Intelligent rating method for banded structures in steel based on artificial intelligence and related device

Abstract

The invention discloses an intelligent rating method for banded structures in steel based on artificial intelligence and a related device, and relates to the field of artificial intelligence, and the method comprises the steps: cutting a banded structure image in the steel to be detected into a plurality of sub-images to be detected, and for each sub-image to be detected, inputting the sub-image to be detected into a banded grade evaluation model, outputting a prediction probability that the rating of the to-be-detected sub-image is not lower than each grade through a strip-shaped grade evaluation model; and for each to-be-detected sub-image, determining a target grade of the to-be-detected sub-image based on the condition that the grade of the to-be-detected sub-image is not lower than the prediction probability corresponding to each grade. And determining the maximum target grade in the target grades corresponding to the plurality of to-be-detected sub-images as the grade of the banded structure image in the steel. Manual evaluation is not needed in the whole process, and rapid and objective grading is achieved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an intelligent rating method and related apparatus for strip structure in steel based on artificial intelligence. Background Technology

[0002] Banded structures in steel are microstructural defects caused by compositional segregation and improper hot working processes, significantly affecting the uniformity of mechanical properties. In related technologies, the rating of banded structures in steel is determined manually, using the following methods:

[0003] Extract banded features from images of banded structures in steel, calculate the geometric features of the banded features such as size, quantity, and location; and manually grade based on the geometric features.

[0004] The subjective nature of manual assessment in related technologies can lead to inaccurate ratings; a single manual assessment takes a long time, generally about ten minutes to assess a single image of banded tissue in steel. Summary of the Invention

[0005] In view of the above problems, this application provides an intelligent rating method and related apparatus for banded microstructures in steel based on artificial intelligence, so as to achieve rapid and accurate rating of banded microstructure images in steel. The specific solution is as follows:

[0006] The first aspect of this application provides an intelligent rating method for banded microstructure in steel based on artificial intelligence, including:

[0007] Acquire multiple sub-images of the banded microstructure in the steel to be tested;

[0008] For each of the sub-images to be tested, the sub-image is input into a pre-constructed strip-shaped rating model, and the strip-shaped rating model outputs that the rating of the sub-image to be tested is not lower than the predicted probability corresponding to each level.

[0009] The banded grade assessment model is trained by taking a sample sub-image of the banded structure image in the sample steel as input and taking the rating of the sample sub-image of the banded structure image in the sample steel not lower than the annotation probability corresponding to each grade as the training objective.

[0010] For each sub-image to be tested, the target level of the sub-image to be tested is determined based on the prediction probability corresponding to each level, which is not lower than the rating of the sub-image to be tested.

[0011] The highest target level among the target levels corresponding to the plurality of sub-images to be tested is determined as the level of the banded structure image in the steel.

[0012] In one possible implementation, determining the target level of the sub-image under test based on the fact that the rating of the sub-image under test is not lower than the predicted probability corresponding to each level includes:

[0013] Based on the predicted probability that the rating of the sub-image to be tested is not lower than the predicted probability corresponding to each level, the predicted probability that the sub-image to be tested belongs to each level is determined.

[0014] The level corresponding to the highest predicted probability among the predicted probabilities that are greater than or equal to a preset threshold is determined as the target level of the sub-image to be tested.

[0015] In one possible implementation, determining the target level of the sub-image under test based on the fact that the rating of the sub-image under test is not lower than the predicted probability corresponding to each level includes:

[0016] The rating of the sub-image to be tested is not lower than the sum of the predicted probabilities corresponding to each level, so as to obtain the target expected value;

[0017] From the preset correspondence between expected range and level, find the target level corresponding to the expected range to which the target expected value belongs.

[0018] In one possible implementation, the strip-shaped rating model includes: a first convolutional layer, a first MBConv6 module whose input is connected to the output of the first convolutional layer, a second MBConv6 module whose input is connected to the output of the first MBConv6 module, an (i+1)th MBConv6 module whose input is connected to the output of the ith MBConv6 module, a second convolutional layer whose input is connected to the output of the Nth MBConv6 module, a pooling layer whose input is connected to the second convolutional layer, a fully connected layer whose input is connected to the output of the pooling layer, a first neuron whose input is connected to the output of the fully connected layer, and a plurality of second neurons whose inputs are respectively connected to the outputs of the first neuron;

[0019] Wherein, the second neuron outputs a predicted probability that the rating of the banded tissue in the steel is not lower than one level, and different second neurons correspond to different levels; i is a positive integer greater than or equal to 2 and less than or equal to N-1, and N is a positive integer greater than 2; the input end of the first convolutional layer is the input end of the banded tissue rating model.

[0020] In one possible implementation, the formula for the k-th second neuron is: y = zb k Where b0=s1, b1=b0+s2, b2=b1+s3, ..., b k =b k-1 +s k ; where s1, s2, ..., s kare the parameters that need to be trained in the strip-shaped rating model, and z is the output feature of the fully connected layer.

[0021] In one possible implementation, the step of acquiring multiple sub-images of the banded microstructure image in the steel to be tested includes:

[0022] Obtain an image of the banded microstructure in the steel to be tested;

[0023] Determine the target cutting diameter and overlap ratio;

[0024] The horizontal sliding step and the vertical sliding step are calculated based on the target cutting diameter and the overlap rate.

[0025] According to the said lateral sliding step size, the sub-image of the test with the said target cutting diameter is laterally cropped from the strip structure image of the steel to be tested;

[0026] According to the longitudinal sliding step size, the sub-image of the test with the target cutting diameter is longitudinally cropped from the strip structure image of the steel to be tested.

[0027] A second aspect of this application provides an artificial intelligence-based intelligent rating device for banded microstructure in steel, comprising:

[0028] The first acquisition module is used to acquire multiple sub-images of the banded microstructure in the steel to be tested;

[0029] The second acquisition module is used to input the sub-image to be tested into a pre-constructed strip-shaped rating model for each sub-image to be tested, and output the rating of the sub-image to be tested as not lower than the predicted probability corresponding to each level through the strip-shaped rating model.

[0030] The banded grade assessment model is trained by taking a sample sub-image of the banded structure image in the sample steel as input and taking the rating of the sample sub-image of the banded structure image in the sample steel not lower than the annotation probability corresponding to each grade as the training objective.

[0031] The first determining module is used to determine the target level of each sub-image to be tested based on the prediction probability that the rating of the sub-image to be tested is not lower than the respective level.

[0032] The second determining module is used to determine the maximum target level among the target levels corresponding to the plurality of sub-images to be tested as the level of the banded structure image in the steel.

[0033] A third aspect of this application provides a computer program product, including computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the AI-based intelligent rating method for strip-like structures in steel as described in the first aspect or any implementation thereof.

[0034] A fourth aspect of this application provides an electronic device, including at least one processor and a memory connected to the processor, wherein:

[0035] The memory is used to store computer programs;

[0036] The processor is used to execute the computer program so that the electronic device can implement the artificial intelligence-based intelligent rating method for strip structure in steel, which is based on the first aspect or any implementation thereof.

[0037] The fifth aspect of this application provides a computer storage medium carrying one or more computer programs that, when executed by an electronic device, enable the electronic device to implement the artificial intelligence-based intelligent rating method for strip structure in steel as described in the first aspect or any implementation thereof.

[0038] Using the above technical solution, this application provides an intelligent rating method for banded microstructure in steel based on artificial intelligence. The method acquires multiple sub-images of a banded microstructure image in the steel to be tested. For each sub-image, the sub-image is input into a pre-constructed banded microstructure rating model. The model outputs the predicted probability that the rating of the sub-image is not lower than the corresponding level. Since the banded microstructure rating model is trained using sub-images of sample sub-images of a sample steel banded microstructure image as input and the labeled probability that the rating of the sample sub-image is not lower than the corresponding level as the training objective, the banded microstructure rating model can accurately output the predicted probability that the rating of the sub-image is not lower than the corresponding level. For each sub-image, based on the predicted probability that the rating is not lower than the corresponding level, the target level of the sub-image is determined. A higher level indicates a lower quality of the sample containing banded microstructure characteristics corresponding to the banded microstructure image in the steel to be tested; therefore, the maximum target level is determined as the level of the banded microstructure image in the steel to be tested. This application introduces an artificial intelligence model, eliminating the need for human assessment of the grade of the banded microstructure image in the steel to be tested. Attached Figure Description

[0039] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0040] Figure 1 A schematic diagram of a system architecture is provided for this application;

[0041] Figure 2 A flowchart illustrating an intelligent rating method for banded microstructure in steel based on artificial intelligence, provided for an embodiment of this application;

[0042] Figure 3 This is a schematic diagram of the structure of the strip-shaped rating model provided in the embodiments of this application;

[0043] Figure 4 A schematic diagram of the image of the sigmoid function provided in the embodiments of this application;

[0044] Figure 5 A schematic diagram of the banded microstructure image and the sub-image of the steel to be tested provided in the embodiments of this application;

[0045] Figure 6 A schematic diagram of the structure of an intelligent rating device for strip-shaped steel structure based on artificial intelligence, provided for an embodiment of this application;

[0046] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0047] The embodiments of this application are described below with reference to the accompanying drawings. The terminology used in the implementation section of this application is for explaining specific embodiments only and is not intended to limit the scope of this application.

[0048] The embodiments of this application will now be described with reference to the accompanying drawings. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are equally applicable to similar technical problems.

[0049] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.

[0050] In related technologies, the rating of banded structures in steel is determined manually, and the specific method is as follows:

[0051] This method extracts banded features from microscopic images of steel microstructures and calculates their geometric characteristics, such as size, quantity, and location. A rating is then assigned manually based on these geometric features. However, this method suffers from poor robustness and generalization in banded feature extraction, and is easily affected by factors such as color and scratches on the metal surface. Consequently, the banded feature extraction rate is low, impacting the rating accuracy.

[0052] Based on this, this application provides an intelligent rating method for banded microstructures in steel based on artificial intelligence. This application mainly utilizes artificial intelligence algorithms to achieve the rating function of banding degree in images of banded microstructures in steel. Samples containing banded microstructure features are photographed using optical and electron microscopes to acquire microscopic images of the banded microstructures. The artificial intelligence algorithm is then used to rate the banding degree of these microscopic images. For example, the rating is generally from 0 to 5 levels, typically integer levels, but half-level ratings are allowed. This application introduces artificial intelligence algorithms to enhance the robustness and generalization ability of the rating.

[0053] The technical solutions involved in this application are described below.

[0054] See Figure 1 , Figure 1 A schematic diagram of a system architecture is shown. The system may include a terminal 100 and a server 200. The server 200 can provide the methods provided in the embodiments of this application to one or more terminals.

[0055] The terminal 100 may have an application installed on it. The application and webpage can provide an interface. The terminal 100 can receive relevant parameters input by the user on the interface, such as multiple sub-images of the banded tissue in the steel to be tested, and send the parameters to the server 200. The server 200 can obtain the processing result based on the received parameters and return the processing result to the terminal 100.

[0056] It should be understood that in some optional implementations, the terminal 100 can also complete the action of obtaining the processing result based on the received parameters on its own, without the need for the server to cooperate. This application embodiment is not limited to this.

[0057] The following description Figure 1 The product form of the mid-terminal 100;

[0058] The terminal 100 in this application embodiment can be a mobile phone, tablet computer, wearable device, vehicle device, augmented reality (AR) / virtual reality (VR) device, laptop computer, ultra-mobile personal computer (UMPC), netbook, personal digital assistant (PDA), etc., and this application embodiment does not impose any restrictions on it.

[0059] Terminal 100 may include a radio frequency unit, memory, input unit, display unit, camera (optional), audio circuitry (optional), speaker (optional), microphone (optional), headphone jack (optional), processor, external interface, power supply, and other components. Those skilled in the art will understand that the above-mentioned components are merely examples and do not constitute a limitation on the terminal or multifunctional device; it may include more or fewer components, or a combination of certain components, or different components.

[0060] The input unit can be used to receive input numeric or character information, and to generate key signal inputs related to user settings and function control of the portable multi-functional device. Specifically, the input unit may include a touchscreen (optional) and / or other input devices. Other input devices may include, but are not limited to, one or more of a physical keyboard, function keys (such as volume control buttons, power buttons, etc.), trackball, mouse, joystick, etc.

[0061] Among them, the input device can receive input data, etc.

[0062] The display unit can be used to display information input by the user or information provided to the user, various menus of the terminal, interactive interfaces, file display, and / or playback of any multimedia file. In the embodiments of this application, the display unit can be used to display interfaces, processing results, etc.

[0063] The memory can be used to store software code related to the AI-based intelligent rating method for banded structures in steel. The processor can execute the steps of the AI-based intelligent rating method for banded structures in steel, and can also schedule other units (such as the aforementioned input unit and display unit) to achieve the corresponding functions.

[0064] This radio frequency unit (optional) can be used to receive and send signals during information transmission or calls.

[0065] In this embodiment of the application, the radio frequency unit can send data to the server 200 and receive the processing results sent by the server 200.

[0066] It should be understood that this radio frequency unit is optional and can be replaced with other communication interfaces, such as a network port.

[0067] Terminal 100 also includes a power source (such as a battery) for supplying power to the various components.

[0068] Terminal 100 also includes an external interface, which can be a standard Micro USB interface or a multi-pin connector, which can be used to connect terminal 100 to other devices for communication or to connect a charger to charge terminal 100.

[0069] Server 200 includes a bus, a processor, a communication interface, and memory. The processor, memory, and communication interface communicate with each other via the bus.

[0070] The memory can be used to store software code related to the AI-based intelligent rating method for banded structures in steel. The processor can execute the steps of the AI-based intelligent rating method for banded structures in steel on the chip, and can also schedule other units to achieve corresponding functions.

[0071] Reference Figure 2 , Figure 2 A flowchart illustrating an artificial intelligence-based intelligent rating method for banded microstructure in steel, as provided in this application embodiment, is shown below. Figure 2 As shown in the embodiment of this application, an intelligent rating method for banded structures in steel based on artificial intelligence may include steps S201 to S204, which are described in detail below.

[0072] Step S201: Obtain multiple sub-images of the banded microstructure in the steel to be tested.

[0073] In this application, the banded microstructure image of the steel grade to be tested is referred to as the banded microstructure image of the steel to be tested.

[0074] For example, the method for obtaining images of banded structures in steel is as follows: by using an optical or electron microscope to photograph a sample containing banded structure features, a microscopic image of the banded structure of the sample can be acquired.

[0075] Multiple sub-images to be tested are cropped from the strip structure image of the steel to be tested, and the multiple sub-images to be tested are the same size.

[0076] Step S202: For each of the sub-images to be tested, the sub-image to be tested is input into a pre-constructed strip rating model, and the strip rating model outputs that the rating of the sub-image to be tested is not lower than the predicted probability corresponding to each level.

[0077] The banded grade assessment model is trained by taking sample sub-images of the banded structure image in the sample steel as input and using the rating of the sample sub-images of the banded structure image in the sample steel not lower than the annotation probability corresponding to each grade as the training objective.

[0078] Multiple sample sub-images were cropped from the strip structure image of the sample steel, and the multiple sample sub-images were of the same size.

[0079] For example, the size of the sample sub-image is the same as the size of the sub-image to be tested.

[0080] For example, the band rating model can be any type of neural network model, and no limitation is made here.

[0081] The following example illustrates the statement that "the rating of the sub-image to be tested is not lower than the prediction probability corresponding to each level".

[0082] Assuming there are 6 levels, and the levels are: Level 0, Level 1, Level 2, Level 3, Level 4, and Level 5, then P(y>0) is the predicted probability that the rating of the sub-image to be tested is not lower than Level 0, i.e., P(y>0) = P(y=1) + P(y=2) + P(y=3) + P(y=4) + P(y=5); P(y>1) is the predicted probability that the rating of the sub-image to be tested is not lower than Level 1, i.e., P(y>1) = P(y=2) + P(y=3) + P(y=4) + P(y=5), and so on, without further explanation.

[0083] Assume there are 8 levels, for example, the levels are: Level 0, Level 1, Level 2, Level 2.5, Level 3, Level 3.5, Level 4, and Level 5. Since Levels 2.5 and 3.5 are very difficult to label manually, only Levels 0, 1, 2, 3, 4, and 5 are labeled manually, i.e., P(y>0) = P(y=1) + P(y=2) + P(y=2.5) + P(y=3) + P(y=3.5) + P(y=4) + P(y=5). P(y>1) = P(y=2) + P(y=2.5) + P(y=3) + P(y=4) + P(y=5), and so on, without further explanation.

[0084] The following example illustrates that "the rating of the sample sub-image of the banded structure image in the sample steel is not lower than the annotation probability corresponding to each level".

[0085] It is understandable that the rating of the sample sub-images of the banded structure image in the sample steel is known, so the rating of the sample sub-images of the banded structure image in the sample steel is not lower than the annotation probability corresponding to each level, which is accurate. Assume there are 6 levels, namely: Level 0, Level 1, Level 2, Level 3, Level 4, and Level 5. Assume the rating of a sample sub-image of the banded structure image in a certain sample steel is Level 3. Then, P(y>0)=1, P(y>1)=1, P(y>2)=1, P(y>3)=0, and P(y>4)=0. For example, the annotation result can be represented by a matrix as [1, 1, 1, 0, 0].

[0086] For example, if there are 6 levels, and the levels are: Level 0, Level 1, Level 2, Level 3, Level 4, and Level 5, then the training set for training the strip rating model can include 6 folders. Each folder includes sample sub-images of the same level and the rating of the sample sub-images is not lower than the annotation probability corresponding to each level. Different folders correspond to different levels.

[0087] Step S203: For each of the sub-images to be tested, based on the fact that the rating of the sub-image to be tested is not lower than the prediction probability corresponding to each level, determine the target level of the sub-image to be tested.

[0088] Step S204: Determine the highest target level among the target levels corresponding to the plurality of sub-images to be tested as the level of the banded structure image in the steel.

[0089] It is understandable that samples containing banded structure features may have scratches, which could affect the banding degree rating. To address this issue, the banded structure image in the steel under test is cropped into multiple sub-images. The probability that all sub-images have scratches is low. Based on obtaining the target grade corresponding to each sub-image, the grade of the banded structure image in the steel under test is determined, significantly reducing the impact of scratches on the banding degree rating.

[0090] Understandably, a higher grade indicates a lower quality of the sample containing banded structure features corresponding to the banded structure image in the steel. Therefore, the maximum target grade is determined to be the grade of the banded structure image in the steel to be tested.

[0091] This application provides an artificial intelligence-based intelligent rating method for banded microstructure in steel. The method acquires multiple sub-images of a sample image of banded microstructure in steel. For each sub-image, it is input into a pre-constructed banded microstructure rating model. The model outputs the predicted probability that the rating of the sub-image is not lower than the corresponding level. Since the banded microstructure rating model is trained using sub-images of sample sub-images of a sample image of banded microstructure in steel as input and the labeled probability that the rating of the sample sub-image is not lower than the corresponding level as the training objective, it can accurately output the predicted probability that the rating of the sub-image is not lower than the corresponding level. For each sub-image, based on the predicted probability that the rating is not lower than the corresponding level, the target level of the sub-image is determined. A higher level indicates a lower quality sample containing banded microstructure features, so the maximum target level is determined as the level of the sample image of banded microstructure in steel. This application introduces an artificial intelligence model, eliminating the need for human assessment of the grade of the banded microstructure image in the steel to be tested.

[0092] It is understood that there are multiple ways to implement step S203, and the embodiments of this application provide, but are not limited to, the following two methods.

[0093] The first method of implementing step S203 includes the following steps A11 to A12.

[0094] Step A11: Based on the prediction probability that the rating of the sub-image to be tested is not lower than the prediction probability corresponding to each level, determine the prediction probability that the sub-image to be tested belongs to each level.

[0095] For example, the predicted probability that the sub-image to be tested belongs to rating g is P(y=g)= P(y>g)- P(y>(g+1)), where g is any positive integer greater than or equal to 0.

[0096] Step A12: Determine the level corresponding to the highest predicted probability among the predicted probabilities that are greater than or equal to a preset threshold as the target level of the sub-image to be tested.

[0097] For example, the preset threshold can be determined based on the actual situation, and there is no limitation here. For example, the preset threshold is 0.5.

[0098] The second method of implementing step S203 includes the following steps A21 to A22.

[0099] Step A21: Calculate the sum of the predicted probabilities corresponding to each level for the rating of the sub-image to be tested to obtain the target expected value.

[0100] Step A22: From the preset correspondence between expected range and level, find the target level corresponding to the expected range to which the target expected value belongs.

[0101] It is understandable that the correspondence between expected range and level can be stored in any form. The following example uses a table to illustrate the correspondence between expected range and level, as shown in Table 1.

[0102] Table 1

[0103] Understandably, manual annotation of semi-levels like 2.5 and 3.5 is difficult, and this application allows for semi-level ratings. The following example illustrates this: assuming a sub-image to be tested has a rating no lower than [a certain level], the predicted probabilities for each level are as follows:

[0104] Given P(y>0) = 0.99, P(y>1) = 0.95, P(y>2) = 0.8, P(y>3) = 0.54, and P(y>4) = 0.1, the target expected value = P(y>0) + P(y>1) + P(y>2) + P(y>3) + P(y>4) = 3.38 ∈ [3.5, 4.0). Therefore, the level of the sub-image to be tested is determined to be level 3.5.

[0105] The structure of the strip-shaped rating model is explained below.

[0106] like Figure 3 The diagram shown is a structural schematic of the strip-shaped rating model provided in an embodiment of this application.

[0107] The band-shaped rating model includes: a first convolutional layer, a first MBConv6 module whose input is connected to the output of the first convolutional layer, a second MBConv6 module whose input is connected to the output of the first MBConv6 module, an (i+1)th MBConv6 module whose input is connected to the output of the ith MBConv6 module, a second convolutional layer whose input is connected to the output of the Nth MBConv6 module, a pooling layer whose input is connected to the second convolutional layer, a fully connected layer (FC layer) whose input is connected to the output of the pooling layer, a first neuron whose input is connected to the output of the fully connected layer, and multiple second neurons whose inputs are respectively connected to the outputs of the first neuron.

[0108] Wherein, the second neuron outputs a predicted probability that the rating of the banded tissue in the steel is not lower than one level, and different second neurons correspond to different levels; i is a positive integer greater than or equal to 2 and less than or equal to N-1, and N is a positive integer greater than 2; the input end of the first convolutional layer is the input end of the banded tissue rating model.

[0109] Among them, the MBConv6 module is the Mobile Inverted Bottleneck convolution module.

[0110] Figure 3 The example below uses five second neurons as an example. Assuming that the levels are level 0, level 1, level 2, level 3, level 4 and level 5, the outputs of the five second neurons are: P(y>0), P(y>1), P(y>2), P(y>3) and P(y>4).

[0111] For example, the first convolutional layer has a 3×3 convolutional kernel, and the second convolutional layer has a 1×1 convolutional kernel.

[0112] For example, Figure 3 The output of the pooling layer in the model has multiple features, such as 1280 features.

[0113] It is understandable that either the first neuron or the second neuron represents a firing formula y = f(x), and the connections between neurons follow a set of linear calculation formulas. An example is provided below.

[0114] For example, the 1280 features obtained from the pooling layer are propagated to the first neuron. The activation formula of the first neuron can be z=wx, where z is the output of the first neuron, x is the input of the first neuron, and w is a constant coefficient that is learned through backpropagation and iteratively updated.

[0115] The feature z output by the first neuron can be input to multiple second neurons. For example, the activation formula for the k-th second neuron is: y = zb k Where b0=s1, b1=b0+s2, b2=b1+s3, ..., b k =b k-1 +s k ; where s1, s2, ..., s k are the parameters that need to be trained in the strip-shaped rating model, and z is the output feature of the fully connected layer.

[0116] Among them, b k For the corresponding threshold of each second neuron, the b kIt will learn through backpropagation and then iterate and update.

[0117] For example, the activation function of either the first neuron or the second neuron is sigmoid(y).

[0118] Among them, s k All are positive numbers. For example, s k The way to control it as a positive number is s k = e （qk） Therefore, in actual training, the most crucial parameter to change is qk, and qk can be any number from negative infinity to positive infinity, which aligns with the training logic. This ensures that b... k Since it is monotonically increasing, the output received by multiple second neurons will be monotonically decreasing, thus making sigmoid(y) monotonically decreasing. For example... Figure 4 The image shown is a schematic diagram of the sigmoid function provided in an embodiment of this application.

[0119] Combination Figure 4 As we know, the result of the sigmoid function graph is 0-1 and monotonic. Therefore, the output of multiple second neurons can be 0-1 with a probability. The output of multiple second neurons decreases monotonically according to P(y>0), P(y>1), ... . The following explanation is based on the following levels: level 0, level 1, level 2, level 3, level 4 and level 5. That is, the value of P(y>0) to P(y>4) decreases monotonically, so that the output y of multiple second neurons conforms to the logic of the comment.

[0120] For example, the optimizer for the strip rating model is AdamW, and the learning rate is 3e-4.

[0121] The applicant validated the strip rating model using a validation set, achieving a training accuracy of 93% and an inference time of 0.4 seconds per inference.

[0122] It is understood that there are multiple ways to implement step S201. The embodiments of this application provide, but are not limited to, the following method, which includes the following steps B1 to B5.

[0123] Step B1: Obtain an image of the banded microstructure in the steel to be tested.

[0124] Step B2: Determine the target cutting diameter and overlap ratio.

[0125] Assume the size of the banded microstructure image in the steel to be tested is H×W, the target cropping diameter is t_w, and the overlap rate is r. For example, the overlap rate can be 80%, so as to preserve the banded microstructure of each region in the image of the banded microstructure in the steel to be tested as completely as possible.

[0126] For example, the target cutting diameter can be determined based on the actual situation and is not limited here; for example, it can be 0.8mm.

[0127] Step B3: Calculate the horizontal sliding step length and the vertical sliding step length based on the target cutting diameter and the overlap rate.

[0128] For example, the horizontal sliding step size and the vertical sliding step size are both S_w=t_w×(1-r).

[0129] Step B4: According to the said lateral sliding step length, laterally cut the sub-image of the steel to be tested with the said target cutting diameter from the strip structure image of the steel to be tested.

[0130] Step B5: According to the longitudinal sliding step size, longitudinally cut the sub-image of the steel to be tested with the target cutting diameter from the strip structure image of the steel to be tested.

[0131] like Figure 5 The image shown is a schematic diagram of the banded microstructure image and the sub-image of the steel to be tested provided in the embodiment of this application.

[0132] like Figure 5 As shown, the left side is an image of the banded structure in the steel to be tested, and the right side is a sub-image of the steel to be tested, which is a circular image.

[0133] For example, the sub-image to be tested can be converted into a grayscale image to eliminate the influence of color information on the rating.

[0134] For example, the sub-image to be tested can be flipped or rotated to expand the training set. Since the sub-image to be tested is circular, the image augmentation does not affect the image size.

[0135] The above describes an artificial intelligence-based intelligent rating method for banded structures in steel, as provided in the embodiments of this application. The following describes the apparatus for performing the above-described artificial intelligence-based intelligent rating method for banded structures in steel.

[0136] Please see Figure 6 , Figure 6 This is a schematic diagram of a smart rating device for strip-shaped steel structures based on artificial intelligence, provided as an embodiment of this application. Figure 6 As shown, the AI-based intelligent rating device for banded structures in steel includes:

[0137] The first acquisition module 601 is used to acquire multiple sub-images of the banded microstructure in the steel to be tested;

[0138] The second acquisition module 602 is used to input the sub-image to be tested into a pre-constructed strip-shaped rating model for each sub-image to be tested, and output the rating of the sub-image to be tested as not lower than the predicted probability corresponding to each rating through the strip-shaped rating model.

[0139] The banded grade assessment model is trained by taking a sample sub-image of the banded structure image in the sample steel as input and taking the rating of the sample sub-image of the banded structure image in the sample steel not lower than the annotation probability corresponding to each grade as the training objective.

[0140] The first determining module 603 is used to determine the target level of each sub-image to be tested based on the prediction probability that the rating of the sub-image to be tested is not lower than the respective level.

[0141] The second determining module 604 is used to determine the maximum target level among the target levels corresponding to the plurality of test sub-images as the level of the banded structure image in the steel.

[0142] In one alternative implementation, the first determining module includes:

[0143] The first determining unit is used to determine the predicted probability that the sub-image under test belongs to each rating based on the predicted probability that the rating of the sub-image under test is not lower than the predicted probability corresponding to each rating.

[0144] The second determining unit is used to determine the level corresponding to the highest predicted probability among the predicted probabilities that are greater than or equal to a preset threshold as the target level of the sub-image to be tested.

[0145] In one alternative implementation, the first determining module includes:

[0146] The first calculation unit is used to calculate the sum of the predicted probabilities that the rating of the sub-image to be tested is not lower than the sum of the predicted probabilities of each level, so as to obtain the target expected value.

[0147] The search unit is used to search for the target level corresponding to the expected range to which the target expected value belongs, from the preset correspondence between expected range and level.

[0148] In one optional implementation, the strip-shaped rating model includes: a first convolutional layer, a first MBConv6 module whose input is connected to the output of the first convolutional layer, a second MBConv6 module whose input is connected to the output of the first MBConv6 module, an (i+1)th MBConv6 module whose input is connected to the output of the ith MBConv6 module, a second convolutional layer whose input is connected to the output of the Nth MBConv6 module, a pooling layer whose input is connected to the second convolutional layer, a fully connected layer whose input is connected to the output of the pooling layer, a first neuron whose input is connected to the output of the fully connected layer, and a plurality of second neurons whose inputs are respectively connected to the outputs of the first neuron;

[0149] Wherein, the second neuron outputs a predicted probability that the rating of the banded tissue in the steel is not lower than one level, and different second neurons correspond to different levels; i is a positive integer greater than or equal to 2 and less than or equal to N-1, and N is a positive integer greater than 2; the input end of the first convolutional layer is the input end of the banded tissue rating model.

[0150] In one alternative implementation, the formula for the k-th second neuron is: y = zb k Where b0=s1, b1=b0+s2, b2=b1+s3, ..., b k =b k-1 +s k ; where s1, s2, ..., s k are the parameters that need to be trained in the strip-shaped rating model, and z is the output feature of the fully connected layer.

[0151] In one optional implementation, the first acquisition module includes:

[0152] The first acquisition unit is used to acquire an image of the banded microstructure in the steel to be tested;

[0153] The third determining unit is used to determine the target cutting diameter and overlap rate;

[0154] The second calculation unit is used to calculate the lateral sliding step and the longitudinal sliding step based on the target cutting diameter and the overlap rate.

[0155] The first trimming unit is used to horizontally cut out the sub-image of the steel to be tested with the target trimming diameter from the strip structure image of the steel to be tested according to the horizontal sliding step length.

[0156] The second trimming unit is used to longitudinally cut out the sub-image of the steel to be tested with the target trimming diameter from the strip structure image of the steel to be tested according to the longitudinal sliding step length.

[0157] This application also provides an electronic device in its embodiments. (See reference...) Figure 7 The diagram illustrates a structural schematic suitable for implementing the electronic device in the embodiments of this application. The electronic device in the embodiments of this application may include, but is not limited to, fixed terminals such as mobile phones, laptops, PDAs (personal digital assistants), PADs (tablet computers), desktop computers, etc. Figure 7 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0158] like Figure 7 As shown, the electronic device may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage device 708 into a random access memory (RAM) 703. When the electronic device is powered on, the RAM 703 also stores various programs and data required for the operation of the electronic device. The processing unit 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0159] Typically, the following devices can be connected to I / O interface 705: input devices 706 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 707 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 708 including, for example, memory cards, hard drives, etc.; and communication devices 709. Communication device 709 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 7 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or have alternatively.

[0160] This application also provides a computer program product including computer-readable instructions, which, when executed on an electronic device, cause the electronic device to implement any of the artificial intelligence-based intelligent rating methods for steel banded structures provided in this application.

[0161] This application also provides a computer-readable storage medium carrying one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any of the artificial intelligence-based intelligent rating methods for steel banded structures provided in this application.

[0162] It should also be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the device embodiment drawings provided in this application, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines.

[0163] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware, or it can be implemented by special-purpose hardware including application-specific integrated circuits, special-purpose CPUs, special-purpose memory, special-purpose components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or special-purpose circuits. However, for this application, software program implementation is more often the preferred implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a computer floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, training equipment, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0164] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0165] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

Claims

1. An artificial intelligence-based intelligent banding rating method for steel, characterized by, The method comprises the following steps: obtaining a plurality of to-be-tested sub-images of a to-be-tested banded structure image of a steel; for each to-be-tested sub-image, inputting the to-be-tested sub-image into a pre-constructed banded structure grade evaluation model, and outputting a prediction probability that the to-be-tested sub-image is not lower than each grade corresponding to the prediction probability through the banded structure grade evaluation model; wherein the banded structure grade evaluation model is obtained by taking a sample sub-image of a sample banded structure image of a sample steel as input and taking a labeled probability that the sample sub-image of the sample banded structure image of the sample steel is not lower than each grade corresponding to the labeled probability as a training target; for each to-be-tested sub-image, determining a target grade of the to-be-tested sub-image based on the prediction probability that the to-be-tested sub-image is not lower than each grade corresponding to the prediction probability; determining a maximum target grade in the target grades corresponding to the plurality of to-be-tested sub-images as a grade of the banded structure image of the steel. 2.The method of claim 1, wherein the method further comprises: The method further comprises the following steps: determining a prediction probability that the to-be-tested sub-image belongs to each grade based on the prediction probability that the to-be-tested sub-image is not lower than each grade corresponding to the prediction probability; determining a target grade of the to-be-tested sub-image corresponding to a maximum prediction probability greater than or equal to a preset threshold. 3.The method of claim 1, wherein the method further comprises: The method further comprises the following steps: calculating a sum of the prediction probabilities that the to-be-tested sub-image is not lower than each grade corresponding to the prediction probability to obtain a target expectation value; finding a target grade corresponding to an expectation range to which the target expectation value belongs from a preset correspondence between expectation ranges and grades.

4. The intelligent rating method for banded microstructure in steel based on artificial intelligence according to claim 1, characterized in that, The banded structure grade evaluation model comprises: a first convolutional layer, a first MBConv6 module connected to an output end of the first convolutional layer, a second MBConv6 module connected to an output end of the first MBConv6 module, an i+1th MBConv6 module connected to an output end of the ith MBConv6 module, a second convolutional layer connected to an output end of the Nth MBConv6 module, a pooling layer connected to the second convolutional layer, a fully connected layer connected to an output end of the pooling layer, a first neuron connected to an output end of the fully connected layer, and a plurality of second neurons connected to the output end of the first neuron respectively. The second neurons output prediction probabilities that the banded structure of the steel is not lower than one grade, and different second neurons correspond to different grades; i is a positive integer greater than or equal to 2 and less than or equal to N-1, and N is a positive integer greater than 2; and an input end of the first convolutional layer is an input end of the banded structure grade evaluation model. 5.The method of claim 4, wherein the method further comprises: The formula of the kth second neuron is: y = z - b k ; wherein b0 = s1, b1 = b0 + s2, b2 = b1 + s3, …, b k = b k-1 + s k ; wherein s1, s2, …, s k are parameters to be trained in the strip rating model, and z is an output feature of the full connection layer. 6.The method of claim 1, wherein the method further comprises: determining a steel grade of the steel based on the steel grade information; and determining a banding level of the steel based on the banding information and the steel grade. The method further comprises the following steps: obtaining the banded structure image of the steel; determining a target cutting diameter and an overlap rate; and determining a plurality of to-be-tested sub-images of the banded structure image of the steel. The horizontal sliding step and the vertical sliding step are calculated based on the target cropping diameter and the overlap rate; The target sub-image with the target cropping diameter is horizontally cut from the banded structure image of the steel to be tested according to the horizontal sliding step; The target sub-image with the target cropping diameter is vertically cut from the banded structure image of the steel to be tested according to the vertical sliding step.

7. An artificial intelligence-based intelligent banding rating device for steel, characterized by, Comprise: A first acquisition module is configured to acquire a plurality of sub-images of a banded structure image of a steel to be tested; A second acquisition module is configured to input each of the sub-images into a pre-constructed banded structure grade evaluation model, and output a prediction probability that the grade of each of the sub-images is not lower than a respective grade of the banded structure grade evaluation model; The banded structure grade evaluation model is obtained by training a sample sub-image of a sample banded structure image of a sample steel as input and a label probability that the grade of the sample sub-image of the sample banded structure image of the sample steel is not lower than a respective grade as a training target; A first determination module is configured to determine a target grade of each of the sub-images based on the prediction probability that the grade of each of the sub-images is not lower than a respective grade; A second determination module is configured to determine a maximum target grade in the target grades corresponding to the plurality of sub-images as a grade of the banded structure image of the steel.

8. A computer program product, characterised in that, The computer readable instructions, when executed on an electronic device, cause the electronic device to implement the artificial intelligence-based intelligent grade evaluation method for banded structure in steel according to any one of claims 1 to 6.

9. An electronic device, comprising: Comprise at least one processor and a memory connected to the processor, wherein: The memory is configured to store a computer program; The processor is configured to execute the computer program to enable the electronic device to implement the artificial intelligence-based intelligent grade evaluation method for banded structure in steel according to any one of claims 1 to 6.

10. A computer storage medium, characterized in that, The storage medium carries one or more computer programs, which, when executed by an electronic device, enable the electronic device to implement the artificial intelligence-based intelligent grade evaluation method for banded structure in steel according to any one of claims 1 to 6.

Images