Spray code identification method and device based on artificial intelligence, equipment and storage medium
Through the deep learning model based on artificial intelligence, the inaccuracy and efficiency of traditional manual detection methods is solved, and high-precision, automated and intelligent incantation recognition is achieved.
Patent Information
- Application Number
- CN202510048726.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-13
- Publication Date
- 2025-06-03
AI Technical Summary
Traditional manual detection methods have accuracy and efficiency problems in inkjet identification, which is difficult to meet the needs of large-scale production lines and increases production costs.
The deep learning model based on artificial intelligence is used for ink coding identification, and precise identification of ink coding is achieved by creating detection templates, collecting and annotating ink coding samples, and generating and optimizing deep learning models.
It improves the accuracy and reliability of inkjet identification, improves the recognition efficiency, reduces labor costs, and realizes the automation and intelligence of inkjet identification.
Smart Images
Figure CN120088535A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of artificial intelligence technology, and particularly to an inkjet code recognition method, device, equipment and storage medium based on artificial intelligence. Background Art
[0002] In traditional implementation methods, product inkjet coding is an important link in product quality control and traceability. The inkjet code usually contains key information such as production date, batch number, serial number, etc., which is crucial for product management, inventory control, and consumer rights protection. Traditional product inkjet code detection mainly relies on manual visual inspection. However, this detection method has many defects.
[0003] First of all, manual detection is easily affected by human factors, making it difficult to ensure the accuracy and consistency of detection results. Long hours of work can cause visual fatigue of inspectors, thus increasing the risk of missed and false detections. In addition, there may be differences in detection standards among different inspectors, further affecting the reliability of detection results. Secondly, manual detection is inefficient and difficult to meet the needs of large-scale production lines. As the degree of automation of production lines continues to increase, manual detection has gradually become a bottleneck in the production process, restricting the improvement of production efficiency. Especially in high-value product industries such as liquor, each set of products needs to undergo strict inkjet code detection to ensure the accuracy and integrity of information. Manual detection is not only time-consuming and laborious but also increases production costs.
[0004] In view of the above problems, image recognition technology has been tried in the prior art for inkjet code detection. However, most of these technologies only stay at the level of simple recognition of the presence or absence of inkjet codes and cannot accurately recognize and compare each character. Therefore, there is an urgent need for a method to improve the accuracy and reliability of inkjet code recognition. Summary of the Invention
[0005] Based on this, the present application provides an inkjet code recognition method, device, equipment and storage medium based on artificial intelligence to solve the problems existing in the prior art.
[0006] In a first aspect, an inkjet code recognition method based on artificial intelligence is provided, and the method includes:
[0007] Create a detection template and set a detection area, where the detection area corresponds to the inkjet code position;
[0008] Collect a first inkjet code sample and a corresponding first annotation file as a training set, and generate a deep learning model based on the training set, where the first annotation file includes feature information and recognition results corresponding to the first inkjet code sample;
[0009] Collect the second inkjet coding sample and the corresponding second annotation file as the test set, test the deep learning model based on the test set, and output the second inkjet coding samples that fail the test. Train the deep learning model based on the second inkjet coding samples that fail the test and the corresponding second annotation files to generate an optimized deep learning model;
[0010] Input the inkjet coding to be detected into the optimized deep learning model, and output the recognition result of the inkjet coding to be detected through the optimized deep learning model.
[0011] According to an implementable manner in the embodiments of the present application, the method further includes:
[0012] Place the inkjet coding of the target product in the direct position of the industrial camera, and set the annular light source to evenly irradiate the inkjet coding of the target product;
[0013] Obtain the best recognition effect by adjusting the focal length of the camera and the color and brightness of the annular light source, and collect the inkjet coding based on the best recognition effect.
[0014] According to an implementable manner in the embodiments of the present application, the target product includes wine bottles.
[0015] According to an implementable manner in the embodiments of the present application, collect the first inkjet coding sample and the corresponding first annotation file as the training set, and generate a deep learning model based on the training set. The first annotation file includes feature information and recognition results corresponding to the first inkjet coding sample, including:
[0016] Collect the first inkjet coding sample and the corresponding first annotation file as the training set. The first annotation file includes feature information and recognition results. The recognition results include qualified inkjet coding or unqualified inkjet coding. The unqualified inkjet coding includes one or more of skewed inkjet coding, incomplete inkjet coding, inkjet coding position exceeding the preset range, repeated inkjet coding, and non-existent inkjet coding. The feature information represents the features on the inkjet coding corresponding to the recognition results;
[0017] Generate a deep learning model based on the training set.
[0018] According to an implementable manner in the embodiments of the present application, the steps of collecting the second inkjet coding sample and the corresponding second annotation file as the test set, testing the deep learning model based on the test set, and outputting the second inkjet coding samples that fail the test, and training the deep learning model based on the second inkjet coding samples that fail the test and the corresponding second annotation files to generate an optimized deep learning model include:
[0019] Collect a preset number of second inkjet coding samples and corresponding second annotation files as a test set, test the deep learning model based on the test set, output the second inkjet coding samples with test failures, and generate a test success rate;
[0020] Train the deep learning model based on the second inkjet coding samples with test failures and the corresponding second annotation files, and generate a test success rate;
[0021] When the test success rate is 100%, stop the test and generate an optimized deep learning model.
[0022] According to an implementable manner in the embodiments of the present application, the collecting the first inkjet coding samples and the corresponding first annotation files as a training set includes:
[0023] Label the first inkjet coding samples with qualified inkjet coding with feature information conforming to the qualified standard and generate a first annotation file, label the first inkjet coding samples with unqualified inkjet coding with feature information conforming to the unqualified standard and generate a first annotation file, and use the first inkjet coding samples with qualified inkjet coding and the first inkjet coding samples with unqualified inkjet coding as the training set.
[0024] According to an implementable manner in the embodiments of the present application, the generating a deep learning model based on the training set includes:
[0025] Perform batch training and optimization on the deep learning model through a training strategy of adjusting the learning rate and setting the batch size.
[0026] In a second aspect, an inkjet coding recognition device based on artificial intelligence is provided. The device includes:
[0027] A creation module: used to create a detection template and set a detection area, and the detection area corresponds to the inkjet coding position;
[0028] A training module: used to collect the first inkjet coding samples and the corresponding first annotation files as a training set, and generate a deep learning model based on the training set, where the first annotation file includes feature information and recognition results corresponding to the first inkjet coding samples;
[0029] A testing module: used to collect the second inkjet coding samples and the corresponding second annotation files as a test set, test the deep learning model based on the test set and output the second inkjet coding samples with test failures, and train the deep learning model based on the second inkjet coding samples with test failures and the corresponding second annotation files to generate an optimized deep learning model;
[0030] An identification module: used to input the inkjet coding to be detected into the optimized deep learning model, and output the recognition result of the inkjet coding to be detected through the optimized deep learning model.
[0031] In a third aspect, a computer device is provided, including:
[0032] at least one processor; and
[0033] a memory communicatively connected to the at least one processor; wherein,
[0034] the memory stores computer instructions executable by the at least one processor, and the computer instructions are executed by the at least one processor to enable the at least one processor to execute the method involved in the above first aspect.
[0035] In a fourth aspect, a computer-readable storage medium is provided, on which computer instructions are stored, and characterized in that the computer instructions are used to cause a computer to execute the method involved in the above first aspect.
[0036] According to the technical content provided by the embodiments of the present application, by creating a detection template and setting a detection area corresponding to the inkjet printing position, accurate identification of the inkjet printing position is achieved. Further, a deep learning model is used to identify the inkjet printing. By collecting and annotating a large number of inkjet printing samples as a training set, the model can learn the complex features of the inkjet printing. At the same time, the deep learning model is tested with a test set, and the model is optimized according to the test feedback, further improving the accuracy and stability of the identification. The inkjet printing to be detected is input into the optimized deep learning model, and the identification result of the inkjet printing to be detected is output through the optimized deep learning model, which can quickly and accurately identify the inkjet printing to be detected, improving the automation and intelligence of inkjet printing identification. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 is a schematic flowchart of a method for inkjet printing identification based on artificial intelligence in an embodiment;
[0038] Figure 2 is a structural block diagram of a device for inkjet printing identification based on artificial intelligence in an embodiment;
[0039] Figure 3 is a schematic structural diagram of a computer device in an embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0040] The present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining the present application and are not used to limit the present application.
[0041] Figure 1 is a flowchart of a method for inkjet printing identification based on artificial intelligence provided by an embodiment of the present application. As Figure 1 shown, the method may include the following steps:
[0042] Step 101: Create a detection template and set the detection area, where the detection area corresponds to the inkjet coding position.
[0043] Specifically, in the vision detection software, create and save a template by creating a new template area, setting the detection area, and selecting the inkjet coding recognition frame as the detection area.
[0044] Step 102: Collect the first inkjet coding sample and the corresponding first annotation file as the training set, and generate a deep learning model based on the training set. The first annotation file includes the feature information and recognition result corresponding to the first inkjet coding sample.
[0045] Specifically, select a test production line and perform deep learning training by identifying a large number of inkjet codings on target products. After deep learning of a large number of inkjet codings on target products, save the images and annotate the feature information. For example, during the detection process, store the images of the products identified as qualified and annotate the feature information, collect the image sets of the products that do not meet the standards, perform character training, and perform data preprocessing such as image enhancement and normalization to further obtain the first inkjet coding sample and the corresponding first annotation file. Further, collect the first inkjet coding sample and the corresponding first annotation file as the training set, and generate a deep learning model based on the training set. The first annotation file includes the feature information and recognition result corresponding to the first inkjet coding sample.
[0046] Step 103: Collect the second inkjet coding sample and the corresponding second annotation file as the test set, test the deep learning model based on the test set and output the second inkjet coding sample with test failure, and train the deep learning model based on the second inkjet coding sample with test failure and the corresponding second annotation file to generate an optimized deep learning model.
[0047] Specifically, after inputting the inkjet coding feature values or characters of qualified and unqualified products into the neural network model for deep learning training in Step 102, package and copy the trained deep learning model to the industrial control computer production software for running tests. Collect the second inkjet coding sample and the corresponding second annotation file as the test set, and test the deep learning model based on the test set. After the test, output the second inkjet coding sample with test failure, and train the deep learning model based on the second inkjet coding sample with test failure and the corresponding second annotation file to generate an optimized deep learning model. For example, through a large number of product identification and detection tests, if there are still character recognition errors or other unqualified product inkjet codings, collect and package such feature values or error image sets for deep learning and character training until the product inkjet coding test recognition rate reaches 100%, and obtain an optimized deep learning model with a test recognition rate of 100%.
[0048] Step 104: Input the inkjet coding to be detected into the optimized deep learning model, and output the recognition result of the inkjet coding to be detected through the optimized deep learning model.
[0049] Specifically, the inkjet code to be detected is input into the optimized deep learning model, and the recognition result of the inkjet code to be detected is output through the optimized deep learning model. For example, the recognition is passed or not passed. In the later stage of operation, relevant data during the recognition process can also be queried by inputting a query statement through an SQL interactive query tool.
[0050] It can be seen that in the embodiment of the present application, by creating a detection template and setting a detection area corresponding to the inkjet code position, accurate recognition of the inkjet code position is achieved. Further, a deep learning model is used to recognize the inkjet code. By collecting and annotating a large number of inkjet code samples as a training set, the model can learn the complex features of the inkjet code. At the same time, the deep learning model is tested through a test set, and the model is optimized according to the test feedback, further improving the accuracy and stability of the recognition. The inkjet code to be detected is input into the optimized deep learning model, and the recognition result of the inkjet code to be detected is output through the optimized deep learning model, which can quickly and accurately recognize the inkjet code to be detected, improving the automation and intelligence of inkjet code recognition.
[0051] In an embodiment of the present application, the method further includes: placing the inkjet code of the target product at the facing position of an industrial camera, and setting a ring light source to uniformly irradiate the inkjet code of the target product; obtaining the best recognition effect by adjusting the focal length of the camera and the color and brightness of the ring light source, and collecting the inkjet code based on the best recognition effect.
[0052] Specifically, the target product is placed at the facing position of the industrial camera and the ring light source. By adjusting the focal length of the camera and the image effect, and debugging the color and brightness of the ring light source according to the chromaticity of the product surface, the best recognition effect can be obtained.
[0053] In an embodiment of the present application, the target product includes a wine bottle. Specifically, when debugging the image effect, the wine bottle is placed at the facing position of the industrial camera and the ring light source. By debugging the focal length of the camera and the image effect, and then adjusting the color and brightness of the ring light source for supplementary lighting according to the influence of the chromaticity of the wine bottle surface, the industrial camera can achieve the best recognition effect for the product surface.
[0054] In an embodiment of the present application, a first inkjet code sample and a corresponding first annotation file are collected as a training set, and a deep learning model is generated based on the training set. The first annotation file includes feature information and a recognition result corresponding to the first inkjet code sample, including: collecting the first inkjet code sample and the corresponding first annotation file as a training set, the first annotation file includes feature information and a recognition result, where the recognition result includes that the inkjet code is qualified or unqualified, and the unqualified inkjet code includes one or more of inkjet skew, inkjet mutilation, inkjet position exceeding the preset range, inkjet repetition, and inkjet non-existence, and the feature information represents the feature on the inkjet code corresponding to the recognition result; generating a deep learning model based on the training set.
[0055] Specifically, first, by collecting the first inkjet coding samples and the corresponding first annotation files, rich learning materials are provided for the deep learning model. These annotation files not only contain the feature information of the inkjet coding, but also clearly indicate the recognition results of the inkjet coding, that is, whether the inkjet coding is qualified. The unqualified situations are refined into various cases such as skewed, incomplete, position deviation, repeated or non-existent inkjet coding, ensuring that the model can learn various forms of inkjet coding. Then, the deep learning model is trained with the training set, and this model can accurately recognize the inkjet coding and determine whether it is qualified.
[0056] In the embodiments of the present application, in order to ensure that the deep learning model can accurately learn various features of the inkjet coding, the samples of qualified and unqualified inkjet coding are carefully annotated. For the qualified inkjet coding samples, the feature information that meets the qualified standards is annotated; for the unqualified inkjet coding samples, corresponding annotations are made according to the specific unqualified situations. This process not only improves the quality of the training set, but also provides support for the subsequent training of the model.
[0057] In an embodiment of the present application, the second inkjet coding samples and the corresponding second annotation files are collected as the test set, the deep learning model is tested based on the test set and the second inkjet coding samples with test failures are output, and the deep learning model is trained based on the second inkjet coding samples with test failures and the corresponding second annotation files to generate an optimized deep learning model, including: collecting a preset number of second inkjet coding samples and the corresponding second annotation files as the test set, testing the deep learning model based on the test set, outputting the second inkjet coding samples with test failures and generating a test success rate; training the deep learning model based on the second inkjet coding samples with test failures and the corresponding second annotation files, and generating a test success rate; stopping the test and generating an optimized deep learning model when the test success rate is 100%.
[0058] Specifically, the second inkjet coding samples and the corresponding second annotation files are collected as the test set, and the deep learning model is tested based on the test set. After the test, the second inkjet coding samples with test failures are output, and the deep learning model is trained based on the second inkjet coding samples with test failures and the corresponding second annotation files to generate an optimized deep learning model. Through the recognition and detection tests of a large number of products, if there are still character recognition errors or other unqualified product inkjet codings, such characteristic values or error atlases are collected and packaged for deep learning and character training until an optimized deep learning model with a test recognition rate of 100% for the products is obtained. For example, after the deep learning training and character learning training of the product inkjet coding characteristic values or characters, the learning training package is copied and run for testing in the industrial control computer production software, and then through the recognition and detection tests of a large number of products, if there are still character recognition errors or other unqualified product inkjet codings, such characteristic values or error atlases are collected and packaged for deep learning and character training until the test recognition rate of the product inkjet coding reaches 100%.
[0059] Embodiments of the present application construct a test set by collecting second inkjet coding samples and corresponding second annotation files, conduct actual tests on the deep learning model, and optimize the model based on the results of failed tests. This process ensures that the model can continuously learn and improve until a 100% test recognition rate is achieved. The beneficial effect of this embodiment is that through iterative testing and training, the accuracy and stability of the deep learning model are effectively improved, and the risk of incorrect product inkjet coding recognition is reduced.
[0060] In an embodiment of the present application, collecting the first inkjet coding samples and corresponding first annotation files as the training set includes: annotating the feature information that meets the qualified standard for the first inkjet coding samples with qualified inkjet coding to generate the first annotation file, annotating the feature information that meets the unqualified standard for the first inkjet coding samples with unqualified inkjet coding to generate the first annotation file, and using the first inkjet coding samples with qualified inkjet coding and the first inkjet coding samples with unqualified inkjet coding as the training set.
[0061] Specifically, in order to ensure that the deep learning model can accurately learn various features of the inkjet coding, the samples with qualified and unqualified inkjet coding are carefully annotated. For the qualified inkjet coding samples, the feature information that meets the qualified standard is annotated; for the unqualified inkjet coding samples, corresponding annotations are made according to the specific unqualified situations. The first inkjet coding samples with qualified inkjet coding and the first inkjet coding samples with unqualified inkjet coding are used as the training set.
[0062] For example, collect non-standard atlases for deep learning and character training. After establishing a model, collecting a large number of images, and performing deep learning training, continuously collect non-standard products (such as products with unprinted inkjet coding, missing one or more inkjet coding characters, inkjet coding exceeding the area range, having one or more extra inkjet coding characters, and the same inkjet coding content for two adjacent sets of products) during the test process for deep learning training of the packaged atlas and corresponding character training. During character training, character training is required when detecting characters for the first time or when there are regular character misidentifications. Select the character to be added, click character training, click add after entering the interface, enter the correct character, click train, select the character, then test and observe the algorithm recognition result and save it.
[0063] In an embodiment of the present application, generating a deep learning model based on the training set includes: batch training and optimizing the deep learning model through a training strategy of adjusting the learning rate and setting the batch size.
[0064] Specifically, the learning rate is a hyperparameter in deep learning that controls the step size of parameter updates. It determines the magnitude of parameter updates based on the gradient of the loss function in each iteration. The choice of the learning rate is directly related to the performance of the model and the effectiveness of the training process. By adjusting the appropriate learning rate, the training speed can be optimized and the training stability can be improved. At the same time, the batch training method of the model divides the entire dataset into several batches, and the number of samples in each batch is called the batch size. Selecting an appropriate batch size can improve the stability of neural network training.
[0065] It can be seen that the embodiments of the present application realize high-precision identification and detection of product inkjet coding through steps such as image preprocessing, modeling, deep learning training, character training, model optimization, identification and detection, associated data query, elimination of unqualified products, repeated learning training, and system operation and maintenance, improving the identification efficiency and product quality, reducing the labor cost, and having significant technical effects.
[0066] Figure 2 The following is a schematic structural diagram of an inkjet coding recognition device based on artificial intelligence provided by an embodiment of the present application, as Figure 2 shown. The device may include:
[0067] A creation module 201: configured to create a detection template and set a detection area, where the detection area corresponds to the inkjet coding position;
[0068] A training module 202: configured to collect a first inkjet coding sample and a corresponding first annotation file as a training set, and generate a deep learning model based on the training set, where the first annotation file includes feature information and recognition results corresponding to the first inkjet coding sample;
[0069] A testing module 203: configured to collect a second inkjet coding sample and a corresponding second annotation file as a testing set, test the deep learning model based on the testing set and output the second inkjet coding samples that fail the test, and train the deep learning model based on the second inkjet coding samples that fail the test and the corresponding second annotation files to generate an optimized deep learning model;
[0070] An identification module 204: configured to input the inkjet coding to be detected into the optimized deep learning model, and output the recognition result of the inkjet coding to be detected through the optimized deep learning model.
[0071] In an embodiment of the present application, it further includes a camera module 205: configured to place the inkjet coding of the target product at the directly facing position of an industrial camera, and set a ring light source to uniformly irradiate the inkjet coding of the target product; obtain the best recognition effect by adjusting the focal length of the camera and the color and brightness of the ring light source, and collect the inkjet coding based on the best recognition effect.
[0072] In one embodiment of the present application, the target product includes a wine bottle.
[0073] In one embodiment of the present application, the first inkjet coding sample and the corresponding first annotation file are collected as a training set, and a deep learning model is generated based on the training set, where the first annotation file includes feature information and recognition results corresponding to the first inkjet coding sample, including:
[0074] The first inkjet coding sample and the corresponding first annotation file are collected as a training set. The first annotation file includes feature information and recognition results. The recognition results include qualified inkjet coding or unqualified inkjet coding. The unqualified inkjet coding includes one or more of inkjet coding skew, inkjet coding mutilation, inkjet coding position exceeding a preset range, inkjet coding repetition, and inkjet coding non-existence. The feature information represents the features on the inkjet coding corresponding to the recognition results.
[0075] A deep learning model is generated based on the training set.
[0076] In one embodiment of the present application, the second inkjet coding sample and the corresponding second annotation file are collected as a test set, the deep learning model is tested based on the test set, and the second inkjet coding sample with a test failure is output. The deep learning model is trained based on the second inkjet coding sample with a test failure and the corresponding second annotation file to generate an optimized deep learning model, including:
[0077] A preset number of second inkjet coding samples and the corresponding second annotation files are collected as a test set. The deep learning model is tested based on the test set, the second inkjet coding sample with a test failure is output, and a test success rate is generated.
[0078] The deep learning model is trained based on the second inkjet coding sample with a test failure and the corresponding second annotation file, and a test success rate is generated.
[0079] When the test success rate is 100%, the test is stopped and an optimized deep learning model is generated.
[0080] In one embodiment of the present application, the collection of the first inkjet coding sample and the corresponding first annotation file as a training set includes:
[0081] The feature information conforming to the qualified standard is marked for the first inkjet coding sample with qualified inkjet coding to generate a first annotation file, the feature information conforming to the unqualified standard is marked for the first inkjet coding sample with unqualified inkjet coding to generate a first annotation file, and the first inkjet coding sample with qualified inkjet coding and the first inkjet coding sample with unqualified inkjet coding are used as the training set.
[0082] In one embodiment of the present application, generating the deep learning model based on the training set includes: batch training and optimizing the deep learning model by adjusting the learning rate and setting the training strategy of the batch size.
[0083] According to the specific embodiments provided by the present application, the technical solutions provided by the present application may have the following advantages:
[0084] Through steps such as image preprocessing, modeling, deep learning training, character training, model optimization, recognition detection, associated data query, elimination of unqualified products, repeated learning training, and system operation and maintenance, high-precision recognition detection of product inkjet coding is achieved, the recognition efficiency and product quality are improved, the labor cost is reduced, and remarkable technical effects are obtained.
[0085] It can be understood that implementing any method or product of the present application does not necessarily need to achieve all the above-mentioned advantages at the same time.
[0086] For the same and similar parts among the above-mentioned various embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment.
[0087] It should be noted that the use of user data may be involved in the embodiments of the present application. In actual applications, user-specific personal data can be used in the solutions described herein within the scope permitted by applicable laws and regulations (for example, when the user clearly consents, notifies the user effectively, and the user clearly authorizes, etc.) in compliance with the requirements of applicable laws and regulations in the country where it is located.
[0088] According to the embodiments of the present application, the present application also provides a computer device and a computer-readable storage medium.
[0089] As Figure 3 shown, it is a block diagram of a computer device according to an embodiment of the present application. The computer device is intended to represent various forms of digital computers or mobile devices. Among them, the digital computer may include a desktop computer, a portable computer, a workbench, a personal digital assistant, a server, a mainframe computer, and other suitable computers. The mobile device may include a tablet computer, a smart phone, a wearable device, etc.
[0090] As Figure 3 shown, the device 300 includes a computing unit 301, a ROM 302, a RAM 303, a bus 304, and an input / output (I / O) interface 305. The computing unit 301, the ROM 302, and the RAM 303 are connected to each other through the bus 304. The input / output (I / O) interface 305 is also connected to the bus 304.
[0091] The computing unit 301 can execute various processes in the method embodiments of the present application according to computer instructions stored in the read-only memory (ROM) 302 or computer instructions loaded from the storage unit 308 into the random access memory (RAM) 303. The computing unit 301 can be various general and / or special processing components with processing and computing capabilities. The computing unit 301 can include, but is not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. In some embodiments, the method provided by the embodiments of the present application can be implemented as a computer software program, which is tangibly contained in a computer-readable storage medium, such as the storage unit 308.
[0092] The RAM 303 can also store various programs and data required for the operation of the device 300. Part or all of the computer programs can be loaded and / or installed onto the device 300 via the ROM 802 and / or the communication unit 309.
[0093] The input unit 306, output unit 307, storage unit 308, and communication unit 309 in the device 300 can be connected to the I / O interface 305. Among them, the input unit 306 can be, such as, a keyboard, a mouse, a touch screen, a microphone, etc.; the output unit 307 can be, such as, a display, a speaker, an indicator light, etc. The device 300 can exchange information, data, etc. with other devices through the communication unit 309.
[0094] It should be noted that the device may also include other components necessary for normal operation. It may also only include the components necessary to implement the solution of the present application, and does not necessarily include all the components shown in the figure.
[0095] Various embodiments of the systems and technologies described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof.
[0096] The computer instructions for implementing the method of the present application can be written in any combination of one or more programming languages. These computer instructions can be provided to the computing unit 301, such that when the computer instructions are executed by a computing unit 301 such as a processor, the steps involved in the method embodiments of the present application are executed.
[0097] The computer-readable storage medium provided by the present application may be a tangible medium that can contain or store computer instructions for executing the various steps involved in the method embodiments of the present application. The computer-readable storage medium may include, but is not limited to, storage media in the forms of electronic, magnetic, optical, electromagnetic, etc.
[0098] The above specific embodiments do not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principle of the present application shall be included within the protection scope of the present application.
Claims
1. A coding recognition method based on artificial intelligence, characterized in that: The method includes: Create a detection template and set the detection area, the detection area corresponds to the coding position; Collecting a first coding sample and a corresponding first annotation file as a training set, and generating a deep learning model based on the training set, wherein the first annotation file includes feature information and a recognition result corresponding to the first coding sample; Collecting a second coding sample and a corresponding second annotation file as a test set, testing the deep learning model based on the test set and outputting a second coding sample that fails the test, and training the deep learning model based on the second coding sample that fails the test and the corresponding second annotation file to generate an optimized deep learning model; The inkjet code to be detected is input into the optimized deep learning model, and the recognition result of the inkjet code to be detected is output through the optimized deep learning model.
2. The method for inkjet coding recognition based on artificial intelligence according to claim 1, characterized in that: The method further comprises: Place the target product's inkjet printer directly opposite the industrial camera, and set a ring light source to evenly illuminate the target product's inkjet printer; The best recognition effect is obtained by adjusting the focal length of the camera and the color and brightness of the annular light source, and the inkjet code is collected based on the best recognition effect.
3. The method for inkjet coding recognition based on artificial intelligence according to claim 2 is characterized in that: The target products include wine bottles.
4. The method for inkjet coding recognition based on artificial intelligence according to claim 1, characterized in that: The collecting of the first coding sample and the corresponding first annotation file as a training set, and generating a deep learning model based on the training set, wherein the first annotation file includes feature information and recognition results corresponding to the first coding sample, including: A first coding sample and a corresponding first annotation file are collected as a training set, wherein the first annotation file includes feature information and a recognition result, wherein the recognition result includes a qualified coding or an unqualified coding, wherein the unqualified coding includes one or more of a skewed coding, a defective coding, a coding position exceeding a preset range, a repeated coding, and a non-existent coding, and the feature information indicates a feature on the coding corresponding to the recognition result; A deep learning model is generated based on the training set.
5. The method for inkjet coding recognition based on artificial intelligence according to claim 1, characterized in that: The collecting of the second coding sample and the corresponding second annotation file as a test set, testing the deep learning model based on the test set and outputting the second coding sample that failed the test, and training the deep learning model based on the second coding sample that failed the test and the corresponding second annotation file to generate an optimized deep learning model include: Collect a preset number of second coding samples and corresponding second annotation files as a test set, test the deep learning model based on the test set, output the second coding samples that failed the test and generate a test success rate; Training the deep learning model based on the second coding sample that failed the test and the corresponding second annotation file, and generating a test success rate; When the test success rate reaches 100%, the test is stopped and an optimized deep learning model is generated.
6. The method for inkjet coding recognition based on artificial intelligence according to claim 4 is characterized in that: The collecting of the first coding sample and the corresponding first annotation file as a training set includes: The first coding sample that passes the coding is labeled with characteristic information that meets the qualified standard and generates a first labeling file. The first coding sample that fails the coding is labeled with characteristic information that meets the unqualified standard and generates a first labeling file. The first coding sample that passes the coding and the first coding sample that fails the coding are used as training sets.
7. The method for inkjet coding recognition based on artificial intelligence according to claim 4, characterized in that: Generating a deep learning model based on the training set includes: Batch training and optimization of deep learning models are performed by adjusting the learning rate and setting the batch size training strategy.
8. A coding recognition device based on artificial intelligence, characterized in that: The device includes: Creation module: used to create a detection template and set the detection area, the detection area corresponds to the coding position; Training module: used for collecting a first coding sample and a corresponding first annotation file as a training set, and generating a deep learning model based on the training set, wherein the first annotation file includes feature information and recognition results corresponding to the first coding sample; Testing module: used for collecting the second coding sample and the corresponding second annotation file as a test set, testing the deep learning model based on the test set and outputting the second coding sample that failed the test, training the deep learning model based on the second coding sample that failed the test and the corresponding second annotation file to generate an optimized deep learning model; Recognition module: used to input the inkjet code to be detected into the optimized deep learning model, and output the recognition result of the inkjet code to be detected through the optimized deep learning model.
9. A computer device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores computer instructions that can be executed by the at least one processor, and the computer instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 7.
10. A computer-readable storage medium having computer instructions stored thereon, characterized in that: The computer instructions are used to make a computer execute the method according to any one of claims 1 to 7.