Industrial key point detection method and device based on deep learning and computer equipment

Abstract

The application discloses a kind of industrial key point detection method, device and computer equipment based on deep learning, which is by the way of marking point to the key point of image point marking in step S101;Step S102: each marking point is expanded according to the set pixel to the square region with the marking point as center;Step S103: the marking point of each square region is converted into the marking point of circular heat map;Step S104: each marked image is trained to generate key point detection model;Step S105: the image to be detected is input into the key point detection model to detect the key point, and the key point coordinate output is detected, since the way of point marking is used, it is simpler to mark, and the detection accuracy of key point can be greatly improved by marking point pixel expansion and heat map conversion.In addition, the maximum point of output heat map distribution is used as the key point coordinate output during detection, which can greatly improve the detection efficiency.

Description

Technical Field

[0001] This invention relates to the field of computer vision inspection technology, and in particular to a method, apparatus and computer equipment for industrial key point detection based on deep learning. Background Technology

[0002] In the context of Industry 4.0, intelligent and automated production processes have become the core driving force for the development of the manufacturing industry. Computer vision, as one of the key technologies for realizing intelligent manufacturing, has greatly promoted the progress of industrial automation through its applications in image processing, object detection, and automatic recognition. In particular, key point detection technology plays a crucial role in ensuring production quality and improving production efficiency. With the continuous emergence and verification of new technologies, such as deep learning, key point detection technology is also constantly evolving and upgrading to meet the ever-increasing demands of industrial production.

[0003] Currently, although existing keypoint detection technologies have achieved certain results, they still face problems such as insufficient accuracy, strong dependence on large amounts of high-quality labeled data, and inadequate generalization ability. These problems are particularly prominent in complex industrial production environments, becoming key factors restricting the development of industrial automation and intelligence. Therefore, how to optimize keypoint detection algorithms, improve their accuracy and real-time performance, reduce dependence on high-quality labeled data, and enhance the generalization ability of the algorithms has become a research hotspot and focus. Summary of the Invention

[0004] This invention aims to at least partially solve one of the technical problems in related technologies. Therefore, the objective of this invention is to propose a method, apparatus, and computer device for industrial keypoint detection based on deep learning.

[0005] To achieve the above objectives, in a first aspect, according to embodiments of the present invention, a deep learning-based industrial keypoint detection method includes: Image data annotation steps: Annotate key points in the image using annotation points; Annotation point expansion steps: Expand each annotation point into a square area centered on the annotation point by a set number of pixels; Heatmap conversion steps: Convert the markers in each square area into markers for the circular heatmap; Model training steps: Train the model on each labeled image to generate a keypoint detection model; Key point detection steps: The image to be detected is input into the key point detection model to detect key points, and the coordinates of the detected key points are output.

[0006] Furthermore, according to one embodiment of the present invention, the output of the detected key point coordinates includes: Obtain information from each output heatmap; Obtain the coordinates of the maximum pixel value of each output heatmap based on the information from each output heatmap; The coordinates of the maximum pixel value of each output heatmap are used as the coordinates of each key point.

[0007] Furthermore, according to one embodiment of the present invention, converting the markers of each square region into markers for a circular heatmap includes: Find the center point of the square by using its four corners; Generate a circular heat map with the center point of the square as the center.

[0008] Furthermore, according to one embodiment of the present invention, the circular heat map marker is a circular marker whose image heat value gradually decreases from the center point outwards.

[0009] Furthermore, according to one embodiment of the present invention, before expanding each marker point into a square region centered on the marker point by a set number of pixels, the method further includes: Image data conversion steps: Convert the labeled JSON format image data into TXT text format image data.

[0010] Furthermore, according to one embodiment of the present invention, expanding each marker point into a square region centered on the marker point according to a set number of pixels specifically means expanding each marker point into a square region centered on the marker point with a size of 50 pixels.

[0011] Furthermore, according to one embodiment of the present invention, the keypoint detection model is generated by training a segmentation task model of a YOLOv8 deep learning model.

[0012] Secondly, embodiments of the present invention also provide an apparatus for industrial keypoint detection based on deep learning, comprising: Image data annotation module: The image data annotation module is used to annotate key points in an image by using annotation points; Annotation point expansion module: The annotation point expansion module is used to expand each annotation point into a square area centered on the annotation point according to a set number of pixels; Heatmap Conversion Module: The heatmap conversion module is used to convert the markers of each square area into markers of a circular heatmap; Model training module: The model training module is used to train the model on each labeled image to generate a key point detection model; Key point detection module: The key point detection module is used to input the image to be detected into the key point detection model to detect key points and output the coordinates of the detected key points.

[0013] Thirdly, a computer device according to an embodiment of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the deep learning-based industrial key point detection method as described above.

[0014] According to embodiments of the present invention, a deep learning-based industrial keypoint detection method, apparatus, and computer device are provided. The method includes: an image data annotation step: annotating keypoints in an image using annotation points; an annotation point expansion step: expanding each annotation point into a square region centered on the annotation point by a set number of pixels; a heatmap conversion step: converting the annotation points in each square region into annotation points in a circular heatmap; a model training step: training a model on each annotated image to generate a keypoint detection model; and a keypoint detection step: inputting the image to be detected into the keypoint detection model to detect keypoints and outputting the detected keypoint coordinates. Because point annotation is used, annotation is simpler, and the pixel expansion of annotation points and heatmap conversion can greatly improve the detection accuracy of keypoints. Furthermore, using the point with the maximum output heatmap distribution as the keypoint coordinate output can greatly improve detection efficiency.

[0015] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the deep learning-based industrial key point detection method provided in an embodiment of the present invention; Figure 2 This is a flowchart of the method for detecting and outputting key point coordinates provided in an embodiment of the present invention; Figure 3 This is a flowchart of another deep learning-based industrial key point detection method provided in an embodiment of the present invention; Figure 4 This is a structural block diagram of an industrial key point detection device based on deep learning provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of heat point conversion provided in an embodiment of the present invention; Figure 6 This is an example of the output effect diagram provided by an embodiment of the present invention, showing the detection of key point coordinates based on a heat map; Figure 7 This is a structural block diagram of a computer device provided in an embodiment of the present invention.

[0018] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0020] On the one hand, refer to Figure 1 This invention provides a deep learning-based method for detecting key points in industry, comprising the following steps: Image data annotation step S101: Annotate key points in the image using annotation points.

[0021] Annotation point expansion step S102: Expand each annotation point into a square area centered on the annotation point according to the set number of pixels.

[0022] Heatmap conversion step S103: Convert the markers of each square area into markers of the circular heatmap.

[0023] Model training step S104: Train the model on each labeled image to generate a key point detection model.

[0024] Key point detection step S105: Input the image to be detected into the key point detection model to detect key points, and output the coordinates of the detected key points.

[0025] Specifically, in this embodiment of the invention, the YOLOv8 model can be used for deep learning and object detection of image data. In the training data preparation for the Segment task branch of the YOLOv8 model, traditional methods typically require manually labeling and connecting keypoints around the image to create connected components. This is not only time-consuming and labor-intensive, but also lacks accuracy and is extremely complex for tasks requiring precise identification of small keypoints. To optimize this process, in the image data annotation step S101 of this embodiment, keypoints in the image are annotated using point annotation. This employs a simpler pose task annotation method for pose detection. In the pose task annotation method, the annotator only needs to perform simple point annotations at keypoint locations, greatly simplifying the data preparation process and reducing time consumption.

[0026] After key points in the image are labeled using point annotation, the original labeled data needs to be converted into a format suitable for training the YOLOv8 segmentation model (YOLOv8-segment) through a processing function. In the annotation point expansion step S102: each annotation point is expanded into a square region centered on the annotation point by a set number of pixels; during the conversion process, the annotation point is converted into a square ROI region centered on the key point. Specifically, expanding each annotation point into a square region centered on the annotation point by a set number of pixels means expanding each annotation point into a square region centered on the annotation point by 50 pixels. Experimental verification shows that, taking the industrial parts inspection project in the actual application of this embodiment as an example, the segmentation effect of expanding by 50 pixels is the best. Using 50 pixels to expand the annotation points ensures the size adaptability of the mask region during model training, and also ensures the accuracy of the predicted values ​​when displaying the original image size. The core of this method is to optimize the model's recognition and segmentation performance of the target region by reasonably setting the size of the prediction box. In the field of deep learning object detection and segmentation, the size of the prediction box has a significant impact on the performance of the model. If the predicted bounding box is too small, it may miss some key information, leading to inaccurate segmentation; conversely, if the predicted bounding box is too large, it may introduce excessive background noise, similarly affecting the accuracy of the segmentation results. Therefore, by appropriately expanding the bounding box, we can ensure that sufficient target information is included while suppressing background noise interference to a certain extent. This annotation point expansion step not only maintains the accuracy of the annotation but also significantly improves the efficiency of training data preparation in subsequent steps, laying a solid foundation for training high-precision industrial vision inspection models. The successful application of this method opens up new paths for data annotation and model training in the field of deep learning and has important reference value for other vision tasks with similar requirements.

[0027] See Figure 5 In the above steps, specifically in the annotation point expansion step S102, square regions of a format suitable for training the YOLOv8 segmentation model (YOLOv8-segment) are generated. When performing object detection and segmentation based on deep learning models, the rotation state of the target often poses a challenge to the accurate prediction of the deep learning model, especially when dealing with target objects in various poses in natural scenes. In traditional polygon region methods, due to the sensitivity of deep learning models to specific shapes, rotated or tilted targets may lead to a decrease in prediction accuracy. To further improve the performance of the YOLOv8-Segment model in fine-grained recognition, this embodiment of the invention, in the heatmap conversion step S103, converts the annotation points of each square region into annotation points of a circular heatmap; that is, it replaces the original method of converting polygons into binary masks by generating a heatmap mask of the center point of the square. This improved strategy aims to more accurately capture the central region and morphological features of the target object, thereby providing more accurate segmentation results. The circular heatmap annotation points are circular annotation points whose image heatmap values ​​gradually decrease from the center point outwards. The core idea is to use heatmaps to represent the probability of a target object's existence, where each point in the heatmap represents the probability that the location is the center of the target. Compared to traditional polygon methods, heatmaps can more intuitively represent the shape and size of a target, especially for targets of varying sizes and irregular shapes, providing a more detailed and accurate description.

[0028] See Figure 5The method of converting the annotation points of each square region into annotation points for a circular heatmap includes the following steps: First, obtaining the center point of the square through its four corners; then generating a circular heatmap with the center point of the square as the center. Technically, this method first determines the position of the target object's center point, then generates square regions of fixed size around these center points, and generates heatmaps within these square regions, where the heatmap value is highest at the center point and decreases outwards. This heatmap-based mask generation method not only simplifies the annotation process and reduces the complexity of data processing, but also effectively improves the model's ability to capture target details, thereby improving segmentation accuracy and the overall performance of the model. Because circular regions are consistent under rotation transformations, they do not change the deep learning model's ability to identify the target center, thus ensuring the accuracy and consistency of the prediction results. By adopting the improved strategy of using a circular heatmap mask, the influence of the target rotation angle on the accuracy of recognition and segmentation can be effectively eliminated, thereby further enhancing the model's prediction accuracy and robustness. Compared to traditional annotation methods based on polygonal or square regions, circular heatmap masks can provide a more uniform and continuous representation of target regions, unaffected by the rotation state of the object, thus solving the problems of traditional polygonal mask methods.

[0029] After keypoints are labeled using a circular heatmap, in step S104 of the model training process, the labeled images are used to train a keypoint detection model. In this step, the labeled images are input into a deep learning model to train the keypoint detection model. In this embodiment, the keypoint detection model is trained using the segmentation task model of the YOLOv8 deep learning model, rather than directly using the pose model of the YOLOv8 deep learning model. This embodiment primarily uses a YOLOv8-segment deep learning network model, which uses the largest feature map in Head1 as the input to the Mask branch, and obtains a mask feature with shape (B, mask_channel, H, W) after processing by the proto_pred convolutional layer. The prediction branch for the detection boxes is basically the same as that of the YOLOv5 Head in object detection, except that parameters for calculating the mask for each instance are added to the channels of the feature_map. The number of parameters is the same as the mask_channel output by proto_pred. Therefore, for an 80x80 / 40x40 / 20x20 feature_map, the number of channels is: 4 + 1 + num_classes + mask_channel. After obtaining the decoded detection boxes, they are processed by NMS to obtain the final detection boxes. The final instance segmentation result is obtained by multiplying the corresponding mask_channel coeffs with the mask_feature and then weighting the result.

[0030] After training the keypoint detection model using the YOLOv8-segment deep learning network model, this embodiment of the invention also requires deployment acceleration using the OpenVINO framework to achieve fast inference under different hardware environments.

[0031] OpenVINO supports acceleration of a wide range of deep learning models. It effectively improves inference efficiency on Intel hardware (including CPUs, GPUs, VPUs, etc.) through various methods such as graph optimization and automatic kernel library tuning. The specific steps for accelerating models using OpenVINO are as follows: Model conversion steps: First, save the trained PyTorch model as a .pt weight file, then use the ONNX tool to convert it to ONNX format. ONNX is an open model representation format that allows developers to easily migrate and deploy models between different deep learning frameworks.

[0032] OpenVINO optimization steps: The converted ONNX format model is then optimized using OpenVINO's ModelOptimizer. In this stage, the ModelOptimizer performs in-depth analysis of the model, performs graphical transformations, prunes redundancy, and optimizes the hierarchical structure, thereby improving model execution efficiency.

[0033] Inference acceleration steps: The optimized model is deployed and executed on the target device through OpenVINO's Inference Engine. The Inference Engine can intelligently call the optimal computing path suitable for the current hardware, achieving efficient model inference whether on a CPU or GPU.

[0034] Through the above process, not only can an inference speed of 50-80 milliseconds per image be achieved on a CPU, but with GPU support, the model's inference performance can even reach within 50 milliseconds, greatly improving the model's response speed and real-time processing capabilities in practical applications. This is of great significance for application scenarios requiring rapid recognition and segmentation (such as real-time video analysis and robot vision).

[0035] After the detection model is deployed to the detection device, key points of the object being detected can be detected using the key point detection model. In key point detection step S105: the image to be detected is input into the key point detection model for key point detection, and the detected key point coordinates are output. In the instance segmentation key point detection strategy based on heatmaps, the model's output channels are designed as 4 + 1 + num_classes + mask_channel to fully meet the needs of segmenting target regions, class prediction, and key point detection. In this design, the original segmentation model output not only includes the segmentation results for each target region but also the corresponding target class prediction.

[0036] See Figure 2 and Figure 6 In processing the output of the detected key point coordinates, two different implementation methods can be adopted, as follows.

[0037] See Figure 2 and Figure 6 In one embodiment of the present invention, such as Figure 2 As shown, the detection of key point coordinates output includes the following steps: S1051a, Obtain information from each output heatmap.

[0038] S1052a. Obtain the coordinates of the maximum pixel value of each output heatmap based on the information of each output heatmap.

[0039] S1053a, The coordinates of the maximum pixel value of each output heatmap are used as the coordinates of each key point.

[0040] Specifically, in step S1051a, information from each output heatmap is obtained. Since the detection output information of the YOLOv8-segment deep learning network model includes heatmap information of the predicted bounding box and the predicted region, the location information of key points can be determined based on these two types of information. In key point detection using heatmap information, key point detection is achieved by finding the maximum value point of the heat distribution within the segmented target region, which is a crucial step in determining the coordinates of the target key points. In step S1052a, the coordinates of the maximum pixel value point of each output heatmap are obtained based on the information from each output heatmap. Since the region segmented by segmentation is not exactly the same as the actual labeled circular mask, and the predicted region is generally not a perfectly round circle, it is only necessary to find the position with the maximum value in its heatmap as the key point location. In practical applications, finding the maximum heatmap value point of each region through pixel iteration ensures the accuracy of the key point coordinates. After obtaining the coordinates of the maximum value point of the heatmap, in step S1053a, the coordinates of the maximum pixel value point of each output heatmap are output as the coordinates of each key point. Figure 7 In the output, A1, A2, and A3 are the key points for detection.

[0041] In the annotation point expansion step S102, a 50-pixel expansion of the annotation points ensures the size adaptability of the mask region during model training and guarantees the accuracy of the predicted values ​​when displayed at the original image size. The core of this method lies in optimizing the model's performance in target region recognition and segmentation by appropriately setting the size of the prediction box. In the field of deep learning object detection and segmentation, the size of the prediction box has a significant impact on model performance. If the prediction box is too small, some key information may be missed, leading to inaccurate segmentation; conversely, if the prediction box is too large, excessive background noise may be introduced, also affecting the accuracy of the segmentation results. Therefore, appropriate expansion ensures sufficient target information while suppressing background noise interference to a certain extent. Specifically, the 50-pixel expansion not only allows the model to learn the shape and feature information of the target more comprehensively but also helps to more accurately delineate target boundaries during instance segmentation, thereby improving segmentation accuracy. Furthermore, when displaying the segmentation results at the original image size, this prediction box size setting helps maintain consistency between the segmented region and the actual target, ensuring accurate reflection of the prediction results. Thus, it can be distinguished that in the key point detection stage, the deep learning model can output a predicted bounding box containing the key points.

[0042] See Figure 3, in one embodiment of the present invention, before the step of expanding each annotation point into a square area centered at the annotation point by a set number of pixels, the method further comprises the step: image data conversion step S2025: converting the annotated image data in json format into image data in txt text format.

[0043] The designed script automatically converts these point-annotated json files into the annotation label format applicable to the Segment segmentation task. After the point-annotated json is converted to txt, the first 5 values of the data format are the top-left, top-right, bottom-left, and bottom-right (x1, y1, x2, y2) of the processed part frame of the workpiece type; starting from the 6th value, every 3 values represent the (x, y) coordinates of the key point and its visibility. It is only necessary to classify the key points according to the workpiece category and index value to which the key points belong, take their coordinates (x, y) as the center coordinates of the square, expand 50 pixels up, down, left and right, and generate the four vertex coordinates of the square (x1, y1, x2, y2, x3, y3, x4, y4).

[0044] Reference Figure 4 , on the other hand, the present invention further provides an industrial key point detection device based on deep learning, comprising: an image data annotation module, an annotation point expansion module, a heat map conversion module, a model training module, and a key point detection module, wherein the image data annotation module is configured to perform point annotation on the key points of an image in the manner of annotation points.

[0045] The annotation point expansion module is configured to expand each annotation point into a square area centered at the annotation point by a set number of pixels.

[0046] The heat map conversion module is configured to convert the annotation points of each square area into annotation points of a circular heat map.

[0047] The model training module is configured to perform model training on each annotated image to generate a key point detection model.

[0048] The key point detection module is configured to input a to-be-detected image into the key point detection model for key point detection, and output the detected key point coordinates.

[0049] Refer to Figure 7 , in yet another aspect, the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored on the memory and operable on the processor, wherein the processor implements the above-mentioned industrial key point detection method based on deep learning when executing the computer program.

[0050] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the computer device.

[0051] The computer device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the figures are merely examples of a computer device and do not constitute a limitation on the computer device. It may include more or fewer components than illustrated, or combine certain components, or different components. For example, the computer device may also include input / output devices, network access devices, buses, etc.

[0052] The processor referred to can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete pre-defined hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0053] The memory can be an internal storage unit of the computer device, such as a hard drive or RAM. The memory can also be an external storage device of the computer device, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the memory can include both internal and external storage units. The memory is used to store the computer program and other programs and data required by the computer device. The memory can also be used to temporarily store data that has been output or will be output.

[0054] Fourthly, embodiments of the present invention also provide a computer storage medium storing a computer program thereon, which, when executed by a processor, implements the deep learning-based industrial key point detection method described above.

[0055] The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0056] It should be noted that the content contained in the computer-readable medium may be appropriately added to or subtracted from the content as required by the legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium may not include electrical carrier signals and telecommunication signals.

[0057] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0058] The steps in the method of this invention can be adjusted, combined, or deleted according to actual needs.

[0059] The modules or units in the system of this invention can be merged, divided, or deleted according to actual needs.

[0060] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented using electronic pre-installed hardware, or a combination of computer software and electronic pre-installed hardware. Whether these functions are implemented in pre-installed hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0061] In the embodiments provided by this invention, it should be understood that the disclosed apparatus / computer device 600 and method can be implemented in other ways. For example, the apparatus / computer device 600 embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0062] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A deep learning-based method for detecting key points in industry, characterized in that, include: Image data annotation steps: Annotate key points in the image using annotation points; Annotation point expansion steps: Expand each annotation point into a square area centered on the annotation point by a set number of pixels; Heatmap conversion steps: Convert the markers in each square area into markers for the circular heatmap; Model training steps: Train the model on each labeled image to generate a keypoint detection model; Key point detection steps: The image to be detected is input into the key point detection model to detect key points, and the coordinates of the detected key points are output.

2. The deep learning-based industrial key point detection method according to claim 1, characterized in that, The detected key point coordinate output includes: Obtain information from each output heatmap; Obtain the coordinates of the maximum pixel value of each output heatmap based on the information from each output heatmap; The coordinates of the maximum pixel value of each output heatmap are used as the coordinates of each key point.

3. The deep learning-based industrial key point detection method according to claim 1 or 2, characterized in that, The process of converting the markers in each square region into markers for a circular heatmap includes: Find the center point of the square by using its four corners; Generate a circular heat map with the center point of the square as the center.

4. The deep learning-based industrial key point detection method according to claim 3, characterized in that, The circular heat map markers are circular markers where the heat map value of the image gradually decreases from the center point outwards.

5. The deep learning-based industrial key point detection method according to claim 1 or 2, characterized in that, Before expanding each marker point into a square region centered on the marker point by a set number of pixels, the process also includes: Image data conversion steps: Convert the labeled JSON format image data into TXT text format image data.

6. The deep learning-based industrial key point detection method according to claim 1 or 2, characterized in that, The specific steps of expanding each marked point into a square region centered on the marked point according to a set number of pixels are as follows: Each marker is expanded into a square area centered on the marker, with each marker being 50 pixels in size.

7. The deep learning-based industrial key point detection method according to claim 1 or 2, characterized in that, The keypoint detection model is generated by training the segmentation task model of the YOLOv8 deep learning model.

8. A deep learning-based industrial key point detection device, characterized in that, include: Image data annotation module: The image data annotation module is used to annotate key points in an image by using annotation points; Annotation point expansion module: The annotation point expansion module is used to expand each annotation point into a square area centered on the annotation point according to a set number of pixels; Heatmap Conversion Module: The heatmap conversion module is used to convert the markers of each square area into markers of a circular heatmap; Model training module: The model training module is used to train the model on each labeled image to generate a key point detection model; Key point detection module: The key point detection module is used to input the image to be detected into the key point detection model to detect key points and output the coordinates of the detected key points.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the deep learning-based industrial key point detection method as described in any one of claims 1 to 7.

Images