low measurement accuracy due to misidentification by the labeling model.

[0074] For example, such as Figure 2 As shown, Figure 2 This is a flowchart illustrating the training and deployment process of the annotation model in this application. Specifically, step 201: Collect multiple material images under different imaging conditions, different material surfaces, and / or different feature types during historical periods, and annotate feature points and their coordinates on each material image to obtain an annotated material image set, which is a large-scale dataset; steps 202-203: Train the initial annotation model based on the transformer architecture using the annotated material image set. When the prediction accuracy of the initial annotation model for feature points and their coordinates in the material images reaches a preset first accuracy threshold, the annotation model is obtained; step 204: Deploy the annotation model to a cloud server.

[0075] Step S30: Based on the spatial relationship between the first coordinate and the second coordinate of the candidate feature point, determine the target feature point from the candidate feature points;

[0076] It should be noted that the measurement system determines the target feature point from the candidate feature points based on the spatial relationship between the first coordinate and the second coordinate of each candidate feature point.

[0077] In a specific implementation, the measurement system calculates the pixel distance between the first coordinates selected manually and the second coordinates of each candidate feature point output by the annotation model. For example, this can be done by calculating Euclidean distance. The measurement system then selects the candidate feature point closest to the first coordinate as the target feature point. For instance, if a measurement engineer intends to annotate the center of a threaded hole but clicks at a slightly off-center position, the candidate feature point whose second coordinate is closest to the first coordinate among the multiple candidate feature points output by the annotation model is selected as the target feature point. This step ensures consistency between the user's intent and the annotation model's output, reduces mismatches, lowers errors and labor costs associated with manual annotation, and improves annotation response speed, thereby increasing annotation efficiency.

[0078] Step S40: Determine the second coordinates of the target feature point as the annotation data of the material image, and display the annotation data.

[0079] It should be noted that the measurement system determines the second coordinates of the target feature point as the annotation data of the material image and displays the annotation data.

[0080] In one embodiment, the labeled data is displayed in a visual form on a host computer interface, for example, by highlighting target feature points in a material image and displaying their coordinate information. Exemplarily, the labeled data can be used for subsequent measurement model training, size calculation, or optimization of the labeled model.

[0081] Understandably, this step not only completes a single annotation, feeding back the results of automated intelligent annotation to measurement engineers for confirmation, improving annotation efficiency and reducing human error, but also provides a data foundation for training higher-precision measurement models, thereby gradually achieving the goal of fully automated AI annotation and measurement, and improving the efficiency of the entire industrial measurement process.

[0082] In one specific implementation, the cloud server in the measurement system returns the target feature points and their second coordinates to the host computer. After receiving the inference results, the host computer converts the second coordinates of the target feature points into a format conforming to annotation specifications. This conversion process includes coordinate system transformation, data format adjustment, and interface adaptation with the annotation tool. The second coordinates of the target feature points conforming to the annotation specifications are then returned to the host computer's display interface, where measurement engineers can view the accuracy of the annotations. The display interface presents images and annotation information in an intuitive way. This achieves the simultaneous annotation of all feature points of a material when completing a measurement task.

[0083] In one feasible implementation, step S40 includes steps S41-S42:

[0084] Step S41: Perform validity verification on the second coordinates of the target feature point, wherein the validity verification includes whether the second coordinate is a non-negative integer and / or whether the second coordinate is within the pixel coordinate system of the material image;

[0085] It should be noted that this step aims to improve the reliability of the annotation process by validating the second coordinates of the target feature points.

[0086] Understandably, validity checks are used to ensure that the coordinates of feature points derived from the annotation model are legal and reasonable. Validity checks can intercept invalid or erroneous coordinates caused by abnormal annotation model inference, data transmission errors, or other unexpected situations, thereby improving the accuracy of annotation data and ensuring the reliability of the entire annotation process.

[0087] Optionally, the validity check includes: checking whether the second coordinate is a non-negative integer and / or checking whether the second coordinate is within the pixel coordinate system of the material image. One or both checks can be performed, depending on the actual requirements.

[0088] Optionally, check whether the value of the second coordinate is a non-negative integer. This is because in an image, pixel positions are usually represented by positive integers starting from zero. For example, the top left corner of the image is the origin (0,0). Negative or decimal coordinates have no practical meaning in the pixel coordinate system and are considered invalid data.

[0089] Optionally, it can be checked whether the second coordinate falls within the valid range of the pixel coordinate system of the material image. The pixel coordinate system refers to a two-dimensional coordinate system established with the image width and height as boundaries. For example, for an image with a width of 1920 pixels and a height of 1080 pixels, the valid X-coordinate should be between 0 and 1919, and the Y-coordinate should be between 0 and 1079. If the coordinate values ​​exceed this range, it means that the feature point is not within the actual image area and is considered unreasonable data. By performing validity checks, the measurement system can filter out coordinate points with correct data format and feasible spatial locations.

[0090] Step S42: Based on the target feature points that have passed the validity check and the second coordinates of the target feature points, the annotation data of the material image is obtained.

[0091] It should be noted that once the second coordinates of the target feature point pass the validity check, i.e., are confirmed as valid data with correct format and reasonable position, the measurement system generates annotation data for the material image based on the valid data.

[0092] In one embodiment, the labeled data is not merely coordinate values ​​that have passed validity verification; it may also include type information of the target feature point (such as indicating whether the feature point is the center of a circle or a corner) and other attributes (such as which edge of the material being measured it belongs to). This step of validating the labeled data improves its quality and reliability, providing a data foundation for subsequent applications such as displaying it to measurement engineers for review, storing it in a database, or using it to train more accurate measurement models, thus ensuring the accuracy and usability of the output results from the labeling process.

[0093] In this embodiment, by validating the coordinates of the target feature points, illegal coordinate data caused by accidental output anomalies of the annotation model, disturbances in the data transmission process, or external interference is prevented from flowing into subsequent processes. This improves the accuracy and reliability of the annotation results. Consequently, the measurement model trained based on the validated annotation data outputs more accurate and reliable dimensional measurement results.

[0094] This embodiment provides a measurement method. Responding to a point-taking operation triggered by a measurement engineer during measurement of the material under test, the measurement system acquires a material image of the material and the first coordinates of the point-taking operation within the material image. This step allows for the acquisition of the material image during measurement, providing an image basis for annotation. Subsequently, the acquired material image is uploaded to a pre-trained annotation model on a cloud server for inference, obtaining at least one candidate feature point in the material image and the second coordinates of the candidate feature point within the material image. This step replaces the existing process of manually identifying and annotating feature points with the computational inference capabilities of a large model, reducing the workload and labor costs of manual annotation and increasing annotation speed. Based on the spatial relationship between the first coordinates and the second coordinates of the candidate feature points, a target feature point is determined from the candidate feature points. This reduces errors caused by misunderstandings and visual fatigue during manual annotation based on objective data. Finally, annotation data is generated based on the second coordinates of the target feature point and displayed on a host computer, achieving visualization of the annotation data. This allows measurement engineers to verify the accuracy of the annotation results, further improving the reliability of the annotation data. In summary, the measurement method of this application enables parallel processing of manual measurement operations and intelligent annotation generation. During parallel processing, the annotation model replaces the traditional feature point identification and annotation process that relies on human experience, overcoming the inherent defects of high labor costs, low consistency, slow speed, and low efficiency. The annotation data is visualized in real time, providing a data foundation for manual review and annotation model optimization. By improving the accuracy and speed of the initial annotation data, the efficiency of industrial measurement is improved.

[0095] For example, please refer to Figure 3 , Figure 3A flowchart of a measurement method is shown. Specifically, step 401: In response to the point-taking operation triggered by the measurement engineer when measuring the material to be measured, the host computer of the measurement system can acquire the material image of the material to be measured and the first coordinate of the point-taking operation in the material image; steps 402-403: The material image and the first coordinate are transmitted to the cloud server, and multiple candidate feature points in the material image and the second coordinates of each candidate feature point in the material image are obtained through inference using a pre-trained annotation model on the cloud server; step 404: Based on the spatial relationship between the first coordinate and the second coordinate, the target feature point is determined and the second coordinate of the target feature point is obtained; steps 405-406: The second coordinate of the target feature point is validated, i.e., the result is converted to obtain the target feature point that passes the validity validation and the second coordinate of the target feature point, thereby obtaining the annotation data; finally, the annotation data is displayed in the host computer.

[0096] Based on the above embodiments of this application, in another embodiment of this application, the same or similar content as the above embodiments can be referred to the above description, and will not be repeated hereafter. Step S10 includes steps S11~S12:

[0097] Step S11: If material image acquisition fails, the step of acquiring the material image of the material to be tested is re-executed, and the abnormal information corresponding to the material image acquisition failure is recorded; and / or,

[0098] It should be noted that the measurement method includes an anomaly handling mechanism during the data acquisition phase. In one embodiment, when the measurement system performs the step of simultaneously acquiring the material image of the material to be measured and recording the first coordinates corresponding to the point acquisition operation, a failure may occur. Once the measurement system detects such an anomaly, it will immediately trigger corresponding remedial measures and record the anomaly information corresponding to the material image acquisition failure.

[0099] Optionally, failure to acquire material images may result from a disconnection with the optical measurement equipment, camera hardware failure, or loss of data packets during image transmission; understandably, these failures will directly prevent subsequent annotation processes from proceeding due to a lack of necessary input data.

[0100] Optionally, if material image acquisition fails, the measurement system will automatically attempt to re-acquire the image, such as re-initializing the connection with the camera or resending the image acquisition command; or prompting the measurement engineer to take a screenshot.

[0101] Step S12: If the first coordinate acquisition fails, the step of acquiring the first coordinate in the material image is re-executed, and the abnormal information corresponding to the failure of the first coordinate acquisition is recorded.

[0102] Optionally, the failure to acquire the first coordinate may be due to the host computer software failing to correctly capture the mouse click event, or a calculation error occurring when converting screen coordinates to image pixel coordinates. Understandably, these failures will directly prevent subsequent annotation processes from proceeding due to a lack of necessary input data. Once the measurement system detects such an anomaly, it will immediately trigger corresponding remedial measures.

[0103] Optionally, if the first coordinate acquisition fails, the measurement system will automatically attempt to reconnect, or will issue a clear prompt to the measurement engineer to guide them to perform the point acquisition operation again, such as prompting "Point acquisition failed, please click again" through a pop-up window or sound alarm on the host computer's display interface.

[0104] It should also be noted that remedial measures (automatic retry and manual retry prompts) can be executed independently or simultaneously, depending on the specific combination of failures. Understandably, this mechanism, combining automatic retries with manual guidance, aims to quickly restore the data acquisition process and minimize the disruption of measurement work caused by anomalies.

[0105] Meanwhile, the measurement system records corresponding anomaly information. For example, this recorded anomaly information can be a log table, typically including the specific time the anomaly occurred, the anomaly type (image acquisition failure, coordinate acquisition failure, or both), possible cause codes, and the remedial measures taken and their results. Understandably, detailed anomaly recording not only helps measurement engineers diagnose and handle problems promptly, but more importantly, it provides data for subsequent measurement system maintenance and optimization. For instance, during subsequent reviews, the frequency and patterns of anomalies can be analyzed to specifically improve the stability of hardware connections or enhance software algorithms, thereby continuously improving the reliability of the entire measurement system.

[0106] Optionally, when a material image acquisition failure occurs, the corresponding remedial measures for the material image acquisition failure are executed and the abnormal information is recorded; when a first coordinate acquisition failure occurs, the corresponding remedial measures for the first coordinate acquisition failure are executed and the abnormal information is recorded; when both failures occur simultaneously, their respective remedial measures are taken and their respective abnormal information is recorded.

[0107] In one feasible implementation, the measurement method further includes:

[0108] If the annotation model fails to infer, a backup annotation model will be used for inference or manual annotation will be triggered, and the abnormal information corresponding to the failure of the annotation model will be recorded.

[0109] It should be noted that the measurement method also includes a fault-tolerance mechanism for the annotation model inference process. Specifically, in the step of inferring the material image using the annotation model pre-trained on the associated cloud server, the annotation model inference may fail.

[0110] Understandably, a labeling model inference failure means that the labeling model is unable to produce a valid output from the input material image. For example, the reasons may include the temporary unavailability of cloud services, errors in loading the labeling model, abnormal format of the input material image that exceeds the processing capacity of the labeling model, or internal errors occurring in the labeling model when processing certain special image features.

[0111] Once the measurement system detects such an inference failure, it immediately initiates remedial measures. In one embodiment, the remedial measures include enabling a backup annotation model for inference. The backup annotation model can be a pre-prepared, computationally lighter, or slightly simplified but more stable alternative model. Exemplarily, it can be a lightweight network deployed on a local server or a cloud server to ensure that basic inference capabilities are still provided and the measurement system continues to operate even if the primary annotation model on the cloud server fails.

[0112] Remedial measures also include triggering manual annotation. In one scenario, when the primary annotation model on the cloud server fails to infer, the measurement engineer can be notified via interface prompts and audible alarms on the host computer, requesting their intervention to manually annotate the current material image. This ensures that the measurement task does not come to a complete halt due to the interruption of the automated process. In another scenario, if the backup annotation model is also unavailable or deemed unsuitable for the current scenario, the measurement system will trigger manual annotation, notifying the measurement engineer via interface prompts and audible alarms on the host computer, requesting their intervention to manually annotate the current material image. This again ensures that the measurement task does not come to a complete halt due to the interruption of the automated process.

[0113] Simultaneously, the measurement system records anomaly information corresponding to the failure of the annotation model inference. In a specific implementation, the recorded anomaly information may include details such as the time of the anomaly, the identifier of the material image involved, the specific error code or description of the inference failure, and the remedial measures taken by the measurement system (whether a backup model was activated or manual annotation was notified) and their results. For example, by analyzing the anomaly information, weaknesses in the annotation model or measurement system can be located, thereby enabling targeted improvements and continuously enhancing the overall reliability of the measurement system.

[0114] In specific implementations, anomalies may occur during the parallel processing of measurement and annotation, such as material image transmission failures or annotation model inference errors. To ensure the stability and reliability of the measurement system, a comprehensive anomaly handling and feedback mechanism is implemented. When an anomaly occurs, the host computer will promptly issue an alarm, prompting the measurement engineer to take appropriate action. Simultaneously, the measurement system will record detailed information about the anomaly, including the time of occurrence, specific manifestations, and relevant material images, for subsequent troubleshooting and analysis. Furthermore, the anomaly handling mechanism will take different measures based on the type and severity of the anomaly. For example, if material image transmission fails, the measurement system will attempt to retransmit; if annotation model inference errors occur, the measurement system may prompt the use of a backup annotation model or manual annotation.

[0115] For example, such as Figure 4 As shown, Figure 4 This is a flowchart illustrating the anomaly handling process of this application. Specifically, when the measurement system performs the steps of acquiring the material image of the material to be measured and acquiring the first coordinates corresponding to the point acquisition operation, after the host computer acquires the material image or the first coordinates, the measurement system monitors whether the material image or the first coordinates can be transmitted to the cloud server. If not, it performs the operation of re-acquiring the image and / or prompts for re-acquiring the points; it records the corresponding anomaly information and reports an error. If transmission is possible, inference is performed using the pre-trained annotation model on the cloud server. If the annotation model cannot perform effective inference, i.e., the annotation model inference fails, the backup annotation model is used for inference or the manual annotation operation is triggered, and the corresponding anomaly information is recorded and reported an error; if the annotation model can perform inference, the output result of the annotation model is converted, i.e., validity is verified, the annotation data is obtained, and it is displayed on the host computer.

[0116] Based on the above embodiments of this application, in another embodiment of this application, the same or similar content as the above embodiments can be referred to the above description, and will not be repeated hereafter. Step S40 is followed by steps A41-A42:

[0117] Step A41: Verify the labeled data to obtain verified labeled data;

[0118] It should be noted that further verification of the labeled data generated in the aforementioned steps can filter out problematic labeled data, correct or remove such labeled data, thereby obtaining high-quality verified labeled data.

[0119] Optionally, the verification here may include the valid verification in the foregoing embodiments, or a more stringent verification than the valid verification in the foregoing embodiments, in order to ensure that the labeled data not only has the correct format, but also meets the measurement requirements in terms of logic and accuracy.

[0120] For example, the verification may include confirmation and review of the annotation results displayed on the host computer interface by the measurement engineer, or automatic logical checks by the measurement system based on preset geometric rules (such as whether the relative positional relationship between feature points is reasonable).

[0121] Step A42: By verifying the labeled data and its corresponding material image, the initial measurement model is trained until the accuracy of the size of the material to be measured output by the initial measurement model reaches the preset second accuracy threshold, and the measurement model is obtained. The measurement model is used to measure the size of the material to be measured according to the labeled data.

[0122] It should be noted that the initial measurement model was trained using high-quality calibration data and its corresponding material images.

[0123] Optionally, the initial measurement model is a model to be trained for direct size calculation. Its function differs from the annotation model; the measurement model is used to calculate the size of the material to be measured based on the coordinates of feature points. The training process involves teaching the initial measurement model the mapping relationship from feature points and their coordinates on the material image to dimensions.

[0124] The measurement system continuously inputs material images and calibration annotation data into the initial measurement model. The initial measurement model outputs the dimensions of the material to be measured, and then compares these dimensions with the standard dimensions calculated using a geometric algorithm based on the coordinates of the annotation points. The parameters of the initial measurement model are adjusted according to the differences. This process iterates repeatedly until the accuracy of the dimensions output by the initial measurement model—that is, the degree of agreement between the predicted and actual values—reaches a preset second accuracy threshold. This second accuracy threshold is set according to the actual industrial measurement precision requirements, for example, allowing dimensional errors within a few micrometers. When the initial measurement model meets the preset second accuracy threshold, a high-precision measurement model ready for practical application is obtained. In essence, this measurement model ultimately completes the automatic, rapid, and accurate measurement of the material dimensions based on the annotation data obtained from the annotation model, achieving full automation from intelligent annotation to automatic measurement.

[0125] It should also be noted that this measurement model is trained on a cloud server and then deployed to the cloud server for practical application. Before the measurement model is deployed, the annotation data obtained from the annotation model can be used to measure the size of the material under test based on a geometric measurement strategy. After the measurement model is deployed, the annotation data obtained from the annotation model is verified and then input into the measurement model to automatically measure the size of the material under test.

[0126] In this specific implementation, to ensure the accuracy of the measurement results, a measurement model is trained using intelligently labeled and error-free data from a labeling model. This model is then used to automatically measure the dimensions of the material under test based on the labeled data. After the intelligently labeled and error-free labeled data and the material image are automatically uploaded to the model training platform on the cloud server, the cloud server automatically performs image preprocessing operations (such as normalization and noise reduction) and fine-tunes a larger model with a different transformer architecture than the labeling model (i.e., the initial measurement model). Once the initial measurement model training is complete, the trained measurement model is automatically deployed to the cloud server. This achieves fully automated integration of the entire process from image labeling to measurement model training, deployment, and final AI measurement.

[0127] During the measurement model training phase, measurement engineers need to upload all images and labeled data to the model training platform and perform preprocessing operations on the images. This process not only consumes the time and effort of measurement engineers but is also prone to human error due to operational negligence, reducing the efficiency and quality of measurement model training. At the same time, excessive manual intervention makes it difficult to automate and scale up measurement model training, failing to meet the needs of industrial production for rapid updates and iterations of AI measurement models.

[0128] In this embodiment, high-quality labeled data is obtained by verifying the labeled data, which effectively improves the rationality and reliability of the training data. Subsequently, a measurement model for size calculation is trained based on the high-quality data, enabling the measurement model to learn the precise mapping relationship from the coordinates of feature points to physical dimensions. This embodiment automates the training and deployment of the measurement model in the cloud, which not only improves the accuracy and stability of the size measurement results, but also realizes full-process automation from intelligent labeling to precise measurement, completely eliminating the continuous dependence on manual intervention. Thus, it provides an efficient, reliable, and scalable measurement method for industrial measurement.

[0129] For example, such as Figure 5 As shown, Figure 5 The flowchart illustrates the training and deployment process of the measurement model provided for this measurement method. Specifically, step 501: Verify the annotation data obtained from the annotation model to obtain high-quality verification annotation data; step 502: Transmit the data to the model training platform on the cloud server; step 503: Use the verification annotation data and its corresponding material image to train the initial measurement model on the cloud server until the accuracy of the size of the material to be measured output by the initial measurement model reaches the preset second accuracy threshold, and finally obtain the measurement model; step 504: Deploy the measurement model on the cloud server.

[0130] In one possible implementation, step S40 is followed by:

[0131] The labeled data is validated to obtain validated labeled data.

[0132] By verifying the labeled data and its corresponding material images, the labeled model is optimized to obtain an optimized labeled model.

[0133] It should be noted that this implementation method aims to achieve self-improvement and continuous optimization of the annotation model. Specifically, the measurement system verifies the annotation data generated in the aforementioned steps to obtain high-quality, high-reliability verification annotation data. The verification here can refer to the verification method in step A41.

[0134] In a specific implementation, the measurement system optimizes the annotation model by verifying the labeled data and its corresponding material images. This optimization refers to the model retraining process, which uses newly generated, validated, high-quality data to update and improve the performance of the deployed annotation model.

[0135] Understandably, as the measurement system processes new materials or encounters new imaging conditions, it will continuously generate new labeled data. This new labeled data contains knowledge that the labeled model has not learned before.

[0136] For example, the measurement system can periodically use accumulated calibration data and their corresponding material images as new training samples to continue training the existing annotation model (for example, this process can be called model fine-tuning). In this way, the annotation model can continuously adapt to new data distributions, learn more complex feature patterns, thereby correcting existing cognitive biases and enhancing its generalization ability. Ultimately, after iterative learning, an optimized annotation model is obtained. Understandably, this gives the entire annotation model the ability to continuously evolve, constantly improving the accuracy and reliability of the annotations over time and with the accumulation of data.

[0137] It should also be noted that the measurement model can be fine-tuned using newly generated data to obtain a measurement model that can accurately measure the material size.

[0138] In this embodiment, the generated labeled data is validated to filter out noisy data and ensure the high accuracy of the training data used for optimizing the labeled model. Furthermore, the deployed labeled model is fine-tuned and optimized using high-quality validation data, enabling the model to continuously learn about new scenarios and features encountered in real-world applications. This allows the model to adapt to new data changes and correct its internal parameters, improving its labeling performance over time and with data accumulation. Ultimately, this enhances the model's accuracy, stability, and adaptability to complex industrial scenarios in long-term operation.

[0139] Based on the above embodiments of this application, in another embodiment of this application, the same or similar content as the above embodiments can be referred to the above description, and will not be repeated hereafter. Step S10 is followed by steps D10~D20:

[0140] Step D10: If there are special feature points in the material image, preprocess the special feature points in the material image to obtain the actual feature points. The special feature points are feature points that cannot be identified in the material image.

[0141] It should be noted that the measurement method further includes processing procedures for special cases to enhance the applicability and robustness of the measurement system. In one embodiment, when special feature points exist in the material image, these special feature points may be feature points that cannot be directly and accurately identified by the annotation model or manual annotation due to the material's abnormally complex structure (e.g., assemblies with multiple occlusions, contours with free-form surfaces) and / or limitations imposed by imaging conditions (e.g., strong reflections, shadow occlusions, insufficient depth of field leading to blurring). For example, a special feature point may be the inner wall edge of a deep hole, or the center point of a light spot formed by specular reflection from a metal surface, etc.

[0142] In a specific implementation, after the measurement system identifies such special feature points, it performs preprocessing. This preprocessing aims to transform the special feature points, which are difficult to process directly, into actual feature points that can be used in subsequent measurement processes. For example, preprocessing may include identifying the type of special feature point and then using image enhancement algorithms to suppress reflections and improve contrast based on that type; or using multi-view image fusion technology to reconstruct the three-dimensional information of the occluded area; in some cases, interactive guidance may be provided to the measurement engineer, for example, the engineer may specify an approximate feature area on the material image, and the measurement system will then perform precise positioning based on this instruction. It is understood that preprocessing transforms the special feature point situation into a normal situation, thereby obtaining accurately located and clearly defined actual feature points and their coordinates.

[0143] Step D20: Based on the geometric measurement strategy, measure the size of the material to be measured according to the coordinates of the actual feature points.

[0144] It should be noted that, based on the geometric measurement strategy, the coordinates of actual feature points are used to complete the measurement of the final size of the material to be measured.

[0145] Understandably, geometric measurement strategies can be computational methods based on mathematical geometric principles. Specifically, when the actual feature points are a series of points on the boundary of a circle, the measurement system can use a least-squares circle fitting algorithm to calculate the diameter of the circle. If the actual feature points represent the two endpoints of a straight line, the pixel distance between the two points can be directly calculated and then converted into physical dimensions by combining camera calibration parameters. Through this geometric calculation based on precise coordinates, even for difficult-to-identify special feature points, the measurement system can output high-precision dimensional measurement results, ensuring the effectiveness and accuracy of the measurement method in various complex industrial scenarios.

[0146] In one specific implementation, because the point structure of individual feature points of some materials is relatively complex, special processing, i.e., preprocessing, is required for the points of special feature points. During the automatic measurement process, after acquiring the material image, the measurement system first identifies feature points. If the feature point is not a special feature point, the measurement system executes the step of inferring the material image through a pre-trained annotation model on the associated cloud server to obtain at least one candidate feature point and its second coordinates in the material image, until the size of the material to be measured is obtained. Once a special feature point is determined, the measurement system identifies the corresponding special feature point type. After preprocessing, the actual feature point and its coordinates are obtained and finally returned to the host computer.

[0147] Then, using a geometric measurement strategy, the various dimensional parameters of the material to be measured are calculated comprehensively and accurately. Specifically, for sections where the actual feature points are straight lines, coordinate calculations are used to accurately calculate the distance between the two endpoints, thus obtaining the accurate length of the line. For sections where the actual feature points are circular, the radius of the circle is calculated based on the coordinate data, and then the diameter is calculated from the radius. Alternatively, after the measurement model is put into practical application, the coordinates of the actual feature points can be input into the measurement model to automatically measure the dimensions of the material to be measured.

[0148] In this embodiment, the size of the material to be measured is measured by preprocessing special feature points and combining them with a geometric measurement strategy. Specifically, this embodiment transforms special feature points that are difficult to identify directly into coordinates of actual feature points that can be accurately quantified. Based on the coordinates of the actual feature points, a geometric measurement strategy is used to calculate the size, ensuring that high-precision results can still be output in complex measurement tasks. This effectively solves the problem of special feature points being unidentifiable or inaccurately identified due to abnormal material structure and / or poor imaging conditions.

[0149] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the measurement method of this application. Any simple variations based on this technical concept, such as the interaction and combination of various embodiments, are all within the protection scope of this application.

[0150] This application provides a host computer, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the measurement method in the first embodiment described above.

[0151] The following is for reference. Figure 6 It shows a schematic diagram of the structure of a host computer suitable for implementing the embodiments of this application. Figure 6 The host computer shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0152] like Figure 6 As shown, the host computer may include a processing device 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in the read-only memory 1002 or a program loaded from the storage device 1003 into the random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the host computer. The processing device 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows the host computer to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows host computers with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0153] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0154] The host computer provided in this application, employing the measurement method described in the above embodiments, can solve the technical problem of low efficiency in industrial measurement due to reliance on manual annotation in existing dimensional measurements. Compared with the prior art, the beneficial effects of the host computer provided in this application are the same as those of the measurement method provided in the above embodiments, and other technical features of the host computer are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0155] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0156] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0157] This application provides a measurement system, the measurement system including: as shown in the example Figure 6 The host computer and cloud server shown in the figure enable the measurement system to implement the steps of the above measurement method, and solve the technical problem that existing size measurement relies on manual annotation and industrial measurement is inefficient.

[0158] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to perform the measurement method in the above embodiments.

[0159] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0160] The aforementioned computer-readable storage medium may be included in the host computer; or it may exist independently and not be assembled into the host computer.

[0161] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0162] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0163] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0164] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for performing the above-described measurement method. This solves the technical problem of low efficiency in industrial measurement due to reliance on manual annotation in existing dimensional measurements. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the measurement method provided in the above embodiments, and will not be repeated here.

[0165] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method of measurement, characterized by, The method is applied to a measurement system, and the measurement system comprises a host computer, and the measurement method comprises: In response to a point picking operation triggered for a display interface in a display screen of the host computer when measuring a material to be measured, a material image of the material to be measured is acquired, and a first coordinate of the point picking operation in the material image is acquired, wherein the material image is an image currently displayed in the display interface captured by the host computer, and the first coordinate is a coordinate in a pixel coordinate system of the material image; A pre-trained labeling model on an associated cloud server is used to infer the material image, to obtain at least one candidate feature point and a second coordinate of the candidate feature point in the material image; A target feature point is determined from the candidate feature points based on a spatial position relationship between the first coordinate and the second coordinate of the candidate feature point; The second coordinate of the target feature point is determined as labeling data of the material image, and the labeling data is displayed, so that a size of the material to be measured is measured according to the labeling data; The labeling data is verified to obtain verification labeling data; An initial measurement model is trained through the verification labeling data and the corresponding material image, until an accuracy of a size of the material to be measured output by the initial measurement model reaches a preset second accuracy threshold, to obtain a measurement model, wherein the measurement model is used to measure a size of the material to be measured according to labeling data.

2. The measurement method of claim 1, wherein, Before the step of using the pre-trained labeling model on the associated cloud server to infer the material image to obtain at least one candidate feature point and a second coordinate of the candidate feature point in the material image, the following step is further included: A set of labeled material images is acquired, wherein the set of labeled material images comprises a plurality of material images under different imaging conditions, different material material surfaces and / or different feature types, and each material image is labeled with a feature point and a coordinate of the feature point; An initial labeling model is trained through the set of labeled material images, until a prediction accuracy of the initial labeling model for the feature point in the material image and the coordinate of the feature point reaches a preset first accuracy threshold, to obtain a labeling model.

3. The method of claim 1, wherein, The step of determining the second coordinate of the target feature point as the labeling data of the material image comprises: The second coordinate of the target feature point is subjected to validity verification, wherein the validity verification comprises whether the second coordinate is a non-negative integer and / or whether the second coordinate is within a pixel coordinate system of the material image; The labeling data of the material image is obtained according to the target feature point and the second coordinate of the target feature point that pass the validity verification.

4. The method of claim 1, wherein, The step of acquiring the material image of the material to be measured and acquiring the first coordinate of the point picking operation in the material image comprises: If the material image acquisition fails, the step of acquiring the material image of the material to be measured is re-executed, and abnormal information corresponding to the material image acquisition failure is recorded; and / or, If the first coordinate acquisition fails, the step of acquiring the first coordinate of the sampling point in the material image is re-executed, and abnormal information corresponding to the first coordinate acquisition failure is recorded.

5. The method of claim 1, wherein, The step of inferring the material image by the pre-trained labeling model on the associated cloud server comprises: If the labeling model inference fails, a backup labeling model is enabled for inference or manual labeling is triggered, and abnormal information corresponding to the labeling model inference failure is recorded.

6. The method of any one of claims 1 to 5, wherein, After the step of acquiring the material image of the material to be measured, the method further comprises: In the case that there is a special feature point in the material image, the special feature point in the material image is preprocessed to obtain an actual feature point, wherein the special feature point is a feature point that cannot be recognized in the material image; Based on a geometric measurement strategy, the size of the material to be measured is measured according to the coordinates of the actual feature point.

7. A host computer, characterized by The host computer comprises a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the measurement method according to any one of claims 1 to 6.

8. A measurement system characterized by, The measurement system comprises a cloud server and a host computer according to claim 7.

9. A storage medium, characterized by The storage medium is a computer readable storage medium, and the storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the measurement method according to any one of claims 1 to 6. The storage medium is a computer readable storage medium, and the storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the measurement method according to any one of claims 1 to 6.

Images