Method and system for detecting surface roughness of steel body based on millimeter wave radar
Through the detection method based on millimeter wave radar, a steel surface roughness detection model is established using CNN-LSTM regression neural network, which solves the cost, complexity and damage problems of steel surface detection in the prior art, and achieves high efficiency and automation , non-damage steel surface roughness detection.
Patent Information
- Application Number
- CN202410956952.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-17
- Publication Date
- 2024-11-08
AI Technical Summary
Among the existing steel surface roughness detection methods, traditional contact methods are susceptible to human factors and may damage the surface. However, non-contact methods have high equipment costs and complexity, making it difficult to achieve low-cost and easy-to-operate automated detection.
Using a detection method based on millimeter wave radar, the millimeter wave radar echo signals under different roughness of the steel surface are collected, preprocessed and feature vector extraction are performed, and inputted into the CNN-LSTM regression neural network to establish the roughness of the steel surface. The degree detection model realizes contactless and automated roughness detection.
It realizes accurate detection of steel surface roughness, avoids damage to the surface, reduces detection costs, improves detection efficiency and adaptability, and is suitable for complex industrial environments.
Smart Images

Figure CN118913164A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of steel surface roughness detection, and in particular to a steel surface roughness detection method and system based on millimeter wave radar. Background Art
[0002] The detection of steel surface roughness is crucial for industrial manufacturing and quality control. The current detection methods are mainly divided into two categories: traditional contact and modern non-contact. Traditional contact methods include comparison method, impression method and stylus method. The comparison method relies on human subjective judgment and is easily affected by the operator's skill level and subjective factors, so it is not conducive to automation and standardization. The impression method usually requires leaving an impression on the surface of the steel body or using a specific impression material, which may damage the surface or affect its subsequent performance. The stylus method measures the slight height difference of the surface through a mechanical stylus. Although it can provide accurate data, the measurement time is long and there is a possibility of damage to the surface.
[0003] In contrast, modern non-contact methods show broader application prospects. Methods such as speckle contrast method, real-time holography method and light section method use laser technology or optical principles to accurately measure the roughness of steel without contacting the surface. Although these methods have high accuracy, they also face challenges such as high equipment cost, complex technology and sensitivity to ambient light interference.
[0004] Therefore, in view of the current technical situation, it is particularly urgent and important to develop a low-cost, easy-to-operate and automated non-contact steel surface roughness detection method. Summary of the invention
[0005] In order to solve the above problems, the purpose of the present invention is to provide a method and system for detecting the surface roughness of a steel body based on millimeter wave radar.
[0006] A method for detecting the surface roughness of a steel body based on millimeter wave radar, comprising:
[0007] Collect millimeter-wave radar echo signals under different roughness of steel surface;
[0008] Preprocessing the millimeter-wave radar echo signal to obtain a three-dimensional array; the three-dimensional array includes a millimeter-wave radar sampling point for each frame and each chirp;
[0009] Calculate the modulus values of all millimeter-wave radar sampling points in the first frame and the first chirp, and use the modulus values of the millimeter-wave radar sampling points as feature vectors;
[0010] The feature vector is input as a sample into the CNN-LSTM recurrent neural network for training to obtain a steel surface roughness detection model;
[0011] The steel body surface roughness detection model is used to complete the detection of the target steel body surface roughness.
[0012] Preferably, the step of inputting the feature vector as a sample into a CNN-LSTM recurrent neural network for training to obtain a steel surface roughness detection model comprises:
[0013] Normalizing the feature vector to obtain a normalized feature vector;
[0014] The normalized feature vector is input into the CNN-LSTM recurrent neural network for training to obtain a steel surface roughness detection model; the CNN-LSTM recurrent neural network includes the first CNN convolutional layer, the second CNN convolutional layer, the LSTM layer, the Dropout layer, the activation layer, the output layer, and the regression layer. The first CNN convolutional layer includes 10 filters with a filter length of 3, and the second CNN convolutional layer includes 15 filters with a filter length of 3; the Dropout layer sets 20% of the neurons to be randomly inactivated during training; the output layer includes one neuron for generating prediction results; and the regression layer is used to calculate the loss between the prediction results and the true results.
[0015] Preferably, in the steel surface roughness detection model training process, the mean square error is selected as the loss function, and the loss function is:
[0016]
[0017] Among them, y i is the true value of the i-th sample, is the predicted value of the steel surface roughness detection model for the i-th sample, and n is the number of samples.
[0018] Preferably, the use of the steel body surface roughness detection model to complete the detection of the target steel body surface roughness includes:
[0019] Acquire the millimeter-wave radar echo signal of the target to be measured, and extract the feature vector of the millimeter-wave radar echo signal of the target to be measured;
[0020] The extracted feature vector is compared with the feature vector of the steel body surface to obtain the similarity between the measured target and the steel body;
[0021] When the similarity is less than the threshold, "non-steel surface" is output; when the similarity is greater than or equal to the threshold, the target is judged to be a steel surface, and the steel surface roughness detection model is used to complete the prediction of the steel surface roughness and output it.
[0022] Preferably, the similarity calculation formula is:
[0023]
[0024] Where A is the eigenvector of the target, B is the eigenvector of the steel surface, ||A|| represents the modulus of the eigenvector of the target, and ||B|| represents the modulus of the eigenvector of the steel surface.
[0025] The present invention also provides a steel surface roughness detection system based on millimeter wave radar, comprising:
[0026] The echo signal acquisition module is used to collect millimeter wave radar echo signals under different roughness of steel surface;
[0027] A signal preprocessing module, used for preprocessing the millimeter wave radar echo signal to obtain a three-dimensional array; the three-dimensional array contains the millimeter wave radar sampling points of each frame and each chirp;
[0028] A feature vector calculation module is used to calculate the modulus values of all millimeter-wave radar sampling points in the first frame and the first chirp, and use the modulus values of the millimeter-wave radar sampling points as feature vectors;
[0029] A neural network training module is used to input the feature vector as a sample into a CNN-LSTM recurrent neural network for training to obtain a steel surface roughness detection model;
[0030] The roughness detection module is used to detect the surface roughness of the target steel body using the steel body surface roughness detection model.
[0031] The present invention also provides an electronic device, comprising a bus, a transceiver, a memory, a processor, and a computer program stored in and run on the memory, wherein the transceiver, the memory and the processor are connected via the bus, and is characterized in that when the computer program is executed by the processor, the steps in the above-mentioned method for detecting the surface roughness of a steel body based on millimeter-wave radar are implemented.
[0032] The present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps in the above-mentioned method for detecting the surface roughness of a steel body based on millimeter-wave radar are implemented.
[0033] The beneficial effect of the steel surface roughness detection method and system based on millimeter wave radar provided by the present invention is that: compared with the prior art, the present invention obtains a steel surface roughness detection model by extracting feature vectors from millimeter wave radar echo signals of different steel surface roughnesses, and making them into samples to input into a deep learning regression model for training. The model can be used to accurately predict the roughness of different steel surfaces being tested. Compared with traditional contact and non-contact steel surface roughness detection methods, it not only improves the detection efficiency, but also reduces interference and damage to the steel surface. It has significant advantages in real-time, non-destructiveness, ease of operation, accuracy, and adaptability to complex industrial environments.
[0034] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0036] Figure 1 A flow chart of a method for detecting steel surface roughness based on millimeter wave radar provided in an embodiment of the present invention is shown;
[0037] Figure 2 A schematic diagram of a steel surface roughness detection system based on millimeter wave radar provided in an embodiment of the present invention is shown. DETAILED DESCRIPTION
[0038] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.
[0039] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.
[0040] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0041] See also Figure 1-2 , a steel surface roughness detection method based on millimeter wave radar, comprising:
[0042] Step 1: Collect millimeter-wave radar echo signals under different roughness of steel surface;
[0043] In practical applications, the present invention can use TI millimeter wave radar "IWR1843BOOST" and "DCA1000EVM" to collect radar echo signals under different roughness of steel surface.
[0044] Step 2: preprocessing the millimeter-wave radar echo signal to obtain a three-dimensional array; the three-dimensional array contains the millimeter-wave radar sampling points of each frame and each chirp;
[0045] The present invention forms a three-dimensional array by performing complex number processing, reshaping and rearranging of the millimeter wave radar echo signal, wherein the three-dimensional array contains millimeter wave radar data organized according to each frame, each chirp and sampling point, and the raw data of the millimeter wave radar is stored in the form of a binary (.bin) file, each file containing real and imaginary data. These real and imaginary data (every two adjacent data are real and imaginary) need to be combined in a complex form during processing, and then further operated.
[0046] Step 3: Calculate the modulus values of all millimeter-wave radar sampling points in the first frame and the first chirp, and use the modulus values of the millimeter-wave radar sampling points as feature vectors;
[0047] Step 4: Input the feature vector as a sample into the CNN-LSTM recurrent neural network for training to obtain a steel surface roughness detection model;
[0048] Furthermore, step 4 includes:
[0049] Step 4.1: normalizing the feature vector to obtain a normalized feature vector;
[0050] Step 4.2: The normalized feature vector is input into the CNN-LSTM recurrent neural network for training to obtain a steel surface roughness detection model; wherein the CNN-LSTM recurrent neural network includes the first CNN convolutional layer, the second CNN convolutional layer, the LSTM layer, the Dropout layer, the activation layer, the output layer, and the regression layer, and the first CNN convolutional layer includes 10 filters with a filter length of 3, and the second CNN convolutional layer includes 15 filters with a filter length of 3; the Dropout layer sets 20% of the neurons to be randomly inactivated during training; the LSTM layer specifies the number of 256 units; the output layer contains one neuron for generating prediction results; and the regression layer is used to calculate the loss between the prediction results and the true results.
[0051] In the steel surface roughness detection model training process, the present invention uses mean square error as the loss function, and the loss function is:
[0052]
[0053] Among them, y i is the true value of the i-th sample, is the predicted value of the steel surface roughness detection model for the i-th sample, and n is the number of samples.
[0054] Step 5: Use the steel body surface roughness detection model to complete the detection of the target steel body surface roughness.
[0055] It should be noted that before the steel surface roughness test is performed, it is necessary to determine whether the current test target is a steel surface; only when the target is a steel surface, the corresponding roughness is output;
[0056] Furthermore, step 5 includes:
[0057] Step 5.1: Obtain the millimeter-wave radar echo signal of the target to be measured, and extract the feature vector of the millimeter-wave radar echo signal of the target to be measured;
[0058] Step 5.2: Compare the extracted feature vector with the feature vector of the steel body surface to obtain the similarity between the measured target and the steel body;
[0059] The similarity calculation formula is:
[0060]
[0061] Where A is the eigenvector of the target, B is the eigenvector of the steel surface, ||A|| represents the modulus of the eigenvector of the target, and ||B|| represents the modulus of the eigenvector of the steel surface.
[0062] Step 5.3: When the similarity is less than the threshold, "non-steel surface" is output; when the similarity is greater than or equal to the threshold, the target is judged to be a steel surface, and the steel surface roughness detection model is used to complete the prediction of the steel surface roughness and output it.
[0063] The present invention further describes the above-mentioned steel surface roughness detection method based on millimeter wave radar in combination with specific embodiments below:
[0064] Step 1. An experimental plan was developed based on the actual situation. The plan includes roughness comparison samples made using different processing methods (vertical milling, flat milling, planing). The material is made of 45 high-quality carbon structural steel. The specifications of the three processing methods are 0.8 microns, 1.6 microns, 3.2 microns, and 6.3 microns, all of which meet the national implementation standard GB / T6060.2-2006.
[0065] Step 2: Use TI millimeter wave radar "IWR1843BOOST" and "DCA1000EVM" sensors to transmit "FMCW frequency modulated continuous wave" signals to steel surfaces of different roughness, and receive the echo signal reflected from the steel surface. The echo signal is processed by the mixer to generate a lower frequency and easier to process intermediate frequency signal, so as to perform digital signal processing on the intermediate frequency signal later. On the received signal, each chirp signal is discretely sampled, and the discretely sampled data is converted into radar data, and a three-dimensional array is formed through complex number processing, reshaping and rearrangement. And select all the sampling points in the first frame and the first chirp, and use the modulus of their complex data as feature vectors. It should be noted that the TI millimeter wave radar sensor system consists of a transmitting system, a receiving system, and a signal processing system.
[0066] Furthermore, step 2 includes:
[0067] (1) The received millimeter-wave radar echo signal is mixed, and the transmitted signal and the received signal are passed through a mixer, and the two signals are multiplied to produce an intermediate frequency (IF) signal with a lower frequency and easier to process.
[0068] The mathematical expression of the intermediate frequency signal is:
[0069]
[0070]
[0071] Among them, A is the amplitude of the intermediate frequency signal, S is the gain factor, f is the intermediate frequency, is the intermediate frequency phase, τ is the round trip time of the transmitted signal, c is the speed of light, and d is the distance between the target and the radar.
[0072] (2) Using the millimeter wave radar system "IWR1843BOOST" and the "DCA1000EVM" sensor system, the intermediate frequency signal is transmitted to the PC via Ethernet in the form of a 16-bit complex ".bin" file. On the PC, these files are parsed in a fixed format using Matlab software to obtain the sampled data of each frame and each chirp.
[0073] (3) Process the intermediate frequency signal and generate a three-dimensional array through complex complex number processing, data reshaping and rearrangement.
[0074] (4) For the obtained three-dimensional array, it is necessary to calculate the modulus values of all sampling points of the first frame and the first chirp.
[0075] (5) Extract the sampling points after modulus calculation and input them into the neural network as feature vectors.
[0076] Step 3: Construct a CNN-LSTM recurrent neural network, and select all sampling points in the first frame and the first chirp in the obtained three-dimensional array. After taking the modulus of the complex data, input it into the neural network as a feature vector for training to obtain a steel surface roughness detection model.
[0077] Furthermore, step three includes:
[0078] (1) Data of different roughness (0.8 μm, 1.6 μm, 3.2 μm, and 6.3 μm) of calibration samples under three processing methods (vertical milling, flat milling, and planing) were collected, and their feature vectors were created as a data set.
[0079] (2) Use CNN-LSTM recurrent neural network for training. First, the initial feature extraction is performed through two convolutional layers: the first convolutional layer contains 10 convolutional kernels of length 3, and the second convolutional layer contains 15 convolutional kernels of the same length. The features processed by the convolutional layer are passed to an LSTM layer, which contains 256 hidden units and only outputs the hidden state of the last time step as the prediction task. Next, the Dropout layer is applied, randomly discarding 20% of the neurons to reduce the risk of overfitting. After that, the output of the previous layer is passed through a fully connected layer (128 nodes) and processed by the ReLU (rectified linear unit) activation function layer. The final prediction result is generated through the output layer (1 neuron, performing a single value regression task). Finally, the loss between the predicted result and the true result is calculated through the regression layer. The ReLU activation function formula is as follows:
[0080] f(x)=max(0,x);
[0081] Among them, x represents the input and f(x) represents the output. That is, when the input is greater than 0, the output is equal to the input; when the input is less than or equal to 0, the output is 0. This nonlinear activation function has a simple calculation method and effectively propagates the gradient, which helps to reduce the problem of gradient disappearance.
[0082] Among them, the CNN-LSTM recurrent neural network combines the convolutional neural network (CNN) and the long-term memory network (LSTM), and shows significant advantages in processing the extracted feature vectors. The CNN layer focuses on extracting local features and is particularly suitable for processing spectral data in signals. It can effectively capture specific patterns at different frequencies or positions. These local features are of great significance for prediction tasks. At the same time, the LSTM layer is responsible for capturing the sequential and long-term dependencies in the data, providing key sequence modeling capabilities and improving the model's ability to understand the complex dynamics behind the data.
[0083] In the CNN-LSTM recurrent neural network, CNN extracts simple but critical local features from data at a low level, while LSTM processes data sequences through memory units at a higher level, enabling the model to learn and understand input data at different levels of abstraction. This multi-level feature learning and sequence modeling capability enables the steel surface roughness detection model to effectively handle complex nonlinear relationships and improves the model's comprehensive understanding and characterization of data features.
[0084] In the process of steel surface roughness detection model training, mean square error (MSE) is selected as the loss function, which has a larger penalty for larger error values, making the model more robust during the training process and less susceptible to outliers. By minimizing MSE, the model is more inclined to reduce prediction errors, thereby improving the overall prediction accuracy and model performance, being able to better fit the data and having better generalization ability.
[0085] Mean square error (MSE) expression:
[0086]
[0087] Among them, yi is the true value, is the model prediction value, and n is the number of samples. The smaller the MSE value, the smaller the difference between the model's prediction result and the true value, and the stronger the model's fitting ability.
[0088] The mean absolute error (MAE) and the coefficient of determination R2 were used as evaluation indicators to evaluate the performance of the CNN-LSTM regression neural network model.
[0089] Step 4: Use deep learning to learn the feature vectors corresponding to the echo signals of steel surfaces with different roughness (0.8 microns, 1.6 microns, 3.2 microns, and 6.3 microns). When the target to be measured is a steel surface, predict its corresponding roughness.
[0090] Furthermore, step four includes:
[0091] (1) Before predicting the surface roughness of a steel body, it is necessary to first determine the target to be measured. Only when it is confirmed that the target to be measured is a steel surface, the roughness prediction is performed. If the judgment result is not a steel surface, the output is "non-steel surface".
[0092] Among them, the original signal of the millimeter-wave radar collected by the current measured target is extracted for feature vector, and the extracted feature vector is compared with the feature vector of the steel surface for similarity. The similarity calculation adopts cosine similarity, and the cosine similarity calculation formula is:
[0093]
[0094] Among them, A and B are two vectors, ||A|| and ||B|| represent the modulus (length) of A and B respectively;
[0095] (2) According to the set threshold, when the cosine similarity is less than the threshold, the system outputs "non-steel surface". When the cosine similarity is greater than or equal to the threshold, it is judged as a steel surface, and the surface roughness is predicted and output.
[0096] Compared with the prior art, the present invention has the following advantages:
[0097] (1) Non-contact prediction of steel surface roughness without damage: As a non-contact technology, millimeter wave radar does not need to be in direct contact with the surface being measured, avoiding any physical damage or deformation that may be caused to the steel surface, maintaining its integrity and original state. It also does not need to apply any marks on the steel surface or prepare special surface materials, making the operation more convenient and efficient.
[0098] (2) High-precision measurement and strong adaptability: Millimeter-wave radar can accurately detect tiny changes and fine structures on the surface of steel, can capture roughness features at the micron level, and can accurately predict the surface roughness of steel. Its electromagnetic waves in the millimeter-wave frequency band can effectively penetrate air, water mist, and light dust under most environmental conditions, and are not affected by light. It is suitable for roughness detection needs in various industrial environments and complex conditions.
[0099] (3) Real-time and high efficiency: It can quickly scan and acquire data in real time, which is suitable for application scenarios of fast feedback and real-time detection, and realizes surface roughness detection on the production line.
[0100] (4) Advanced deep learning regression model: By combining advanced deep learning technology, the millimeter wave radar data features can be deeply mined and analyzed. The model can autonomously learn signal patterns of different intensities, capture subtle changes and details, and accurately predict the surface roughness information of steel bodies.
[0101] (5) The present invention has significant advantages such as real-time monitoring, adaptability to complex industrial environments, and the combination of deep learning technology to achieve non-contact and non-destructive prediction of steel surface roughness. These advantages make millimeter wave radar an efficient and reliable method for detecting different roughness of steel surfaces.
[0102] The present invention also provides a steel surface roughness detection system based on millimeter wave radar, comprising:
[0103] The echo signal acquisition module is used to collect millimeter wave radar echo signals under different roughness of steel surface;
[0104] A signal preprocessing module, used for preprocessing the millimeter wave radar echo signal to obtain a three-dimensional array; the three-dimensional array contains the millimeter wave radar sampling points of each frame and each chirp;
[0105] A feature vector calculation module is used to calculate the modulus values of all millimeter-wave radar sampling points in the first frame and the first chirp, and use the modulus values of the millimeter-wave radar sampling points as feature vectors;
[0106] A neural network training module is used to input the feature vector as a sample into a CNN-LSTM recurrent neural network for training to obtain a steel surface roughness detection model;
[0107] The roughness detection module is used to detect the surface roughness of the target steel body using the steel body surface roughness detection model.
[0108] Compared with the prior art, the beneficial effects of the steel surface roughness detection system based on millimeter wave radar provided by the present invention are the same as the beneficial effects of the steel surface roughness detection method based on millimeter wave radar described in the above technical solution, which will not be repeated here.
[0109] The present invention also provides an electronic device, comprising a bus, a transceiver, a memory, a processor, and a computer program stored in the memory and executable on the memory, wherein the transceiver, the memory, and the processor are connected via the bus, and is characterized in that when the computer program is executed by the processor, the steps in the above-mentioned method for detecting surface roughness of a steel body based on millimeter-wave radar are implemented. Compared with the prior art, the beneficial effects of the electronic device provided by the present invention are the same as the beneficial effects of the method for detecting surface roughness of a steel body based on millimeter-wave radar described in the above-mentioned technical solution, and are not elaborated herein.
[0110] The present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps in the above-mentioned method for detecting surface roughness of a steel body based on millimeter-wave radar are implemented. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present invention are the same as the beneficial effects of the method for detecting surface roughness of a steel body based on millimeter-wave radar described in the above-mentioned technical solution, which will not be repeated here.
[0111] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technical solution that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.
Claims
1. A method for detecting steel surface roughness based on millimeter wave radar, characterized in that: include: Collect millimeter-wave radar echo signals under different roughness of steel surface; Preprocessing the millimeter wave radar echo signal to obtain a three-dimensional array; The three-dimensional array contains the millimeter-wave radar sampling points of each frame and each chirp; Calculate the modulus values of all millimeter-wave radar sampling points in the first frame and the first chirp, and use the modulus values of the millimeter-wave radar sampling points as feature vectors; The feature vector is input as a sample into the CNN-LSTM recurrent neural network for training to obtain a steel surface roughness detection model; The steel body surface roughness detection model is used to complete the detection of the target steel body surface roughness.
2. The method for detecting steel surface roughness based on millimeter wave radar according to claim 1, characterized in that: The feature vector is input as a sample into the CNN-LSTM recurrent neural network for training to obtain a steel surface roughness detection model, including: Normalizing the feature vector to obtain a normalized feature vector; The normalized feature vector is input into the CNN-LSTM recurrent neural network for training to obtain a steel surface roughness detection model; the CNN-LSTM recurrent neural network includes the first CNN convolutional layer, the second CNN convolutional layer, the LSTM layer, the Dropout layer, the activation layer, the output layer, and the regression layer. The first CNN convolutional layer includes 10 filters with a filter length of 3, and the second CNN convolutional layer includes 15 filters with a filter length of 3; the Dropout layer sets 20% of the neurons to be randomly inactivated during training; the output layer includes one neuron for generating prediction results; and the regression layer is used to calculate the loss between the prediction results and the true results.
3. The method for detecting steel surface roughness based on millimeter wave radar according to claim 2, characterized in that: In the process of steel surface roughness detection model training, the mean square error is selected as the loss function, and the loss function is: Among them, y i is the true value of the i-th sample, is the predicted value of the steel surface roughness detection model for the i-th sample, and n is the number of samples.
4. The method for detecting steel surface roughness based on millimeter wave radar according to claim 3, characterized in that: The method of using the steel body surface roughness detection model to complete the detection of the surface roughness of the target steel body includes: Acquire the millimeter-wave radar echo signal of the target to be measured, and extract the feature vector of the millimeter-wave radar echo signal of the target to be measured; The extracted feature vector is compared with the feature vector of the steel body surface to obtain the similarity between the measured target and the steel body; When the similarity is less than the threshold, "non-steel surface" is output; when the similarity is greater than or equal to the threshold, the target is judged to be a steel surface, and the steel surface roughness detection model is used to complete the prediction of the steel surface roughness and output it.
5. The method for detecting steel surface roughness based on millimeter wave radar according to claim 4, characterized in that: The similarity calculation formula is: Where A is the eigenvector of the target, B is the eigenvector of the steel surface, ||A|| represents the modulus of the eigenvector of the target, and ||B|| represents the modulus of the eigenvector of the steel surface.
6. A steel surface roughness detection system based on millimeter wave radar, characterized in that: include: The echo signal acquisition module is used to collect millimeter wave radar echo signals under different roughness of steel surface; A signal preprocessing module, used for preprocessing the millimeter wave radar echo signal to obtain a three-dimensional array; the three-dimensional array contains the millimeter wave radar sampling points of each frame and each chirp; A feature vector calculation module is used to calculate the modulus values of all millimeter-wave radar sampling points in the first frame and the first chirp, and use the modulus values of the millimeter-wave radar sampling points as feature vectors; A neural network training module is used to input the feature vector as a sample into a CNN-LSTM recurrent neural network for training to obtain a steel surface roughness detection model; The roughness detection module is used to detect the surface roughness of the target steel body using the steel body surface roughness detection model.
7. An electronic device comprising a bus, a transceiver, a memory, a processor, and a computer program stored in the memory and executable on the memory, wherein the transceiver, the memory, and the processor are connected via the bus, wherein: When the computer program is executed by the processor, the steps in the method for detecting the surface roughness of a steel body based on millimeter-wave radar as described in any one of claims 1 to 5 are implemented.
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 steps in the method for detecting surface roughness of a steel body based on millimeter wave radar as described in any one of claims 1 to 8 are implemented.
