Steel pile identification method and equipment and computer readable storage medium
By adopting the steel stack identification method based on GRU and CNN in the field of steel metallurgy, the problem of relying on manual and requiring transformation of control systems in the prior art is solved, and automatic identification is achieved, and safety and efficiency are improved.
Patent Information
- Application Number
- CN202411953737.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-27
- Publication Date
- 2025-04-25
AI Technical Summary
In the field of steel metallurgy, the monitoring of steel piles mainly relies on manual labor, which poses safety risks and non-manual inspection methods require the transformation of existing control systems, resulting in low production impact and reliability.
The steel stack recognition method based on GRU and CNN is adopted to realize automatic identification of steel stacks by building data sets, training of steel stack recognition models, and using Adam optimizer to optimize model parameters.
Automatic identification of steel piles is realized, the safety and efficiency of monitoring is improved, the need for transformation of existing control systems is reduced, and the intelligence, efficiency and precision of production is improved.
Smart Images
Figure CN119888332A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of defect detection, and in particular relates to a steel pile identification method, equipment and computer-readable storage medium. Background Art
[0002] In the smelting process of steel metallurgy, during the rolling process of high-speed wire rods, steel piles may occur due to incorrect rolling speed and roll diameter settings, large steel temperature fluctuations, sudden roll breakage, etc. Once steel piles occur, they must be dealt with in a timely manner. When steel piles occur, wire rods may overflow the rolling track on a large scale. If observed closely by humans, there is a risk of accidental safety accidents. Therefore, how to identify and monitor steel piles is closely related to the safe production of high-speed wire rods during rolling.
[0003] At present, the existing monitoring of steel piles still relies mainly on manual monitoring, while most of the other non-human detection methods need to be connected to the existing control system and need to modify the existing control system. Installing a monitoring system on an ordinary steel rolling production line usually requires a long debugging and modification time, which affects production. On the other hand, there is also the factor of low reliability, which cannot guarantee the quality of steelmaking. Therefore, a new steel pile identification method is needed to realize the intelligence, efficiency and precision of industrial scenes.
[0004] The information disclosed in this background technology section is only intended to enhance the understanding of the overall background of the invention and should not be regarded as an acknowledgment or any form of suggestion that the information constitutes the prior art already known to a person skilled in the art. Summary of the invention
[0005] The purpose of the present invention is to solve the deficiencies in the above-mentioned prior art and to provide a steel pile identification method, device and computer-readable storage medium.
[0006] A first aspect of the present invention provides a steel pile identification method, comprising the following steps:
[0007] Building a data set based on the steel pile image data, wherein the data set includes a training set and a validation set;
[0008] The training set and the validation set are used in sequence to train and test the steel pile recognition model; wherein the steel pile recognition model is constructed by GRU, and feature extraction is performed based on CNN;
[0009] The steel pile image to be inspected is input into the trained steel pile recognition model, and the steel surface defect detection results are output.
[0010] In one embodiment of the present invention, the GRU processes data information, wherein:
[0011] The expression of the forget gate is as follows:
[0012] z t =g(W (x) x t +U (x) h t-1 )
[0013] The expression of the update gate is as follows:
[0014] r t =g(W (N) x t +U (r) h t-1 )
[0015] The expression of the current memory content is as follows:
[0016] h′ t =tanh(Wx t +r t ) t-1 )
[0017] The expression of the final memory of the previous time is as follows:
[0018] h t =z t ⊙h t-1 +(1-z t )⊙h′ i
[0019] Among them, W (x) , W (N) , U (x) , U (r) , W and U t-1 Represents different weight matrices, g() represents the sigmoid function, tanh represents the hyperbolic tangent function, x t Indicates the input at the current moment, h t-1 represents the hidden state at the previous moment, and ⊙ represents the dot product.
[0020] In one embodiment of the present invention, the loss function of the steel pile identification model is as follows:
[0021]
[0022] Where N is the length of the input sequence, is the predicted value of the model, y i is the true value.
[0023] In one embodiment of the present invention, an Adam optimizer is applied to the steel pile identification model.
[0024] In one embodiment of the present invention, the CNN is 1DCNN.
[0025] In one embodiment of the present invention, the 1DCNN includes an input layer, several convolutional layers, a pooling layer, a fully connected layer and an output layer.
[0026] A second aspect of the present invention provides a device, including a processor and a memory communicatively connected to the processor, wherein the memory stores computer instructions executable by the processor, and the steel pile identification method is implemented when the computer instructions are executed.
[0027] A third aspect of the present invention provides a computer-readable storage medium storing computer instructions, which implement the steel pile identification method when executed.
[0028] Compared with the prior art, the technical effects achieved by the present invention are as follows:
[0029] 1. GRU uses reset gate and update gate to regulate the transmission and retention of information. The reset gate is responsible for screening and deciding which information should be discarded, while the update gate is responsible for determining the degree of integration of old information and new input information. Thanks to its streamlined structural design, GRU has excellent computational efficiency and faster computing speed.
[0030] 2. Extract high-dimensional features in time or space from one-dimensional sequence data to perform effective data recognition and classification.
[0031] 3. By applying the Adam optimizer, the weight parameters and bias parameters in the model can be effectively optimized to minimize the loss. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] Figure 1 is a flow chart of a steel pile identification method according to an embodiment of the present invention;
[0033] Figure 2 is a block diagram of a GRU unit structure according to an embodiment of the present invention;
[0034] Figure 3 1DCNN module structure block diagram according to one embodiment of the present invention. DETAILED DESCRIPTION
[0035] The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the invention are shown in the accompanying drawings.
[0036] like Figures 1 to 3As shown, a steel pile identification method according to a preferred embodiment of the present invention comprises the following steps:
[0037] S1. Construct a data set based on the steel pile image data, wherein the data set includes a training set and a validation set.
[0038] The sources of steel pile image data include but are not limited to photos, videos or databases. The occurrence of steel pile can be artificial or real.
[0039] The defects in the steel surface defect image can be real damage, that is, damage caused during one or more normal, standard or recognized operating processes; it can also be artificial damage, that is, the steel surface is deliberately damaged by humans; it can also be synthetic damage, that is, defect image blocks are inserted into the image of the steel surface to represent or emphasize the defects.
[0040] The images and labels of the dataset can be divided into a training set and a validation set. The images and labels of the training set are used to train a steel pile recognition model (see below) to recognize steel pile conditions in one or more images, and the images and labels of the validation set are used to test or confirm the effectiveness of the trained steel pile recognition model.
[0041] S2. Use the training set and the verification set to train and test the steel pile recognition model in sequence; wherein, the steel pile recognition model is constructed through GRU, and feature extraction is performed based on CNN.
[0042] After careful adjustment of the parameters and function configuration, an efficient network training process can be designed. This process follows the established training cycles (epochs) and continuously improves the parameters of the steel pile recognition model until its performance reaches the preset qualified level. In other words, if the model's performance reaches this set performance standard, the training will be terminated early even if the scheduled training cycle has not been completed. In addition, if the value of the loss function fails to achieve a significant reduction after three consecutive training cycles, the training of the steel pile recognition model will be stopped immediately.
[0043] In step S2, we use GRU (Gated Recurrent Unit) to deeply process the data information. As an advanced recurrent neural network variant, GRU has a unique basic structure consisting of two key gating mechanisms: update gate and reset gate. These two gating mechanisms finely regulate the flow of information through carefully designed sigmoid functions, allowing the GRU model to intelligently and selectively retain or forget information from previous time steps, thereby effectively capturing the temporal dependencies in the data.
[0044] Specifically, the contents of GRU are as follows:
[0045] The expression of the forget gate is as follows:
[0046] z t =g(W (x) x t +U (x) h t-1 )
[0047] The expression of the update gate is as follows:
[0048] r t =g(W (N) x t +U (r) h t-1 )
[0049] The expression of the current memory content is as follows:
[0050] h′ t =tanh(Wx t +r t ) t-1 ) The expression of the final memory before time is as follows:
[0051] h t =z t ⊙h t-1 +(1-z t )⊙h′ i
[0052] Among them, W (x) , W (N) , U (x) , U (r) , W and U t-1 Represents different weight matrices, g() represents the sigmoid function, tanh represents the hyperbolic tangent function, x t Indicates the input at the current moment, h t-1 represents the hidden state at the previous moment, and ⊙ represents the dot product.
[0053] GRU controls the flow of information through update gates and reset gates. The update gate determines the degree of retention of the state information of the previous moment, affecting the long-term memory of the model. The reset gate controls the degree to which the previous state information is written into the current candidate set, helping the model "forget" unimportant information. This design makes GRU more efficient and accurate when processing complex time series data.
[0054] In some embodiments, CNN adopts 1DCNN (one-dimensional convolutional neural network). 1DCNN is a variant of CNN, which is specially used to process sequence data. It is usually composed of an input layer, several convolutional layers, pooling layers, fully connected layers and output layers. The alternating use of convolutional layers and pooling layers not only increases the depth of the model, but also reduces the data dimension while maintaining key information. This structural design makes 1DCNN particularly good at extracting high-dimensional features in time or space from one-dimensional sequence data, thereby performing effective data recognition and classification. Each convolution layer applies multiple convolution kernels to filter the input data. These convolution kernels can capture local features in the data; the subsequent pooling layer is responsible for reducing the feature dimension and sub-sampling the features to reduce the amount of calculation and prevent overfitting.
[0055] S3. Input the steel surface defect image to be detected into the trained steel pile recognition model, and output the steel surface defect detection result.
[0056] In some embodiments, the loss function of the fault detection model is as follows:
[0057]
[0058] Where N is the length of the input sequence, is the predicted value of the model, y i is the true value.
[0059] In order to improve the accuracy of the steel pile identification model, the loss function was optimized to accurately measure and minimize the difference between the prediction and the actual. This optimization strategy accelerates the error correction process by increasing the attention to larger error terms, that is, giving them higher weights, and ensures that the entire training process remains continuous and differentiable, making the optimization process smoother. In addition, this optimization is also intended to speed up the convergence of the model, improve the accuracy of the regression task, and effectively curb the occurrence of overfitting, thereby enhancing the model's generalization ability on new data.
[0060] In some embodiments, an Adam optimizer is applied to the fault detection model. The Adam optimizer is a gradient-based optimization algorithm that can effectively optimize weight parameters and bias parameters in the steel pile identification model.
[0061] Optimizing parameters is mainly achieved through the following four key steps:
[0062] 1. Momentum mechanism (first-order moment estimation): The Adam optimizer tracks the gradient momentum of each parameter, i.e. its first-order moment estimation. This step helps to accelerate the optimization trajectory, especially when facing many flat areas, it can effectively make the parameters approach the optimal solution more quickly.
[0063] 2. Weighted average of squared gradients (second-order moment estimate): The Adam optimizer also calculates the exponentially weighted average of the squared gradients of each parameter, i.e., its second-order moment estimate. This step enables the optimizer to dynamically adjust the learning rate of each parameter based on the size of the gradient, ensuring that a more cautious update step is taken in the direction of a larger gradient, and a larger update step is taken in the direction of a smaller gradient, thereby improving the efficiency of the optimization.
[0064] 3. Adaptive learning rate adjustment: Based on the results of first-order and second-order moment estimation, the Adam optimizer can independently calculate an adaptive learning rate for each parameter. This mechanism ensures that the learning rate can be dynamically adjusted according to historical gradient information, thereby ensuring the stability and efficiency of the update process.
[0065] 4. Bias correction: In the initial stage of training, since the initial values of the first-order and second-order moment estimates may be small, the Adam optimizer will apply bias correction techniques to eliminate this effect. This step ensures that the model can be updated with an appropriate step size at the beginning of training.
[0066] In steel surface defect detection, the Adam optimizer shows the following advantages:
[0067] 1. Adaptive learning rate adjustment: The Adam optimizer has the ability to dynamically adjust the learning rate, which is particularly suitable for complex steel surface defect detection tasks and can significantly improve the accuracy and reliability of detection.
[0068] 2. Efficient convergence characteristics: By combining the weighted average of momentum and squared gradient, the Adam optimizer can converge quickly in the early stages of training, thereby improving training efficiency and detection accuracy.
[0069] 3. Bias correction technology: The Adam optimizer uses bias correction technology to effectively reduce the bias in the early stages of training, improve training efficiency and accuracy, and make it more suitable for handling complex and changeable training tasks.
[0070] 4. Large-scale data processing capabilities: The Adam optimizer can efficiently process large-scale data sets, speed up training, reduce the risk of overfitting, and improve the generalization performance of the model.
[0071] 5. High stability: In variable gradients and complex loss spaces, the Adam optimizer shows higher stability and reliability by combining momentum and adaptive adjustment mechanisms.
[0072] According to another embodiment of the present invention, the device comprises a processor and a memory, wherein the processor is electrically connected to the memory, and the memory is provided with computer instructions, and the steel pile identification method is implemented when the computer instructions are executed by the processor.
[0073] A computer-readable storage medium according to another embodiment of the present invention includes computer instructions, and when the computer instructions are executed, the steel pile identification method is implemented.
[0074] The foregoing description of specific exemplary embodiments of the present invention is for the purpose of illustration and demonstration. These descriptions are not intended to limit the present invention to the precise form disclosed, and it is clear that many changes and variations can be made based on the above teachings. The purpose of selecting and describing the exemplary embodiments is to explain the specific principles of the present invention and its practical application, so that those skilled in the art can realize and utilize various different exemplary embodiments of the present invention and various different selections and changes. The scope of the present invention is intended to be limited by the claims and their equivalents.
Claims
1. A steel pile identification method, characterized in that: The steps include: Building a data set based on the steel pile image data, wherein the data set includes a training set and a validation set; The training set and the validation set are used in sequence to train and test the steel pile recognition model; wherein the steel pile recognition model is constructed by GRU, and feature extraction is performed based on CNN; The steel pile image to be inspected is input into the trained steel pile recognition model, and the steel surface defect detection results are output.
2. The steel pile identification method according to claim 1, characterized in that: The GRU processes the data information, wherein: The expression of the forget gate is as follows: z t =g(W (x) x t +U (x) h t-1 ) The expression of the update gate is as follows: r t =g(W (N) x t +U (r) h t-1 ) The expression of the current memory content is as follows: h t =tanh(Wx t +r t )U t-1 ) The expression of the final memory of the previous time is as follows: h t =z t ⊙h t-1 +(1-z t )⊙h′ i Among them, W (x) , W (N) , U (x) , U (r) , W and U t-1 Represents different weight matrices, g() represents the sigmoid function, tanh represents the hyperbolic tangent function, x t Indicates the input at the current moment, h t-1 represents the hidden state at the previous moment, and ⊙ represents the dot product.
3. The steel pile identification method according to claim 1, characterized in that: The loss function of the steel pile identification model is as follows: Where N is the length of the input sequence, is the predicted value of the model, y i is the true value.
4. The steel pile identification method according to claim 1, characterized in that: The Adam optimizer is applied to the steel pile identification model.
5. The steel pile identification method according to claim 1, characterized in that: The CNN is 1DCNN.
6. The steel pile identification method according to claim 5, characterized in that: The 1DCNN includes an input layer, several convolutional layers, a pooling layer, a fully connected layer and an output layer.
7. A device, characterized in that: It comprises a processor and a memory communicatively connected to the processor, wherein the memory stores computer instructions executable by the processor, and when the computer instructions are executed, the steel pile identification method as described in any one of claims 1 to 6 is implemented.
8. A computer-readable storage medium storing computer instructions, characterized in that: When the computer instructions are executed, the steel pile identification method as described in any one of claims 1 to 6 is implemented.