A recognition method based on laser marking positioning and a recognition electronic device thereof

Abstract

The application provides a kind of identification method and identification electronic device based on laser marking positioning, with laser source, comprising the following steps: S1, obtaining the content to be marked and positioned;S2, judge whether the area of marking and positioning in the content to be marked and positioned is coding information;S3, if yes, the recognition algorithm scans the coding information and extracts the corresponding content;If not, the recognition algorithm carries out picture interception on the area of marking and positioning in the file;S4, target detection is carried out on the intercepted picture to obtain a prediction result graph;S5, the prediction result graph is identified by local or cloud OCR algorithm, and the corresponding content is output.The above technical scheme solves the problem that the existing business content needs to be input by multiple devices or manual input, and the operation process is complex.

Description

Technical Field

[0001] This invention relates to the field of laser marking recognition technology, and in particular to a recognition method and electronic device based on laser marking positioning. Background Technology

[0002] The input of business information includes coded information (including barcodes and QR codes) and other business content. Existing laser-marked positioning-based identification methods and electronic devices only address the input of coded information for some business content and cannot meet the requirements for inputting other business content. Other business content includes different business information and different processing methods. Different business information comes from different sources. For example, one business requires inputting a customer's code, ID number, and the type of business being processed. The code comes from the customer's mobile phone, while the ID information comes from the customer's ID number. The processing method includes optional fields on paper forms and optional fields on electronic forms. The current business system requires multiple devices or manual input to solve this problem. For example, codes are scanned with a barcode scanner, ID numbers are read using an ID card reader, and optional fields on paper forms are entered manually. The entire operation is complex and requires multiple devices. Summary of the Invention

[0003] Therefore, there is a need to provide a recognition method based on laser marking and positioning to solve the problem that the existing methods of inputting business content require multiple devices or manual input, which is complicated in operation.

[0004] To achieve the above objectives, the present invention provides a laser marking-based identification method, applied to an identification electronic device, wherein the identification electronic device includes a laser source, and includes the following steps:

[0005] S1. Obtain the content to be labeled and positioned;

[0006] S2. Determine whether the area to be marked within the content to be marked is coded information;

[0007] S3. If the area to be marked within the content to be marked is coded information, the laser source performs laser marking on the area to be marked within the content, and the recognition electronic device scans the coded information within the marking area to extract the corresponding content; if the area to be marked within the content to be marked is non-coded information, the laser source performs laser marking on the area to be marked within the content, and the recognition electronic device extracts the image within the marking area.

[0008] S4. Perform object detection on the captured image to obtain the prediction result;

[0009] S5. Based on the prediction results, call the corresponding local or cloud-based OCR algorithm to recognize the captured image and obtain the corresponding content output by the local or cloud-based OCR algorithm.

[0010] Furthermore, the content to be labeled and positioned can be paper content or electronic content.

[0011] Furthermore, the electronic content includes content displayed on the screen of an electronic device and content displayed on the screen of another electronic device, which are obtained through a wired or wireless connection to the electronic device. The other electronic devices include computers, cash registers, or mobile phones with displays.

[0012] Furthermore, the content to be tagged and located consists of multiple pages, which can be automatically or manually turned. This method also includes the following steps:

[0013] Automatically recognize page-turning information. After page-turning is completed, the laser source marks and identifies the encoded or non-encoded information of the marked and positioned areas in the same location of multiple pages.

[0014] Furthermore, the content to be tagged and located consists of multiple pages, which can be automatically or manually turned. This method also includes the following steps:

[0015] Automatically recognize page-turning information. After page-turning is completed, the laser source marks and identifies the encoded or non-encoded information of the marked and positioned areas in different locations of the multi-page content.

[0016] Furthermore, determining whether the area to be marked within the content to be marked is coded information includes the following steps:

[0017] By judging the different pressed states of a single button, it is possible to determine whether the area to be marked within the content to be marked is encoded information.

[0018] Further, step S4 includes an object detection model, which comprises a backbone network, a bottleneck network, and a decoupling head. The backbone network includes a Focus module and four Darknet modules. The bottleneck network includes three branches. The captured image is input into the backbone network of the object detection model built based on the YOLOX network, passing through the Focus module and the four Darknet modules in series, and outputting three effective feature maps of different sizes after passing through the second, third, and fourth Darknet modules. The effective feature maps are then input into the three branches of the bottleneck network, and after adaptive feature correction, three enhanced feature maps are output. The three enhanced feature maps are then input into the corresponding decoupling heads. The decoupling heads convolve the input enhanced feature maps, decomposing them into regression branches, confidence branches, and classification branches, and outputting three predicted targets. The predicted targets include the coordinate information of the predicted bounding box, foreground and background, and object type. The three predicted targets are stacked to output a prediction result image.

[0019] Furthermore, the OCR algorithm recognition in step S5 includes:

[0020] S51. Construct a convolutional neural network (CNN) model, which includes a StemConv module, a convolutional layer, a global average pooling layer, and a representative fully connected layer connected in sequence.

[0021] S52. Select text information from specific areas within the content to be labeled and located, crop the image, input it into the convolutional neural network (CNN) model for training and testing, and adjust the network parameters in the CNN model.

[0022] S53. When the loss of the Convolutional Neural Network (CNN) model training and testing converges, stop training and obtain the CNN recognition model.

[0023] S54. Use the CNN recognition model generated in S53 to recognize the predicted result image.

[0024] Furthermore, the convolutional layer includes four depthwise separable convolutional layers, wherein the first three depthwise separable convolutional layers have 3×3 convolutional kernels with a convolution stride of 1×1; the last depthwise separable convolutional layer has a 5×5 convolutional kernel with a convolution stride of 1×1; or the convolutional layer includes two depthwise separable convolutional layers, a DW convolutional layer and a PW convolutional layer, wherein the depthwise separable convolutional layer has a 3×3 convolutional kernel with a convolution stride of 1×1.

[0025] An identification electronic device includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the program, it implements the above-described identification method based on laser marking positioning.

[0026] Unlike existing technologies, the above technical solution uses laser source marking and positioning combined with different recognition algorithms to input coded information and other business content. For coded information, the laser source marks the marked area within the content to be marked, and the recognition electronic device scans the coded information within the marked area to extract the corresponding content. For other business content input, the laser source marks the marked area within the content, the recognition electronic device extracts an image from the marked area, and then performs target detection and OCR algorithm recognition on the extracted image to output the corresponding other business content. In use, it first determines whether the marked area within the content to be marked is coded information, solving the problem of the complex operation process of existing methods that require multiple devices or manual input to input business content. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the process of the present invention;

[0028] Figure 2 This is a flowchart illustrating the specific implementation method;

[0029] Figure 3 This is a schematic diagram illustrating the target detection model process for a specific implementation method.

[0030] Figure 4 This is a schematic diagram illustrating the OCR algorithm recognition process in a specific implementation. Detailed Implementation

[0031] To explain in detail the technical content, structural features, objectives, and effects of the technical solution, the following description is provided in conjunction with specific embodiments and accompanying drawings.

[0032] In this document, the term "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The term "embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment, nor does it specifically limit its independence or connection with other embodiments. In principle, in this application, as long as there are no technical contradictions or conflicts, the technical features mentioned in each embodiment can be combined in any way to form corresponding implementable technical solutions.

[0033] Unless otherwise defined, the technical terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the use of related terms herein is merely for the purpose of describing particular embodiments and is not intended to limit this application.

[0034] In the description of this application, the term "and / or" is used to describe the logical relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A exists, B exists, and A and B exist simultaneously. Additionally, the character " / " in this document generally indicates that the preceding and following objects have an "or" logical relationship.

[0035] In this application, terms such as “first” and “second” are used only to distinguish one entity or operation from another, and do not necessarily require or imply any actual quantity, hierarchy or order relationship between these entities or operations.

[0036] Unless otherwise specified, the use of terms such as “comprising,” “including,” “having,” or other similar expressions in this application is intended to cover non-exclusive inclusion, which does not exclude the presence of additional elements in a process, method, or product that includes the stated elements, such that a process, method, or product that includes a list of elements may include not only those defined elements but also other elements not expressly listed, or elements inherent to such a process, method, or product.

[0037] Similar to the understanding in the Examination Guidelines, in this application, expressions such as "greater than," "less than," and "exceeding" are understood to exclude the stated number; expressions such as "above," "below," and "within" are understood to include the stated number. Furthermore, in the description of the embodiments in this application, "multiple" means two or more (including two), and similar expressions related to "multiple" are also understood in this way, such as "multiple groups" and "multiple times," unless otherwise explicitly specified.

[0038] In the description of the embodiments of this application, the space-related expressions used, such as "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "vertical," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential," indicate the orientation or positional relationship based on the orientation or positional relationship shown in the specific embodiments or drawings. They are only for the purpose of describing the specific embodiments of this application or for the reader's understanding, and do not indicate or imply that the device or component referred to must have a specific position, a specific orientation, or be constructed or operated in a specific orientation. Therefore, they should not be construed as limitations on the embodiments of this application.

[0039] Unless otherwise expressly specified or limited, the terms "installation," "connection," "linking," "fixing," and "setting," as used in the description of the embodiments of this application, should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral setting; it can be a mechanical connection, an electrical connection, or a communication connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be the internal connection of two components or the interaction between two components. For those skilled in the art to which this application pertains, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0040] See Figures 1 to 4 As shown, this embodiment provides a laser marking-based identification method applied to an electronic identification device. The electronic identification device includes a laser source, which marks and locates coded information and other business content input. For coded information, the laser source marks the marked area within the content to be marked, and the electronic identification device scans the coded information within the marked area to extract the corresponding content. For other content, the laser source marks the marked area within the content, the electronic identification device extracts an image from the marked area, and then performs target detection and OCR algorithm recognition on the extracted image to output the corresponding other business content. In use, it first determines whether the marked area within the content to be marked is coded information, solving the problem of the complex operation process requiring multiple devices or manual input for business content input.

[0041] See Figure 1 As shown, the specific implementation method of this embodiment includes the following steps:

[0042] The identification electronic device includes a laser source, and includes the following steps:

[0043] S1. Obtain the content to be labeled and positioned;

[0044] S2. Determine whether the area to be marked within the content to be marked is coded information;

[0045] S3. If the area to be marked within the content to be marked is coded information, the laser source performs laser marking on the area to be marked within the content, and the recognition electronic device scans the coded information within the marking area to extract the corresponding content; if the area to be marked within the content to be marked is non-coded information, the laser source performs laser marking on the area to be marked within the content, and the recognition electronic device extracts the image within the marking area.

[0046] S4. Perform object detection on the captured image to obtain the prediction result;

[0047] S5. Based on the prediction results, call the corresponding local or cloud-based OCR algorithm to recognize the captured image and obtain the corresponding content output by the local or cloud-based OCR algorithm.

[0048] Finally, based on the results, the corresponding business operations are performed. For example, operations related to the ID card number are performed.

[0049] The encoded information mentioned in the above steps can be a QR code, barcode, or similar encoded information with hidden content. The area to be marked within the content to be marked in the above steps is identified by a prominent identification pattern, which includes, but is not limited to, symbols such as "", "|", "+", and "()". The laser source's method of alerting the prominent identification pattern includes, but is not limited to, flashing and constant illumination. The color of the laser source includes, but is not limited to, red, green, white, and purple.

[0050] The content to be marked and positioned mentioned above can be paper content or electronic content. The electronic content can be obtained by connecting the identification electronic device to other electronic devices via wired or wireless communication, displaying the content from those other electronic devices on the identification electronic device's screen, or it can be content displayed on the screen of other electronic devices. The laser source switches its emission frequency synchronously to mark the coded or non-coded information within the marking and positioning area on the paper content or the display screen. Here, "other devices" refers to computers, cash registers, mobile phones, etc., that have displays.

[0051] The content to be marked and located may be a single page or multiple pages. When the content is multiple pages, it can be automatically or manually turned. The laser source marks the encoded or non-encoded information of the marking and location areas at the same or different positions on different pages. Manual page turning here refers to manually turning over paper content or electronic content (such as PDF files). Automatic page turning can be timed by setting a page turning duration within the program of the device containing the electronic content; or it can be set within the device containing the electronic content, where the electronic recognition device executes the marking of encoded or non-encoded information on the current page, and only turns the page after the scanning extraction or image capture is completed. This realizes automatic page turning of electronic content and automatic input of business content. Furthermore, the electronic recognition device automatically identifies page-turning information. After page-turning is completed, the laser source marks and identifies the encoded or non-encoded information of the marked and positioned areas in the same or different positions of multiple pages. This is achieved in the following way: by detecting the proportion of changes in the page (the amount of pixel changes in the image during page-turning), if the amount of changes in the page is greater than a certain preset value, it is considered a page turn. Then, if the amount of changes is less than a second preset value within a certain period of time, it is considered that the page turn is complete.

[0052] Steps S2 and S3 above can be determined by a computer program to determine whether the information is coded, and whether to adjust the recognition algorithm by executing the computer program; alternatively, it can be determined by judging different pressed states of a single key to determine whether the area to be marked within the content to be marked is coded information. In this case, the recognition algorithm is manually adjusted, such as by switching different recognition algorithms using different keys, or by switching using the same key, such as by switching the number of times the same key is pressed. In practical applications, a single click can be used for the first recognition algorithm, and a double click for the second recognition algorithm (see...). Figure 2 (as shown); or by the duration of the same button press, in practical applications, a short press can be used for the first recognition algorithm, and a long press for the second recognition algorithm.

[0053] See Figure 3 As shown, the specific implementation of step S4 above includes an object detection model, which includes a backbone network, a bottleneck network, and a decoupling head. The backbone network includes a Focus module and four Darknet modules. The bottleneck network includes three branches. The captured image is input into the backbone network of the object detection model built based on the YOLOX network. After passing through the Focus module and the four Darknet modules in series, three effective feature maps of different sizes are output, which have passed through the second, third, and fourth Darknet modules. The effective feature maps are input into the three branches of the bottleneck network respectively. After adaptive feature correction, three enhanced feature maps are output. The three enhanced feature maps are then input into the corresponding decoupling heads. The decoupling heads perform convolution on the input enhanced feature maps, decomposing them into regression branches, confidence branches, and classification branches, and output three predicted targets. The predicted targets include the coordinate information of the predicted bounding box, foreground and background, and object type. The three predicted targets are stacked to output a prediction result image.

[0054] The Focus module is used to obtain a value for every pixel in the input sample image, acquire four independent feature layers, and then stack the four feature layers to concentrate the width and height information into channel information, expanding the input channels fourfold. The concatenated feature layer becomes twelve channels compared to the original three channels, which enhances the feature extraction capability. Then, after convolution, batch normalization and activation by the Conv2D_BN_SiLU module, it is output to the first Darknet module.

[0055] The Conv2D_BN_SiLU module includes convolutional units, batch regularization units, and activation function units connected in series along the forward propagation direction. The activation function is the SiLU function. The first, second, third, and fourth Darknet modules each include a Conv2D_BN_SiLU module and a CspLayer module connected in series along the forward propagation direction. The CspLayer module is used to perform residual calculation after 1×1 convolution of the input feature image; the output after residual calculation is stacked with the output after only 1×1 convolution, and then subjected to another 1×1 convolution to obtain the final output. An SPPBottleneck module is added between the Conv2D_BN_SiLU module and the CspLayer module of the fourth Darknet module. The SPPBottleneck module pools the feature images using pooling kernels of different sizes, which can significantly improve the receptive field of the network. The CspLayer modules of the second, third, and fourth Darknet modules each output an effective feature map, namely the first feature map, the second feature map, and the third feature map, which are fed into the bottleneck network. The shapes of the first, second, and third feature maps are f1 = (80, 80, 256), f2 = (40, 40, 512), and f3 = (20, 20, 1024), respectively. The first, second, and third feature maps are input into the bottleneck network. The three branches of the algorithm perform adaptive feature correction and output three enhanced feature maps. These three enhanced feature maps are then input to their respective decoupling heads. Each decoupling head, YoloHead, performs convolution decomposition on the input enhanced feature map into a regression branch, a confidence branch, and a classification branch, outputting three prediction targets. The acquisition of prediction targets includes the coordinate information of the prediction box, foreground and background, and the object class. The three prediction targets are then stacked to output the prediction result map. Specifically, each enhanced feature map can acquire three prediction targets: Reg(h,w,4) to determine the regression parameters (i.e., coordinate information) of each feature point (adjusting the regression parameters yields the prediction box); Obj(h,w,1) to determine whether each feature point contains an object; and Cls(h,w,num_classes) to determine the object class contained in each feature point. The three prediction targets are stacked to obtain the prediction result of each feature layer as (h, w, 4+1+num_classes), where 4 represents the regression parameter of each feature point, 1 represents whether an object is contained, num_classes represents the types of objects contained in each feature point, and (h, w) represents the length and width of the prediction box.

[0056] See Figure 4As shown, the specific implementation of the OCR algorithm recognition in step S5 includes:

[0057] S51. Construct a convolutional neural network (CNN) model, which includes a StemConv module, a convolutional layer, a global average pooling layer, and two fully connected layers connected in sequence.

[0058] S52. Select text information from specific areas within the content to be labeled and located, crop the image, input it into the convolutional neural network (CNN) model for training and testing, and adjust the network parameters in the CNN model.

[0059] S53. When the loss of the Convolutional Neural Network (CNN) model training and testing converges, stop training and obtain the CNN recognition model.

[0060] S54. Use the CNN recognition model generated in S53 to recognize the predicted result image.

[0061] The OCR algorithm formed by the CNN recognition model trained by the convolutional neural network (CNN) model has good generalization ability and fault tolerance. A Convolutional Neural Network (CNN) model can be constructed by sequentially connecting a Stem Conv module, convolutional layers, a Global Average Pooling (GAP) layer, and two fully connected layers (FC). The Stem Conv module uses the h-swish activation function. The convolutional layers include four depthwise separable convolutional layers (DepthSepConv). The first three depthwise separable convolutional layers have 3×3 kernels with a stride of 1×1; the last depthwise separable convolutional layer has a 5×5 kernel with a stride of 1×1. The output feature maps of the Stem Conv module and the four depthwise separable convolutional layers have shapes of (16×112×112), (64×56×56), (128×28×28), (256×14×14), and (512×7×7), respectively. The first fully connected layer uses the h-swish activation function, and the number of channels in the first and second fully connected layers are 1280 and 1000, respectively. In some embodiments, a Convolutional Neural Network (CNN) model can be constructed by sequentially connecting a Stem Conv module, convolutional layers, a Global Average Pooling (GAP) layer, and two fully connected layers (FC). The Stem Conv module uses the h-swish activation function. The convolutional layers include two depthwise separable convolutional layers (DepthSepConv), a DW convolutional layer, and a PW convolutional layer. The depthwise separable convolutional layers have 3×3 kernels and a 1×1 stride. The two depthwise separable convolutional layers output feature maps with shapes of (64×56×56) and (128×28×28), respectively. Both the DW and PW convolutional layers use the h-swish activation function. The first fully connected layer uses the ReLU activation function, and the second fully connected layer uses h-sigmoid to represent the HardSigmoid activation function. Alternatively, an SE layer can be added after the PW convolutional layer to extract multi-scale spatial information, achieving more spatial feature fusion while improving sensitivity to channel features.

[0062] An identification electronic device includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the program, it implements the above-described identification method based on laser marking positioning.

[0063] It should be noted that although the above embodiments have been described herein, this does not limit the scope of patent protection for this invention. Therefore, any changes and modifications made to the embodiments described herein based on the innovative concept of this invention, or equivalent structural or procedural transformations made using the description and drawings of this invention, directly or indirectly applying the above technical solutions to other related technical fields, are all included within the scope of protection of this invention.

Claims

1. A method for identifying based on laser marking positioning, applied to an electronic device, and characterized in that: The recognition electronic device comprises a laser source, comprising the following steps: S1, obtaining the content to be marked and positioned; S2, judging whether the marked and positioned area in the content to be marked and positioned is coding information; S3, if the marked and positioned area in the content to be marked and positioned is coding information, the laser source performs laser marking on the marked and positioned area in the content to be marked and positioned, and the recognition electronic device scans and extracts the corresponding content in the coding information in the marked area; if the marked and positioned area in the content to be marked and positioned is non-coding information, the laser source performs laser marking on the marked and positioned area in the content, and the recognition electronic device performs picture interception in the marked area; S4, target detection is performed on the intercepted picture to obtain a prediction result; S5, according to the prediction result, a corresponding local or cloud OCR algorithm is called to recognize the intercepted picture, and corresponding content output by the local or cloud OCR algorithm is obtained; The content to be marked and positioned is multi-page content, and the multi-page content is automatically or manually turned over, and the method further comprises the following steps: The electronic recognition device automatically recognizes the page turning information, including detecting the change amount ratio of the page, and determining that the page is turned over when the change amount of the page is greater than a certain preset value, and then determining that the page turning is completed when the change amount is less than a second preset value within a period of time; the detection of the change amount ratio of the page is how many pixel points of the picture change during the page turning process.

2. The identification method based on laser marking positioning according to claim 1, characterized in that: The content to be marked and positioned is paper content or electronic content.

3. The identification method based on laser marking positioning according to claim 2, characterized in that: The electronic content includes the content displayed on the screen of the electronic device and the content displayed on the screen of other electronic devices, which are obtained through wired or wireless connection of the electronic device, and the other electronic devices include computers, cash registers or mobile phones with displays.

4. The identification method based on laser marking positioning according to claim 1, characterized in that: After the page turning is completed, the laser source marks and positions the coding information or non-coding information of the marked and positioned area at the same position of the multi-page content and recognizes it.

5. The identification method based on laser marking positioning according to claim 1, characterized in that: After the page turning is completed, the laser source marks and positions the coding information or non-coding information of the marked and positioned area at different positions of the multi-page content and recognizes it.

6. The identification method based on laser marking positioning according to claim 1, characterized in that: The judgment of whether the marked and positioned area in the content to be marked and positioned is coding information comprises the following steps: The judgment of whether the marked and positioned area in the content to be marked and positioned is coding information is realized by judging different pressing states of a single key.

7. The identification method based on laser marking positioning according to claim 1, characterized in that: The step S4 comprises a target detection model comprising a backbone network, a bottleneck network and a decoupling head; the backbone network comprises a Focus module and four darknet modules; the bottleneck network comprises three branches; the intercepted picture is input into the backbone network of the target detection model based on the YOLOX network, and is sequentially input into the Focus module and the four darknet modules, and three effective feature maps of different sizes output by the second darknet module, the third darknet module and the fourth darknet module; the effective feature maps are input into the three branches of the bottleneck network respectively, and three enhanced feature maps are output after adaptive feature correction; then the three enhanced feature maps are input into the corresponding decoupling heads respectively; the decoupling head convolves the input enhanced feature maps, decomposes them into a regression branch, a confidence branch and a classification branch, and outputs three prediction targets, the prediction targets comprising coordinate information of a prediction box, foreground and background, and object categories; the three prediction targets are stacked to output a prediction result.

8. The identification method based on laser marking positioning according to claim 1, characterized in that: The step S5 comprises an OCR algorithm recognition comprising: S51, a convolutional neural network (CNN) model is built, the convolutional neural network (CNN) model comprising a StemConv module, a convolutional layer, a global average pooling layer and a full connection layer connected in sequence; S52, the text information in a specific area of the content in the content to be marked and positioned is screened out and input into the convolutional neural network (CNN) model for training and testing, and the network parameters in the convolutional neural network (CNN) model are adjusted; S53, when the loss of the convolutional neural network (CNN) model training and testing converges, the training is stopped, and a CNN recognition model is obtained; S54, the CNN recognition model generated in S53 is used to recognize the prediction result picture.

9. The identification method based on laser marking positioning according to claim 8, characterized in that: The convolutional layer comprises four depth separable convolutional layers, wherein the first three depth separable convolutional layers are 3*3 convolutional kernels with a convolution step of 1*1; the last depth separable convolutional layer is a 5*5 convolutional kernel with a convolution step of 1*1; or the convolutional layer comprises two depth separable convolutional layers, a DW convolutional layer and a PW convolutional layer, the depth separable convolutional layer is a 3*3 convolutional kernel with a convolution step of 1*1. 10.An identification electronic device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: The processor executes the program to realize the identification method based on laser marking and positioning according to any one of claims 1-9.

Images