Industrial part identification method, equipment and medium
By acquiring RGB images and depth images of industrial parts, combining feature attribute processing strategies and pre-trained models, the problem of low accuracy in industrial parts recognition is solved, and efficient part recognition and robot grasping is achieved.
Patent Information
- Application Number
- CN202510562914.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-30
- Publication Date
- 2025-08-12
AI Technical Summary
In the prior art, due to the operating environment, operation scenarios and characteristics of industrial parts, the accuracy of industrial parts recognition is poor, affecting the robot grasping efficiency and production line reliability.
By acquiring RGB images and depth images of industrial parts, combining feature attributes such as reflectivity and similarity, different feature processing strategies are used to process part features, and inputting pre-trained category recognition models for identification, using polarization filters and image processing technology to optimize visual features, and optimizing category recognition models with incremental learning algorithms.
It improves the identification accuracy of industrial parts and the accuracy of robot grasping, adapts to changes in complex operation scenarios, and ensures efficient operation of the production line.
Smart Images
Figure CN120472392A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of parts identification in industrial automation, and in particular to an industrial parts identification method, device and medium. Background Art
[0002] In industrial automation production lines, the identification of industrial parts is a critical step in the production process. Existing industrial automation lines use traditional visual recognition methods for industrial part identification, resulting in low accuracy. This, in turn, hinders the robot's ability to grasp industrial parts, ultimately reducing the efficiency and reliability of the automated production line. Summary of the Invention
[0003] In response to the above-mentioned problems and technical needs, the applicant has proposed an industrial parts identification method, equipment and medium to solve the problem of poor industrial parts identification accuracy in the existing technology due to the operating environment, operating scene and inherent characteristics of the industrial parts, and to achieve accurate identification of industrial parts.
[0004] An embodiment of the present application provides a method for identifying industrial parts, the method comprising:
[0005] Acquire characteristic attributes of each industrial part in the current operation scene, as well as an RGB image and a depth image of the industrial part, wherein the characteristic attributes include any one or more of reflectivity and similarity, and the similarity includes a first similarity between parts and a second similarity between the parts and their background;
[0006] Based on the RGB image and the depth image, the part features of each industrial part are obtained. The part features include feature attributes, visual features, and depth features. Different feature attributes correspond to different first feature processing strategies.
[0007] Processing the corresponding part feature using a first feature processing strategy corresponding to the feature attribute to obtain a target part feature;
[0008] The target part features are input into a pre-trained category recognition model to obtain a part category output by the category recognition model, wherein the category recognition model is trained based on part feature samples and part category samples.
[0009] According to the industrial part recognition method provided in the embodiment of the present application, a first feature processing strategy corresponding to a feature attribute is used to process the corresponding part feature to obtain a target part feature, including:
[0010] In the case where the characteristic attribute is similarity, determining that the similarity belongs to the first similarity and / or the second similarity;
[0011] The corresponding part features are processed using a second feature processing strategy corresponding to the similarity to obtain target part features.
[0012] According to the industrial parts recognition method provided by the embodiment of the present application, the visual features include: color features, edge features and texture features;
[0013] The corresponding part features are processed using the second feature processing strategy corresponding to the similarity to obtain the target part features, including:
[0014] When it is determined that the similarity belongs to the first similarity, calculating the similarity of the features of different parts;
[0015] When the similarity is greater than a preset similarity, the color features, edge features, and texture features of the two industrial parts are compared, and the part difference between the two industrial parts is obtained based on the comparison results;
[0016] Enhance the visual features corresponding to the part differences; fuse the visual features and depth features after enhancing the part differences to obtain the target part features.
[0017] According to the industrial parts recognition method provided by the embodiment of the present application, the visual features include: color features, edge features and texture features;
[0018] The corresponding part features are processed using the second feature processing strategy corresponding to the similarity to obtain the target part features, including:
[0019] In cases where it is determined that the similarity belongs to the second similarity, sharpening the visual features of the industrial parts;
[0020] The target part features are obtained by fusing the sharpened visual features and depth features.
[0021] According to the industrial part recognition method provided in the embodiment of the present application, a first feature processing strategy corresponding to a feature attribute is used to process the corresponding part feature to obtain a target part feature, including:
[0022] When the feature attribute is reflective, the visual feature is processed and the target part feature is obtained by fusing the visual feature after image processing with the depth feature;
[0023] And / or, an RGB image and a depth image including the industrial part are acquired based on a polarization filter, and the acquired part features are used as target part features.
[0024] According to the industrial parts recognition method provided in an embodiment of the present application, before inputting the target part features into a pre-trained category recognition model to obtain the part category output by the category recognition model, the method further includes:
[0025] Extracting current operation scene features from the current operation scene;
[0026] Determine whether the extracted current operation scene features belong to known features;
[0027] In the case where it is determined that the extracted current operation scene feature is a known feature, performing the step of inputting the target part feature into a pre-trained category recognition model to obtain the part category output by the category recognition model;
[0028] When it is determined that the extracted features of the current operation scene do not belong to known features, the category recognition model is optimized based on the incremental learning algorithm, and the part category is obtained and output using the optimized category recognition model.
[0029] According to the industrial parts recognition method provided in the embodiment of the present application, after extracting the current operation scene features from the current operation scene, the method further includes:
[0030] Calculate the similarity between the current operation scene characteristics and the preset standard scene characteristics;
[0031] Determine whether the extracted features of the current operation scene are known features, including:
[0032] When it is determined that the similarity is less than the preset similarity, it is determined that the extracted current operation scene feature does not belong to the known feature;
[0033] When it is determined that the similarity is greater than or equal to the preset similarity, it is determined that the extracted current operation scene feature is a known feature.
[0034] According to the industrial parts recognition method provided in an embodiment of the present application, before inputting the target part features into a pre-trained category recognition model to obtain the part category output by the category recognition model, the method further includes:
[0035] Obtain sample images of the static posture of each industrial part in the current operation scene under different viewing angles and different lighting conditions, where the sample images include RGB sample images and depth sample images;
[0036] Part feature samples and part category samples are obtained based on the sample images, and the part feature samples and part category samples are used to train a category recognition model.
[0037] An embodiment of the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of any of the above industrial parts identification methods are implemented.
[0038] An embodiment of the present application further provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of any of the above industrial parts identification methods are implemented.
[0039] The industrial parts identification method, device and medium provided in the embodiments of the present application obtain the characteristic attributes of each industrial part in the current working scene, as well as the RGB image and depth image of the industrial parts, wherein the characteristic attributes include: any one or more of reflectivity and similarity, and the similarity includes: a first similarity between parts and a second similarity between parts and the background in which they are located. The present application fully considers parameters such as the working environment, working scene and the characteristics of the industrial parts themselves in the automated production line, and provides an effective data basis for the subsequent identification of industrial parts; based on the RGB image and the depth image, the part features of each industrial part are obtained, wherein the part features include characteristic attributes, visual features and depth features, and different characteristic attributes correspond to different The first feature processing strategy; use the first feature processing strategy corresponding to the feature attribute to process the corresponding part feature to obtain the target part feature. It can be seen that this application first identifies the part feature, and then adopts different feature processing strategies for different feature attributes to process the part features of industrial parts, providing an effective data basis for subsequent identification; input the target part feature into the pre-trained category recognition model to obtain the part category output by the category recognition model. It can be seen that this application fully considers the parameters such as the working environment, working scene and the characteristics of the industrial parts themselves in the automated production line, and based on this, optimizes the part features extracted for the first time, and then uses the optimized part features (target part features) to identify the part category, ensuring the accuracy of part category identification. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0041] Figure 1 This is one of the flow charts of the industrial parts identification method provided in the embodiment of the present application;
[0042] Figure 2 This is the second flow chart of the industrial parts identification method provided in the embodiment of the present application;
[0043] Figure 3 This is the third flow chart of the industrial parts identification method provided in the embodiment of the present application;
[0044] Figure 4This is the fourth flow chart of the industrial parts identification method provided in the embodiment of the present application;
[0045] Figure 5 This is the fifth flow chart of the industrial parts identification method provided in the embodiment of the present application;
[0046] Figure 6 This is the sixth flow chart of the industrial parts identification method provided in the embodiment of the present application;
[0047] Figure 7 This is the seventh flow chart of the industrial parts identification method provided in the embodiment of the present application;
[0048] Figure 8 This is the eighth flow chart of the industrial parts identification method provided in the embodiment of the present application;
[0049] Figure 9 It is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0050] To make the purpose, technical solutions, and advantages of the embodiments of the present application more clear, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0051] The embodiment of the present application provides a method for identifying industrial parts. The method can be applied to smart terminals and can also be applied to servers. The present application uses the method applied to a server as an example for illustration. This is for illustration only and is not intended to limit the scope of protection of the present application. Some other descriptions in the embodiment are also for illustration only and will not be described one by one later. The specific implementation of the method is as follows: Figure 1 As shown:
[0052] Step 101: Acquire characteristic attributes of each industrial part in the current operation scene, as well as an RGB image and a depth image of the industrial part.
[0053] The characteristic attributes include: any one or more of reflectivity and similarity; and the similarity includes: a first similarity between parts and a second similarity between the parts and their background.
[0054] Specifically, the feature attributes acquired at this stage are the feature attributes corresponding to each industrial part in the known current operation scene.
[0055] Step 102: Obtain the part features of each industrial part based on the RGB image and the depth image.
[0056] Among them, part features include feature attributes, visual features and depth features, and different feature attributes correspond to different first feature processing strategies.
[0057] Specifically, the feature attributes obtained in this stage are the feature attributes of each industrial part in the image obtained by analyzing the RGB image and the depth image. The feature attributes in step 102 belong to the feature attributes in step 101.
[0058] Step 103: Process the corresponding part feature using the first feature processing strategy corresponding to the feature attribute to obtain the target part feature.
[0059] Specifically, processing part features includes processing any one or more of visual features and depth features.
[0060] Specifically, the characteristic attributes of this stage are the characteristic attributes in step 102 .
[0061] The characteristic attributes of each stage can be accurately determined based on their contextual meaning and the content described in the specific embodiment, and will not be described in detail.
[0062] Step 104 : Input the target part features into the pre-trained category recognition model to obtain the part category output by the category recognition model.
[0063] Among them, the category recognition model is trained based on part feature samples and part category samples.
[0064] The industrial parts identification method provided by the embodiment of the present application obtains the characteristic attributes of each industrial part in the current working scene, as well as the RGB image and depth image of the industrial parts, wherein the characteristic attributes include: any one or more of reflectivity and similarity, and the similarity includes: a first similarity between parts and a second similarity between parts and the background in which they are located. The present application fully considers parameters such as the working environment, working scene and the characteristics of the industrial parts themselves in the automated production line, and provides an effective data basis for the subsequent identification of industrial parts; based on the RGB image and the depth image, the part features of each industrial part are obtained, wherein the part features include characteristic attributes, visual features and depth features, and different characteristic attributes correspond to different first features. Feature processing strategy; use the first feature processing strategy corresponding to the feature attribute to process the corresponding part feature to obtain the target part feature. It can be seen that this application first identifies the part features, and then adopts different feature processing strategies for different feature attributes to process the part features of industrial parts, providing an effective data basis for subsequent identification; input the target part features into the pre-trained category recognition model to obtain the part category output by the category recognition model. It can be seen that this application fully considers the parameters such as the working environment, working scene and the characteristics of the industrial parts themselves in the automated production line, and optimizes the part features extracted for the first time based on this, and then uses the optimized part features (target part features) to identify the part category, ensuring the accuracy of part category identification.
[0065] In a specific embodiment, the specific implementation of obtaining the part features of each industrial part based on the RGB image and the depth image includes:
[0066] The RGB image and depth image are input into the pre-trained object detection model to obtain the part features output by the object detection model.
[0067] Among them, the target detection model is trained based on RGB image samples, depth image samples and part feature samples.
[0068] Among them, the part features output by the target inspection model are triples, and the elements in the triples represent feature attributes, visual features, and depth features respectively.
[0069] Specifically, the object detection model is trained using mixed-precision and distributed training methods. Mixed-precision training balances computational speed and accuracy, while distributed training improves training efficiency and scalability. The combination of these two training methods ensures the accuracy of the object detection model, enabling more precise output of part features.
[0070] Specifically, after outputting a triplet using the target detection model, determine whether the content representing the feature attributes in the triplet includes any one or more of reflectivity and similarity; if it is determined that its content includes any one or more of reflectivity and similarity, execute step 103; if it is determined that its content does not include any one or more of reflectivity and similarity, determine the output part feature as the target part feature, and execute step 104.
[0071] This application uses the target detection model to obtain the characteristic attributes of industrial parts, providing an effective data basis for the subsequent optimization of visual features and / or depth features, and further providing an effective data basis for improving the accuracy of category recognition model predictions.
[0072] In a specific embodiment, the first feature processing strategy corresponding to the feature attribute is used to process the corresponding part feature to obtain the specific implementation of the target part feature. Figure 2 As shown:
[0073] Step 201 : When the feature attribute is reflective, image processing is performed on the visual feature.
[0074] Step 202: fuse the visual features and depth features after image processing to obtain target part features.
[0075] The first feature processing strategy includes performing image processing on the visual features and fusing the processed visual features with the depth features. Furthermore, this processing strategy has a mapping relationship with reflectivity.
[0076] Specifically, image processing includes: optimizing the image quality of RGB images through image processing methods such as denoising, sharpening, and color correction to reduce the negative impact of illumination changes, and adjusting the height and contrast of the image through histogram equalization and adaptive filtering technology to better obtain visual features.
[0077] This application removes reflection interference and other noise interference through image processing, and ultimately retains important visual features.
[0078] In a specific embodiment, the specific implementation of processing the corresponding part feature using the first feature processing strategy corresponding to the feature attribute to obtain the target part feature includes:
[0079] An RGB image and a depth image of an industrial part are acquired based on a polarization filter, and the obtained part features are used as target part features.
[0080] Specifically, if the characteristic attribute is determined to be reflective, one or more external polarizers are installed between the industrial part and the camera to enhance contrast, enabling the acquisition of RGB and depth images of the characteristic attribute, which is either non-reflective or poorly reflective. Furthermore, based on the RGB and depth images captured through the polarizing filters, the part features of each industrial part are obtained. The corresponding part features are processed using a first feature processing strategy corresponding to the characteristic attribute to obtain target part features.
[0081] Specifically, processing the RGB image using a polarization filter includes adding a polarization filter inside or outside a camera that collects the RGB image of the industrial part to obtain a polarized RGB image.
[0082] This application solves the reflection problem through a polarizing filter, so that the obtained RGB image can better extract visual features.
[0083] Specifically, for reflective features, a polarized RGB image and depth image can be directly acquired by adding a polarizing filter, and steps 102-104 can be performed based on these images. Alternatively, the RGB image and depth image can be acquired without a polarizing filter, and the target part features can be obtained through image processing. Alternatively, a combination of polarizing filters and image processing can be used to acquire the target part features.
[0084] In a specific embodiment, the first feature processing strategy corresponding to the feature attribute is used to process the corresponding part feature to obtain the specific implementation of the target part feature. Figure 3 As shown:
[0085] Step 301: When the characteristic attribute is similarity, determine whether the similarity belongs to the first similarity and / or the second similarity.
[0086] Step 302: Process the corresponding part features using a second feature processing strategy corresponding to the similarity to obtain target part features.
[0087] The first feature processing strategy further includes: determining a corresponding processing strategy based on a specific category of similarity to process the part feature. The processing strategy has a mapping relationship with the similarity.
[0088] The specific categories include: first similarity and second similarity.
[0089] The content of the element representing the characteristic attribute in the triplet is classified into a specific category.
[0090] In a specific embodiment, the second feature processing strategy corresponding to the similarity is used to process the corresponding part features to obtain the target part features such as Figure 4 As shown:
[0091] Step 401 : When it is determined that the similarity belongs to the first similarity, the similarity of different part features is calculated.
[0092] Step 402 : When the similarity is greater than a preset similarity, the color features, edge features, and texture features of the two industrial parts are compared, and the part difference between the two industrial parts is obtained based on the comparison result.
[0093] Step 403 , enhancing the visual features corresponding to the part differences; fusing the visual features after enhancing the part differences with the depth features to obtain the target part features.
[0094] Specifically, the similarity between the two part features is determined by calculating the Euclidean distance between any two part features. Two part features with a similarity greater than a preset similarity are more difficult to distinguish. In this case, by comparing the color features, edge features, and texture features of the two industrial parts, the detail differences between the two industrial parts are determined, and the detail differences are enhanced to better distinguish the two industrial parts. In a specific embodiment, the second feature processing strategy corresponding to the similarity is used to process the corresponding part features, and the specific implementation of obtaining the target part features is as follows: Figure 5 As shown:
[0095] Step 501: When it is determined that the similarity belongs to the second similarity, sharpen the visual features of the industrial part.
[0096] Step 502: Fusing the sharpened visual features and depth features to obtain target part features.
[0097] Specifically, sharpening visual features is equivalent to sharpening RGB images.
[0098] Specifically, the present application quickly increases the contrast of edge details by adding an isolation layer to the edge of industrial parts and increasing the sharpness of the edge, and two bright and dark lines with very obvious contrast appear on both sides of the edge, making the image look clearer and better distinguishing between industrial parts and background.
[0099] In a specific embodiment, the target part features are input into the pre-trained category recognition model. Before obtaining the part category output by the category recognition model, sample data is obtained to train the category recognition model. The specific implementation method is as follows: Figure 6 As shown:
[0100] Step 601: Acquire sample images of the static postures of various industrial parts in the current operation scene under different viewing angles and different lighting conditions.
[0101] The sample images include RGB sample images and depth sample images.
[0102] Step 602 : Obtain part feature samples and part category samples based on the sample image, and use the part feature samples and part category samples to train a category recognition model.
[0103] This application ensures the diversity and amount of sample data by obtaining sample images of the static postures of various industrial parts under different perspectives and different lighting conditions, providing an effective data basis for training category recognition models.
[0104] Specifically, after obtaining the part category of the industrial part, the present application can also obtain the posture and target position of the industrial part, and then determine the robot's grasping posture and grasping path through the posture and target position of the industrial part, and control the robot to grasp the industrial part and place it at the target position based on the grasping posture and grasping path.
[0105] In a specific embodiment, the target part features are input into the pre-trained category recognition model. Before obtaining the part category output by the category recognition model, scene judgment is performed, and then the category recognition of the target part features is performed. Specifically, Figure 7 As shown:
[0106] Step 701: extract current operation scene features from the current operation scene.
[0107] Step 702 , determine whether the extracted current operation scene feature is a known feature. If so, execute step 703 ; otherwise, execute step 704 .
[0108] Step 703: Input the target part features into the pre-trained category recognition model to obtain the part category output by the category recognition model.
[0109] Step 704 : Optimize the category recognition model based on the incremental learning algorithm, and use the optimized category recognition model to obtain and output the part category.
[0110] Specifically, if the current operating scenario features are determined to be non-known, the classification recognition model is optimized based on the incremental learning algorithm, the current operating scenario, and the corresponding target part features within the current operating scenario. The optimized classification recognition model is then used to determine and output the part category. Of course, a certain amount of historical sample data can also be introduced during this optimization process to optimize the classification recognition model alongside the new sample data.
[0111] Among them, the known features are features that already exist in the category recognition model during the training phase.
[0112] Specifically, the incremental learning algorithm includes: a neural network for self-organizing incremental learning.
[0113] Specifically, the current working scene can be obtained through scene data input by the user, or a picture including the current working scene can be obtained by photographing the current working environment with a camera, etc.
[0114] Specifically, since changes in the operating scene may lead to the extraction results of the target part features, this application collects new data samples for online learning when it is determined that the current operating scene features are not known features, and updates the model parameters of the category recognition model in real time to adapt to the current operating scene. Through the incremental learning algorithm, the weights of the category recognition model are dynamically adjusted so that the model can adapt to new environmental changes (changes in the operating scene) more quickly. By constantly adapting to the new environment in the above manner, the accuracy and robustness of the detection are effectively improved.
[0115] This application uses different category recognition models to identify target part features in different operating scenarios, thereby improving the accuracy of the category recognition model output.
[0116] In a specific embodiment, after extracting the current operation scene feature from the current operation scene, the specific implementation of determining whether the extracted current operation scene feature belongs to a known feature is as follows: Figure 8 As shown:
[0117] Step 801: Calculate the similarity between the current operation scene feature and the preset standard scene feature.
[0118] Step 802: When it is determined that the similarity is less than the preset similarity, it is determined that the extracted current operation scene feature does not belong to the known feature.
[0119] Step 803: When it is determined that the similarity is greater than or equal to the preset similarity, it is determined that the extracted current operation scene feature is a known feature.
[0120] Specifically, the standard scene feature is a scene feature that has been seen before, and there are one or more of them. The similarity between the current task scene feature and the standard scene feature is calculated by calculating the Euclidean distance between the two.
[0121] This application addresses the low accuracy of industrial part recognition through multimodal feature fusion and category recognition models, improving the accuracy and reliability of robot recognition in complex industrial scenarios. Furthermore, by first identifying the features of industrial parts, further optimizing the features based on their characteristic attributes, and using the optimized features to determine the category, the accuracy of category determination is ensured. Furthermore, this provides an effective data foundation for subsequent robot grasping.
[0122] Figure 9 An example of a physical structure diagram of an electronic device is shown below. Figure 9As shown, the electronic device may include: a processor (processor) 901, a communication interface (Communications Interface) 902, a memory (memory) 903, and a communication bus 904. The processor 901, the communication interface 902, and the memory 903 communicate with each other via the communication bus 904. The processor 901 may call logic instructions in the memory 903 to execute the industrial part identification method.
[0123] In addition, the logic instructions in the above-mentioned memory 903 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.
[0124] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the industrial parts identification methods provided by the above methods.
[0125] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is used to execute the industrial parts identification method provided in the above embodiments when the computer program is executed by a processor.
[0126] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.
[0127] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.
[0128] Finally, it should be noted that the above description is merely a preferred embodiment of the present application and the present application is not limited to the above embodiments. It is understood that other improvements and variations directly derived or conceived by those skilled in the art without departing from the spirit and concept of the present application should be considered to be included within the scope of protection of the present application.
Claims
1. A method for identifying industrial parts, characterized in that: The method comprises: Acquire characteristic attributes of each industrial part in the current operation scene, as well as an RGB image and a depth image of the industrial part, wherein the characteristic attributes include any one or more of reflectivity and similarity, and the similarity includes a first similarity between parts and a second similarity between the parts and their background; Based on the RGB image and the depth image, the part features of each industrial part are obtained. The part features include feature attributes, visual features, and depth features. Different feature attributes correspond to different first feature processing strategies. Processing the corresponding part feature using a first feature processing strategy corresponding to the feature attribute to obtain a target part feature; The target part features are input into a pre-trained category recognition model to obtain a part category output by the category recognition model, wherein the category recognition model is trained based on part feature samples and part category samples.
2. The method for identifying industrial parts according to claim 1, characterized in that: The first feature processing strategy corresponding to the feature attribute is used to process the corresponding part feature to obtain the target part feature, including: In the case where the characteristic attribute is similarity, determining that the similarity belongs to the first similarity and / or the second similarity; The corresponding part features are processed using a second feature processing strategy corresponding to the similarity to obtain target part features.
3. The method for identifying industrial parts according to claim 2, characterized in that: The visual features include: color features, edge features and texture features; The corresponding part features are processed using the second feature processing strategy corresponding to the similarity to obtain the target part features, including: When it is determined that the similarity belongs to the first similarity, calculating the similarity of the features of different parts; When the similarity is greater than a preset similarity, the color features, edge features, and texture features of the two industrial parts are compared, and the part difference between the two industrial parts is obtained based on the comparison results; Enhance the visual features corresponding to the part differences; fuse the visual features and depth features after enhancing the part differences to obtain the target part features.
4. The method for identifying industrial parts according to claim 2, characterized in that: The visual features include: color features, edge features and texture features; The corresponding part features are processed using the second feature processing strategy corresponding to the similarity to obtain the target part features, including: In cases where it is determined that the similarity belongs to the second similarity, sharpening the visual features of the industrial parts; The target part features are obtained by fusing the sharpened visual features and depth features.
5. The method for identifying industrial parts according to any one of claims 1 to 4, characterized in that: The first feature processing strategy corresponding to the feature attribute is used to process the corresponding part feature to obtain the target part feature, including: When the feature attribute is reflective, the visual feature is processed and the target part feature is obtained by fusing the visual feature after image processing with the depth feature; And / or, an RGB image and a depth image including the industrial part are acquired based on a polarization filter, and the acquired part features are used as target part features.
6. The method for identifying industrial parts according to any one of claims 1 to 4, characterized in that: Before inputting the target part features into a pre-trained category recognition model to obtain the part category output by the category recognition model, the method further includes: Extracting current operation scene features from the current operation scene; Determine whether the extracted current operation scene features belong to known features; In the case where it is determined that the extracted current operation scene feature is a known feature, performing the step of inputting the target part feature into a pre-trained category recognition model to obtain the part category output by the category recognition model; When it is determined that the extracted features of the current operation scene do not belong to known features, the category recognition model is optimized based on the incremental learning algorithm, and the part category is obtained and output using the optimized category recognition model.
7. The method for identifying industrial parts according to claim 6, characterized in that: After extracting the current job scene features from the current job scene, it also includes: Calculate the similarity between the current operation scene characteristics and the preset standard scene characteristics; Determine whether the extracted features of the current operation scene are known features, including: When it is determined that the similarity is less than the preset similarity, it is determined that the extracted current operation scene feature does not belong to the known feature; When it is determined that the similarity is greater than or equal to the preset similarity, it is determined that the extracted current operation scene feature is a known feature.
8. The method for identifying industrial parts according to any one of claims 1 to 4, characterized in that: Before inputting the target part features into a pre-trained category recognition model to obtain the part category output by the category recognition model, the method further includes: Obtain sample images of the static posture of each industrial part in the current operation scene under different viewing angles and different lighting conditions, where the sample images include RGB sample images and depth sample images; Part feature samples and part category samples are obtained based on the sample images, and the part feature samples and part category samples are used to train a category recognition model.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the industrial part identification method according to any one of claims 1 to 8 are implemented.
10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the industrial part identification method according to any one of claims 1 to 8 are implemented.