A slab number identification method, system, medium and electronic terminal

Abstract

The present invention provides a slab number recognition method, system, medium and electronic terminal, the method comprising: continuously collecting a plurality of slab side images on a slab transport roller; inputting the slab side images into a preset first recognition model for feature extraction and classification, and obtaining images of slabs to be spliced, wherein the images of slabs to be spliced ​​are slab side images with letter and / or number features; inputting the slab images to be spliced ​​into a preset image stitching model for image stitching, and obtaining a stitched image; inputting the stitched image into a preset second recognition model for recognition, and obtaining a slab number recognition value; by comparing the slab number recognition value with a preset slab number expected value, issuing a warning, and completing slab number recognition; the slab number recognition method, system, medium and electronic terminal of the present invention better realize the accurate recognition of slab numbers, have high recognition accuracy, high recognition efficiency, low cost, and are convenient to implement.

Description

Technical Field

[0001] The present invention relates to the field of image recognition technology, and in particular to a slab number recognition method, system, medium and electronic terminal. Background Art

[0002] With the development of the steel industry, people have higher and higher requirements for the degree of automation in the steelmaking process. In order to achieve refined production, it is usually necessary to identify and count the slab numbers during the production process. However, at present, the slab numbers are usually identified manually, which easily leads to low slab number recognition accuracy, low recognition efficiency, and high cost. Summary of the invention

[0003] The present invention provides a slab number recognition method, system, medium and electronic terminal to solve the problems of low slab number recognition accuracy, low recognition efficiency and high cost in the prior art.

[0004] The slab number recognition method provided by the present invention comprises:

[0005] Continuously collect multiple side images of the slab on the slab transport roller;

[0006] Input the slab side image into a preset first recognition model for feature extraction and classification to obtain an image of the slab to be spliced, wherein the image of the slab to be spliced ​​is a slab side image having letter and / or number features;

[0007] Inputting the image of the slab to be spliced ​​into a preset image splicing model to perform image splicing to obtain a spliced ​​image;

[0008] Input the spliced ​​image into a preset second recognition model for recognition, and obtain a slab number recognition value;

[0009] By comparing the slab number identification value with a preset slab number expected value, a warning is issued to complete the slab number identification.

[0010] Optionally, the step of continuously acquiring a plurality of side images of the slab on the slab transport roller table includes:

[0011] Using a pre-set slab in-place detection device to obtain a slab in-place signal;

[0012] According to the slab arrival signal, the roller table running speed is obtained;

[0013] According to the roller running speed and the preset shooting rules, the target shooting interval is obtained;

[0014] According to the target shooting interval, a plurality of side images of the slab on both sides of the slab transport roller are regularly collected.

[0015] Optionally, the step of acquiring the first recognition model includes:

[0016] Acquire a first training set, wherein the first training set includes: a plurality of training samples, and true category information corresponding to the training samples;

[0017] Inputting the training sample into a preset first convolutional neural network for feature extraction to obtain first sample features;

[0018] Obtaining a predicted image category by classifying the first sample features;

[0019] According to the predicted image category and the actual category information, the first convolutional neural network is iteratively trained to obtain the first recognition model.

[0020] Optionally, the step of acquiring the image stitching model includes:

[0021] Acquire a second training set, wherein the second training set includes: a plurality of image samples to be stitched, and real stitched images corresponding to the image samples to be stitched;

[0022] Inputting the plurality of image samples to be stitched into a preset deep neural network for feature extraction, sliding matching and image stitching to obtain a predicted stitched image;

[0023] The deep neural network is trained according to the real stitching image and the predicted stitching image to obtain the image stitching model.

[0024] Optionally, the step of inputting a plurality of the to-be-stitched image samples into a preset deep neural network for feature extraction, sliding matching and image stitching to obtain a predicted stitched image includes:

[0025] Extracting and encoding features of a plurality of continuous image samples to be spliced, and obtaining feature values ​​of a plurality of continuous image samples to be spliced;

[0026] In units of columns, sliding matching is performed on the feature values ​​of two adjacent image samples to be spliced ​​to obtain a matching result;

[0027] Image stitching is performed according to the matching result to obtain the predicted stitched image.

[0028] Optionally, the step of acquiring the second recognition model includes:

[0029] Acquire a third training set, the third training set comprising: a plurality of samples to be identified, and real labeling information corresponding to the samples to be identified;

[0030] Inputting the sample to be identified into a preset second convolutional neural network for feature extraction to obtain second sample features;

[0031] Obtaining a prediction result by performing feature matching on the second sample feature;

[0032] Training the second convolutional neural network once according to the prediction result and the sample content in the real annotation information;

[0033] Comparing the second sample feature with a preset feature to be compared to obtain a predicted comparison result;

[0034] According to the predicted comparison result and the actual comparison result in the actual annotation information, the second convolutional neural network that has been trained once is trained for the second time to obtain the second recognition model.

[0035] Optionally, the step of issuing a warning by comparing the slab number identification value with a preset slab number expected value includes:

[0036] If the slab number identification value is different from the slab number expected value, a warning is issued, and the corresponding stitching image is recorded as an error image;

[0037] By manually verifying the error image, determining whether the slab code in the error image is correct;

[0038] If the slab code in the error image is correct, the error image is marked as a correct image, and the number of markings is recorded;

[0039] When the number of markings exceeds a preset number threshold, the first recognition model, the image stitching model and the second recognition model are retrained and optimized to complete the slab number recognition.

[0040] The present invention also provides a slab number recognition system, comprising:

[0041] An image acquisition module, used for continuously acquiring multiple side images of the slab on the slab transport roller;

[0042] A first recognition module is used to input the slab side image into a preset first recognition model for feature extraction and classification, and obtain the image of the slab to be spliced, wherein the image of the slab to be spliced ​​is a slab side image with letter and / or number features;

[0043] An image stitching module, used for inputting the image of the slab to be stitched into a preset image stitching model to perform image stitching and obtain a stitched image;

[0044] A second recognition module is used to input the spliced ​​image into a preset second recognition model for recognition, and obtain a slab number recognition value;

[0045] The comparison module is used to issue a warning and complete the slab number recognition by comparing the slab number recognition value with the preset slab number expected value; the image acquisition module, the first recognition module, the image stitching module and the second recognition module are connected.

[0046] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements any of the above methods when executed by a processor.

[0047] The present invention also provides an electronic terminal, comprising: a processor and a memory;

[0048] The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the terminal executes any one of the methods described above.

[0049] Beneficial effects of the present invention: The slab number recognition method, system, medium and electronic terminal of the present invention continuously collect multiple side images of the slabs on the slab transport roller; input the side images of the slabs into a preset first recognition model for feature extraction and classification, and obtain the image of the slab to be spliced, wherein the image of the slab to be spliced ​​is a side image of the slab with letter and / or number features; input the image of the slab to be spliced ​​into a preset image stitching model for image stitching to obtain a stitched image; input the stitched image into a preset second recognition model for recognition to obtain a slab number recognition value; by comparing the slab number recognition value with the preset slab number expected value, a warning is issued, and the slab number recognition is completed, so as to better realize the accurate recognition of the slab number, with high recognition accuracy, high recognition efficiency, low cost, convenient implementation and high degree of automation. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a schematic flow chart of a slab number recognition method in an embodiment of the present invention.

[0051] Figure 2 It is a schematic diagram of the process of continuously collecting multiple side images of slabs on a slab transport roller in a slab number recognition method in an embodiment of the present invention.

[0052] Figure 3 It is a schematic diagram of the process of obtaining the first recognition model in the slab number recognition method in an embodiment of the present invention.

[0053] Figure 4 It is a schematic diagram of the process of obtaining an image stitching model in the slab number recognition method in an embodiment of the present invention.

[0054] Figure 5 It is a schematic diagram of the process of obtaining the second recognition model in the slab number recognition method in an embodiment of the present invention.

[0055] Figure 6 It is a flow chart of completing slab number identification by comparing the slab number identification value with a preset slab number expected value in the slab number identification method in an embodiment of the present invention.

[0056] Figure 7 It is a structural schematic diagram of a slab number recognition system in an embodiment of the present invention.

[0057] Figure 8 It is another structural schematic diagram of the slab number recognition system in an embodiment of the present invention.

[0058] Figure ID:

[0059] 01 slab; 02 slab transport roller; 03 laser transmitter; 04 reflector DETAILED DESCRIPTION

[0060] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0061] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and thus the drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0062] The inventors have found that with the development of the steel industry, people have higher and higher requirements for the degree of automation in the steelmaking process. In order to achieve refined production, it is usually necessary to identify and count the slab numbers during the production process. However, at present, the slab numbers are usually identified manually, which easily leads to low slab number recognition accuracy, low recognition efficiency, and high cost. Therefore, the inventors propose a slab number recognition method, system, medium and electronic terminal, which continuously collect multiple side images of slabs on a slab transport roller; input the side images of the slabs into a preset first recognition model for feature extraction and classification, and obtain the image of the slab to be spliced, wherein the image of the slab to be spliced ​​is a side image of the slab with letter and / or number features; input the image of the slab to be spliced ​​into a preset image stitching model for image stitching, and obtain a stitched image; input the stitched image into a preset second recognition model for recognition, and obtain a slab number recognition value; by comparing the slab number recognition value with the preset slab number expected value, a warning is issued, and the slab number recognition is completed, so as to better realize the accurate recognition of the slab number, with high recognition accuracy, high recognition efficiency, low cost, convenient implementation, high degree of automation, and effective prevention of mixed slabs.

[0063] like Figure 1 As shown, the slab number recognition method in this embodiment includes:

[0064] S101: Continuously capture multiple side images of the slab on the slab transport roller. The side images of the slab refer to images of both sides of the slab perpendicular to the running direction of the slab transport roller. By continuously capturing multiple side images of the slab on the slab transport roller, it is convenient to subsequently identify and splice multiple continuous side images of the slab, avoiding incomplete or incorrect slab number recognition caused by capturing only a single slab image. In some embodiments, the capture of the side images of the slab can be completed by setting cameras on both sides of the slab transport roller.

[0065] S102: Input the slab side image into a preset first recognition model for feature extraction and classification, and obtain the slab image to be spliced, wherein the slab image to be spliced ​​is a slab side image with letter and / or number features. By inputting the slab side image into a pre-trained first recognition model for feature extraction and classification, the slab side image with letter features and / or number features can be well screened out, reducing the workload of subsequent slab number recognition, and helping to improve the accuracy of subsequent image splicing.

[0066] S103: Input the slab images to be spliced ​​into a preset image splicing model for image splicing to obtain a spliced ​​image. By inputting the slab images to be spliced ​​into a pre-trained image splicing model for image splicing, a more complete spliced ​​image can be obtained to avoid incomplete slab number recognition caused by a single slab image.

[0067] S104: Input the spliced ​​image into a preset second recognition model for recognition, and obtain a slab number recognition value. By inputting the spliced ​​image into a pre-trained second recognition model for recognition, a relatively accurate slab number recognition value can be obtained.

[0068] S105: By comparing the slab number identification value with the preset slab number expected value, a warning is issued to complete the slab number identification. That is, when the slab number identification is different from the preset slab number expected value, a warning is issued to avoid mixed slabs. The slab number expected value can be obtained through the upper-level server, for example, the slab transportation plan in the upper-level server is obtained, so as to obtain the corresponding slab expected value, and the expected value is the planned value. By comparing the slab number identification value with the preset slab number expected value, the real-time identification and monitoring of the slab number can be better achieved, with a high degree of automation and high identification accuracy. In some embodiments, the first recognition model, the image splicing model and the second recognition model can be set together in the same model, or they can be set separately.

[0069] like Figure 2 As shown, in order to improve the accuracy of collecting the side image of the slab and improve the efficiency of collecting the side image of the slab, the inventor proposes that the steps of continuously collecting multiple side images of the slab on the slab transport roller include:

[0070] S201: Using a pre-set slab in-place detection device, a slab in-place signal is obtained. The slab in-place detection device may be a photoelectric detection element, etc. By collecting the slab in-place signal, it is convenient to collect the slab side image more accurately later.

[0071] S202: According to the slab arrival signal, the roller running speed is obtained. For example, a speed sensor is set on the slab transport roller or on the motor of the slab transport roller to obtain the roller running speed. By obtaining the roller running speed, it is convenient to determine the optimal shooting interval later.

[0072] S203: According to the roller running speed and the preset shooting rules, a target shooting interval is obtained. The shooting rules include: multiple roller running speeds and multiple shooting time intervals, and the roller running speeds and the shooting time intervals have a one-to-one mapping relationship. According to the roller running speed obtained in real time and the one-to-one mapping relationship, the shooting time interval corresponding to the roller running speed obtained in real time is determined, and the shooting time interval corresponding to the roller running speed obtained in real time is used as the target shooting interval.

[0073] S204: regularly collecting multiple side images of the slab on both sides of the slab transport roller according to the target shooting interval. That is, according to the target shooting interval, the side images of the slab are collected at fixed time intervals, which better realizes the accurate collection of the side images of the slab, has high accuracy and is easy to implement.

[0074] like Figure 3 As shown, in order to obtain a better first recognition model, the inventor proposes that the step of obtaining the first recognition model includes:

[0075] S301: Acquire a first training set, where the first training set includes: a plurality of training samples, and true category information corresponding to the training samples.

[0076] S302: Input the training sample into a preset first convolutional neural network for feature extraction to obtain first sample features.

[0077] In some embodiments, the first convolutional neural network includes: an input layer, which is used to normalize the input image; a convolution layer, which is composed of multiple convolution kernels, each element of the convolution kernel is composed of a specific weight coefficient and a bias, and the weight coefficient and the bias can be set according to actual conditions, which will not be repeated here; a pooling layer, which uses random pooling; and an output layer, which uses a normalized exponential function to output classification values.

[0078] S303: Obtain a predicted image category by classifying the first sample features.

[0079] S304: Iteratively train the first convolutional neural network according to the predicted image category and the real category information to obtain the first recognition model. That is, iteratively train the first convolutional neural network according to the difference between the predicted image category and the real category information to obtain a better first recognition model.

[0080] like Figure 4 As shown, in some embodiments, the step of acquiring the image stitching model includes:

[0081] S401: Acquire a second training set, where the second training set includes: a plurality of image samples to be stitched, and real stitched images corresponding to the image samples to be stitched.

[0082] S402: Inputting the plurality of image samples to be stitched into a preset deep neural network for feature extraction, sliding matching and image stitching to obtain a predicted stitched image.

[0083] In some embodiments, the step of inputting a plurality of the to-be-stitched image samples into a preset deep neural network for feature extraction, sliding matching and image stitching, and obtaining a predicted stitched image comprises:

[0084] S4021: extracting and encoding features of a plurality of continuous image samples to be spliced, and obtaining feature values ​​of a plurality of continuous image samples to be spliced.

[0085] S4022: Perform sliding matching on the feature values ​​of two adjacent image samples to be spliced ​​in units of columns to obtain a matching result.

[0086] S4023: Perform image stitching based on the matching result to obtain the predicted stitching image, that is, perform image stitching based on the overlapping parts of adjacent images.

[0087] For example: multiple image samples to be spliced ​​are numbered 1, 2, 3, etc., and the numbered image samples to be spliced ​​are arranged from left to right. The right half of the image of the image sample No. 1 to be spliced ​​is selected for encoding, the middle sample image to be spliced ​​is fully encoded, and the left half of the image of the last image is selected for encoding, and the characteristic values ​​of multiple samples to be spliced ​​are obtained to reduce the encoding workload. The characteristic value of the right half of the image No. 1 is matched with the characteristic value of the left half of the middle image. When the similarity of the characteristic value of the first column on the left side of the next image and the first column of the first image exceeds the preset similarity threshold, the sliding matching is started. When the similarity of the characteristic values ​​of the consecutive columns exceeds the similarity threshold, the image splicing is completed. The splicing of the middle image and the next image is completed in the same way. When there are only two image samples to be spliced, only one step of splicing is required. When there is only one image sample to be spliced, no splicing is required.

[0088] S403: According to the real stitched image and the predicted stitched image, the deep neural network is trained to obtain the image stitching model. That is, according to the difference between the real stitched image and the predicted stitched image, the deep neural network is iteratively trained to obtain a better image stitching model with higher accuracy.

[0089] like Figure 5 As shown, in order to improve the recognition accuracy of the second recognition model, the inventor proposes that the step of obtaining the second recognition model includes:

[0090] S501: Acquire a third training set, where the third training set includes: a plurality of samples to be identified, and real labeling information corresponding to the samples to be identified.

[0091] S502: Input the sample to be identified into a preset second convolutional neural network for feature extraction to obtain second sample features.

[0092] S503: Obtain prediction results by performing feature matching on the second sample features.

[0093] S504: Training the second convolutional neural network once according to the prediction result and the sample content in the real annotation information. That is, training the second convolutional neural network once according to the gap between the prediction result and the sample content in the real annotation information.

[0094] S505: Compare the second sample feature with the preset feature to be compared to obtain a predicted comparison result.

[0095] S506: According to the predicted comparison result and the actual comparison result in the actual annotation information, the second convolutional neural network that has been trained once is trained twice to obtain the second recognition model. That is, according to the gap between the predicted comparison result and the actual comparison result in the actual annotation information, the second convolutional neural network that has been trained once is trained twice to obtain a second recognition model with higher accuracy.

[0096] like Figure 6 As shown, in some embodiments, by comparing the slab number identification value with a preset slab number expected value, the step of issuing a warning includes:

[0097] S601: If the slab number identification value is different from the slab number expected value, a warning is issued and the corresponding stitching image is recorded as an error image.

[0098] S602: manually verifying the error image to determine whether the slab code in the error image is correct, that is, manually determining whether the slab number in the error image is the same as the preset slab number expected value, and if they are the same, determining that the slab code in the error image is correct.

[0099] S603: If the slab code in the error image is correct, the error image is marked as a correct image, and the number of markings is recorded.

[0100] S604: When the number of markings exceeds the preset number threshold, the first recognition model, the image stitching model and the second recognition model are retrained and optimized to complete the slab number recognition. Through the above optimization mechanism, model errors can be avoided, and the first recognition model, the image stitching model and the second recognition model can be iteratively updated in real time, effectively improving the accuracy of slab number recognition.

[0101] like Figure 7 As shown, this embodiment also provides a slab number recognition system, including:

[0102] An image acquisition module, used for continuously acquiring multiple side images of the slab on the slab transport roller;

[0103] A first recognition module is used to input the slab side image into a preset first recognition model for feature extraction and classification, and obtain the image of the slab to be spliced, wherein the image of the slab to be spliced ​​is a slab side image with letter and / or number features;

[0104] An image stitching module, used for inputting the image of the slab to be stitched into a preset image stitching model to perform image stitching and obtain a stitched image;

[0105] A second recognition module is used to input the spliced ​​image into a preset second recognition model for recognition, and obtain a slab number recognition value;

[0106] The comparison module is used to issue a warning and complete the slab number recognition by comparing the slab number recognition value with the preset slab number expected value; the image acquisition module, the first recognition module, the image stitching module and the second recognition module are connected. The slab number recognition system in this embodiment continuously collects multiple slab side images on the slab transport roller; inputs the slab side images into the preset first recognition model for feature extraction and classification to obtain the image of the slab to be spliced, and the slab image to be spliced ​​is a slab side image with letter and / or number features; inputs the slab image to be spliced ​​into the preset image stitching model for image stitching to obtain the stitched image; inputs the stitched image into the preset second recognition model for recognition to obtain the slab number recognition value; by comparing the slab number recognition value with the preset slab number expected value, a warning is issued to complete the slab number recognition, and the accurate recognition of the slab number is better realized, the recognition accuracy is high, the recognition efficiency is high, the cost is low, the implementation is convenient, the degree of automation is high, and the feasibility is strong.

[0107] like Figure 8 As shown, in some embodiments, the image acquisition module includes a first image acquisition module and a second image acquisition module, the first image acquisition module and the second image acquisition module are respectively arranged on both sides of the slab conveying roller, and the first image acquisition device and the second image acquisition device are arranged in opposite directions.

[0108] In some embodiments, it also includes: a switch, and the first image acquisition module and the second image acquisition module are respectively connected to the first recognition module through the switch.

[0109] In some embodiments, it further includes: a control module, and the image acquisition module, the first recognition module, the image stitching module, the second recognition module, and the comparison module are respectively connected to the control module.

[0110] In some embodiments, the step of continuously acquiring a plurality of side images of the slab on the slab transport roller comprises:

[0111] A slab in-place signal is obtained by using a pre-set slab in-place detection device; the slab in-place detection device comprises: a laser emitter and a reflector;

[0112] According to the slab arrival signal, the roller table running speed is obtained;

[0113] According to the roller running speed and the preset shooting rules, the target shooting interval is obtained;

[0114] According to the target shooting interval, a plurality of side images of the slab on both sides of the slab transport roller are regularly collected.

[0115] In some embodiments, the step of acquiring the first recognition model includes:

[0116] Acquire a first training set, wherein the first training set includes: a plurality of training samples, and true category information corresponding to the training samples;

[0117] Inputting the training sample into a preset first convolutional neural network for feature extraction to obtain first sample features;

[0118] Obtaining a predicted image category by classifying the first sample features;

[0119] According to the predicted image category and the actual category information, the first convolutional neural network is iteratively trained to obtain the first recognition model.

[0120] In some embodiments, the step of acquiring the image stitching model includes:

[0121] Acquire a second training set, wherein the second training set includes: a plurality of image samples to be stitched, and real stitched images corresponding to the image samples to be stitched;

[0122] Inputting the plurality of image samples to be stitched into a preset deep neural network for feature extraction, sliding matching and image stitching to obtain a predicted stitched image;

[0123] The deep neural network is trained according to the real stitching image and the predicted stitching image to obtain the image stitching model.

[0124] In some embodiments, the step of inputting a plurality of the to-be-stitched image samples into a preset deep neural network for feature extraction, sliding matching and image stitching, and obtaining a predicted stitched image comprises:

[0125] Extracting and encoding features of a plurality of continuous image samples to be spliced, and obtaining feature values ​​of a plurality of continuous image samples to be spliced;

[0126] In units of columns, sliding matching is performed on the feature values ​​of two adjacent image samples to be spliced ​​to obtain a matching result;

[0127] Image stitching is performed according to the matching result to obtain the predicted stitched image.

[0128] In some embodiments, the step of acquiring the second recognition model includes:

[0129] Acquire a third training set, the third training set comprising: a plurality of samples to be identified, and real labeling information corresponding to the samples to be identified;

[0130] Inputting the sample to be identified into a preset second convolutional neural network for feature extraction to obtain second sample features;

[0131] Obtaining a prediction result by performing feature matching on the second sample feature;

[0132] Training the second convolutional neural network once according to the prediction result and the sample content in the real annotation information;

[0133] Comparing the second sample feature with a preset feature to be compared to obtain a predicted comparison result;

[0134] According to the predicted comparison result and the actual comparison result in the actual annotation information, the second convolutional neural network that has been trained once is trained for the second time to obtain the second recognition model.

[0135] In some embodiments, by comparing the slab number identification value with a preset slab number expected value, the step of issuing a warning includes:

[0136] If the slab number identification value is different from the slab number expected value, a warning is issued, and the corresponding stitching image is recorded as an error image;

[0137] By manually verifying the error image, determining whether the slab code in the error image is correct;

[0138] If the slab code in the error image is correct, the error image is marked as a correct image, and the number of markings is recorded;

[0139] When the number of markings exceeds a preset number threshold, the first recognition model, the image stitching model and the second recognition model are retrained and optimized to complete the slab number recognition.

[0140] This embodiment further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, any one of the methods in this embodiment is implemented.

[0141] This embodiment also provides an electronic terminal, including: a processor and a memory;

[0142] The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the terminal executes any one of the methods in this embodiment.

[0143] The computer-readable storage medium in this embodiment can be understood by ordinary technicians in this field: all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to the computer program. The aforementioned computer program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk and other media that can store program codes.

[0144] The electronic terminal provided in this embodiment includes a processor, a memory, a transceiver and a communication interface. The memory and the communication interface are connected to the processor and the transceiver and complete communication with each other. The memory is used to store computer programs, the communication interface is used to communicate, and the processor and the transceiver are used to run computer programs so that the electronic terminal executes each step of the above method.

[0145] In this embodiment, the memory may include a random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0146] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0147] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by a person of ordinary skill in the art without departing from the spirit and technical concept disclosed by the present invention shall still be covered by the claims of the present invention.

Claims

1. A slab number recognition method, characterized in that: include: Continuously collect multiple side images of the slab on the slab transport roller; Input the slab side image into a preset first recognition model for feature extraction and classification to obtain an image of the slab to be spliced, wherein the image of the slab to be spliced ​​is a slab side image having letter and / or number features; Inputting the image of the slab to be spliced ​​into a preset image splicing model to perform image splicing to obtain a spliced ​​image; Input the spliced ​​image into a preset second recognition model for recognition, and obtain a slab number recognition value; By comparing the slab number identification value with a preset slab number expected value, a warning is issued to complete the slab number identification; The step of acquiring the image stitching model includes: Acquire a second training set, wherein the second training set includes: a plurality of image samples to be stitched, and real stitched images corresponding to the image samples to be stitched; Inputting a plurality of the to-be-stitched image samples into a preset deep neural network for feature extraction, sliding matching and image stitching to obtain a predicted stitched image, wherein the sliding matching refers to sliding a window to continuously compare the similarity of feature values ​​of two to-be-stitched image samples in the window; The deep neural network is trained according to the real stitching image and the predicted stitching image to obtain the image stitching model.

2. The slab number recognition method according to claim 1, characterized in that: The steps of continuously acquiring a plurality of side images of the slab on the slab transport roller table include: Using a pre-set slab in-place detection device to obtain a slab in-place signal; According to the slab arrival signal, the roller table running speed is obtained; According to the roller running speed and the preset shooting rules, the target shooting interval is obtained; According to the target shooting interval, a plurality of side images of the slab on both sides of the slab transport roller are regularly collected.

3. The slab number recognition method according to claim 1, characterized in that: The step of acquiring the first recognition model comprises: Acquire a first training set, wherein the first training set includes: a plurality of training samples, and true category information corresponding to the training samples; Inputting the training sample into a preset first convolutional neural network for feature extraction to obtain first sample features; Obtaining a predicted image category by classifying the first sample features; According to the predicted image category and the actual category information, the first convolutional neural network is iteratively trained to obtain the first recognition model.

4. The slab number recognition method according to claim 1, characterized in that: The step of inputting a plurality of the to-be-stitched image samples into a preset deep neural network for feature extraction, sliding matching and image stitching, and obtaining a predicted stitched image comprises: Extracting and encoding features of a plurality of continuous image samples to be spliced, and obtaining feature values ​​of a plurality of continuous image samples to be spliced; In units of columns, sliding matching is performed on the feature values ​​of two adjacent image samples to be spliced ​​to obtain a matching result; Image stitching is performed according to the matching result to obtain the predicted stitched image.

5. The slab number recognition method according to claim 1, characterized in that: The step of acquiring the second recognition model comprises: Acquire a third training set, the third training set comprising: a plurality of samples to be identified, and real labeling information corresponding to the samples to be identified; Inputting the sample to be identified into a preset second convolutional neural network for feature extraction to obtain second sample features; Obtaining a prediction result by performing feature matching on the second sample feature; Training the second convolutional neural network once according to the prediction result and the sample content in the real annotation information; Comparing the second sample feature with a preset feature to be compared to obtain a predicted comparison result; According to the predicted comparison result and the actual comparison result in the actual annotation information, the second convolutional neural network that has been trained once is trained for the second time to obtain the second recognition model.

6. The slab number recognition method according to claim 1, characterized in that: By comparing the slab number identification value with a preset slab number expected value, the step of issuing a warning includes: If the slab number identification value is different from the slab number expected value, a warning is issued, and the corresponding stitching image is recorded as an error image; By manually verifying the error image, determining whether the slab code in the error image is correct; If the slab code in the error image is correct, the error image is marked as a correct image, and the number of markings is recorded; When the number of markings exceeds a preset number threshold, the first recognition model, the image stitching model and the second recognition model are retrained and optimized to complete the slab number recognition.

7. A slab number recognition system, characterized in that: include: An image acquisition module, used for continuously acquiring multiple side images of the slab on the slab transport roller; A first recognition module is used to input the slab side image into a preset first recognition model for feature extraction and classification, and obtain the image of the slab to be spliced, wherein the image of the slab to be spliced ​​is a slab side image with letter and / or number features; An image stitching module, used for inputting the image of the slab to be stitched into a preset image stitching model to perform image stitching and obtain a stitched image; A second recognition module is used to input the spliced ​​image into a preset second recognition model for recognition, and obtain a slab number recognition value; A comparison module is used to compare the slab number identification value with a preset slab number expected value, issue a warning, and complete the slab number identification; the image acquisition module, the first identification module, the image stitching module and the second identification module are connected; The step of acquiring the image stitching model includes: Acquire a second training set, wherein the second training set includes: a plurality of image samples to be stitched, and real stitched images corresponding to the image samples to be stitched; Inputting a plurality of the to-be-stitched image samples into a preset deep neural network for feature extraction, sliding matching and image stitching to obtain a predicted stitched image, wherein the sliding matching refers to sliding a window to continuously compare the similarity of feature values ​​of two to-be-stitched image samples in the window; The deep neural network is trained according to the real stitching image and the predicted stitching image to obtain the image stitching model.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

9. An electronic terminal, characterized in that: include: Processor and memory; The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the terminal executes the method according to any one of claims 1 to 6.

Images