Improved ResNet50V2-based dual-laser micro-displacement distance measuring device and micro-displacement distance measuring method
Through improved ResNet50V2 model and dual laser dataset training, combined with optical device adjustment and depth separation convolution, the problem of low accuracy and efficiency of deep learning models in micro-displacement ranging is solved, achieving higher measurement accuracy and adaptation sex.
Patent Information
- Application Number
- CN202411416339.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-11
- Publication Date
- 2025-01-14
AI Technical Summary
Existing deep learning models have low measurement accuracy and low prediction efficiency in micro-displacement ranging, especially when facing changes in environmental conditions, and the large amount of training parameters leads to a long training time.
The improved ResNet50V2 model is used to combine red and green laser dataset training, and the optical equipment is adjusted through lifting and horizontal displacement adjustment mechanisms. The dual lasers are used to avoid wavelength limitations. Combined with depth separation convolution and feature extraction optimization, reducing the amount of parameters to improve model adaptability and efficiency.
It improves the accuracy and efficiency of micro-displacement measurement, enhances the model's adaptability in different environments, reduces the computational complexity and training time, and achieves higher measurement accuracy and stability.
Smart Images
Figure CN119309496A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of micro-displacement ranging, and relates to a device and a ranging method for improving the micro-displacement prediction accuracy and efficiency, and specifically to a device and a method for training a captured red and green laser data set and predicting the micro-displacement distance of a measured object by improving a ResNet50V2 deep learning model. Background Art
[0002] In the cutting-edge field of modern manufacturing, micro-displacement ranging technology is a key link in achieving high-precision displacement control. Taking wafer processing in the semiconductor manufacturing field as an example, when a wafer is transferred from one processing equipment to another, the position and angle of the wafer need to be monitored to ensure that it does not tilt or shift during the transmission process.
[0003] At present, micro-displacement ranging methods are generally divided into two categories: contact and non-contact. The contact ranging method usually measures the displacement change by directly contacting the measured object with a mechanical device. Although this method is simple and easy to use, it is easy to cause physical damage to the measured object, and the measurement accuracy is limited by the stability of the mechanical device. In addition, the contact method cannot realize real-time monitoring of small displacement changes, which limits its effect in high-precision applications. Non-contact micro-displacement ranging methods mainly include optical methods, laser methods, and electromagnetic methods. Among them, the laser method has become an important representative of non-contact micro-displacement ranging technology due to its high precision and high resolution. The laser displacement sensor realizes the accurate measurement of small displacement by emitting a laser beam and measuring its reflection time. The laser method has the advantages of high measurement accuracy and fast response speed, and is suitable for various complex working conditions. However, the laser displacement sensor is expensive, sensitive to changes in ambient light, and requires strict working conditions, which limits its universality in practical applications.
[0004] Compared with the traditional micro-displacement ranging method, the visual measurement system uses a deep learning model for micro-displacement ranging, which has lower costs. It is usually composed of a camera, an optical lens, and image processing software. Its hardware equipment (such as a camera) is relatively cheap, especially after the popularization of industrial-grade cameras, and has a high cost-effectiveness. In contrast, the hardware cost of equipment such as laser ranging systems and optical interferometers is relatively high, including lasers, optical components, and precise mechanical structures. Moreover, the visual measurement system has strong adaptability and can work flexibly under different viewing angles, lighting conditions, and complex environments. By adjusting the camera position and parameters, it can easily cope with a variety of different application scenarios. The deep learning model extracts features from a large amount of displacement image data and accurately predicts its displacement. By processing and analyzing image data through a convolutional neural network (CNN), it can achieve high-precision and high-stability displacement measurement in a complex environment. Therefore, this method has significant advantages in efficiency and accuracy.
[0005] However, deep learning still faces some challenges when performing micro-displacement ranging. Although deep learning models have the ability to automatically extract features, they are prone to measurement errors when facing changes in environmental conditions, because the model may not be able to effectively adapt to different measurement environments, and the number of calculated parameters during model training is generally large, and the training time is long. Therefore, how to improve the adaptability of network models in different environments and improve training efficiency has become the focus of current research.
[0006] In order to meet these challenges, some improvements have been made in the existing technology, such as the introduction of multi-scale feature extraction in network model design, optimization of image preprocessing algorithms, and the construction of enhanced data sets. These methods have improved the robustness of the model to illumination changes and background noise to a certain extent, but there is still room for improvement. How to further improve the accuracy and efficiency of deep learning models in micro-displacement ranging is still an urgent problem to be solved. Summary of the invention
[0007] The purpose of the present invention is to address the problems of low micro-displacement ranging accuracy and low prediction efficiency of the existing algorithm, and to propose a dual-laser micro-displacement ranging device and a micro-displacement ranging method based on an improved ResNet50V2, so as to further improve the visual micro-displacement measurement accuracy.
[0008] The improved ResNet50V2 dual laser micro-displacement ranging device provided in this application adopts the following technical solutions:
[0009] A dual laser micro-displacement ranging device based on an improved ResNet50V2, comprising a base plate; characterized in that the device also includes:
[0010] The lifting displacement adjustment mechanism is connected and fixed to the bottom plate and is used for adjusting the displacement of the optical device in the vertical direction;
[0011] A horizontal displacement adjustment mechanism, connected and fixed to the base plate, used for adjusting the displacement of the optical device in the horizontal direction;
[0012] A CCD camera assembly, which is composed of a CCD camera, a horizontal displacement adjustment mechanism and a lifting displacement adjustment mechanism, is placed on one side of the bottom plate as a whole and is used for collecting the light beam spot;
[0013] The tested sample block assembly is composed of the tested sample block and the horizontal displacement adjustment mechanism, and is placed on the other side of the bottom plate as a whole;
[0014] A beam splitter assembly, which is composed of a beam splitter and the lifting displacement adjustment mechanism, is placed as a whole between the CCD camera assembly and the sample block assembly to be tested, and is used for reflecting the laser beam;
[0015] The laser is arranged on the side of the beam splitter assembly and is used for emitting a laser beam.
[0016] Furthermore, the lifting displacement adjustment mechanism is composed of a lifting base, a lifting top plate and a first micrometer; the lifting top plate and the lifting base are slidably connected through a lifting slide groove and a lifting guide rail, a top rod is fixedly provided at the bottom of the lifting top plate, a fixed block is connected to one side of the lifting base, the first micrometer is clamped and fixed in the fixed block, and a V-shaped rotating rod is provided on the lifting base through a rotating shaft, one end of the rotating rod is in contact with the first micrometer, and the other end is in contact with the top rod.
[0017] By adopting the above technical solution, the first micrometer is adjusted to drive the rotating rod to form a small angle deflection around the rotating axis, driving the lifting top plate to form an up and down micro-displacement adjustment relative to the lifting base, thereby driving the lifting and lowering micro-displacement adjustment of the optical device on the top of the lifting top plate.
[0018] Furthermore, the horizontal displacement adjustment mechanism is composed of a horizontal base, a horizontal top plate and a second micrometer; the horizontal top plate and the horizontal base are slidably connected by a horizontal guide rail and a horizontal slide groove, a fixed block is connected to one side of the horizontal base, the second micrometer is fixedly clamped in the fixed block, a push rod is fixedly provided on one side of the horizontal top plate, and a locking screw is provided on the other side, and the push rod is in contact with the second micrometer.
[0019] By adopting the above technical solution, the second micrometer is adjusted to push the top rod and the horizontal top plate to form a micro-displacement movement in the horizontal direction, thereby realizing the micro-displacement adjustment of the optical device above the horizontal top plate.
[0020] Furthermore, the laser is composed of a red laser and a green laser, both lasers are connected and fixed to the base plate via a connecting bracket, the green laser is set at an angle of 45° to the spectroscope assembly, and the red laser is set to align with the sample block to be tested.
[0021] By adopting the above technical solution, the use of dual monochromatic lasers can avoid the micro-displacement information being limited by its wavelength, and richer micro-displacement information can be obtained, making the ranging accuracy higher; the green laser and the beam splitter assembly are set at a 45° angle to ensure that the reflected laser beam is in the center of the camera phase plane, which is suitable as a direct correlation relationship for ranging; the red laser, as an oblique light, has a refraction relationship with the camera lens, which is used to assist in determining the position of the measured sample relative to the camera focus.
[0022] Furthermore, the connecting bracket is composed of a base, a column, a sleeve, a connecting rod and a locking nut. The base is connected and fixed to the bottom plate, the column is fixed in the base, the connecting rod is sleeved on the column and fixed to it through a locking nut, the sleeve is inserted into the connecting rod and fixed to it through a locking nut, and the laser is fixed in the sleeve through the locking nut.
[0023] By adopting the above technical solution, the angle of the incident light beam can be easily adjusted by using the connecting bracket, so as to adjust the incident light beam to correctly incident on the sample block to be tested, so that the CCD camera can capture a clear image.
[0024] The dual laser micro-displacement ranging method based on improved ResNet50V2 provided in this application adopts the following technical solutions:
[0025] A dual laser micro-displacement ranging method based on an improved ResNet50V2, the steps are as follows:
[0026] S1. Use red and green lasers to collect original images and obtain data sets;
[0027] S2. Perform noise reduction on the collected data set;
[0028] S3. Divide the denoised data set into a training set, a validation set, and a test set;
[0029] S4. Use the improved ResNet50V2 to train the training set and obtain the weight file;
[0030] S5. Load the trained weight file and use the validation set to evaluate and adjust the improved ResNet50V2.
[0031] S6. Calculate the weighted coefficients according to the prediction results of the red and green laser validation sets, and fit the true values;
[0032] S7. Use the improved ResNet50V2 on the test set to obtain the prediction results of the red and green lasers, multiply them by their corresponding weighting coefficients, and finally weightedly predict the micro-displacement distance.
[0033] Furthermore, the specific steps of the original image acquisition in step S1 are as follows: after the power of the CCD camera is connected, the red laser is turned on and the green laser is turned off. The red laser beam is irradiated on the sample block to be tested, and a red spot appears on the sample block. At this time, the computer issues a shooting command to the CCD camera to obtain the initial image before the micro-displacement; then the micrometer is rotated one scale, and the workbench drives the sample block to be tested to displace 0.01mm. The computer captures the red spot image on the sample block again, and the process is repeated until 800 images are captured. The shooting process of the green laser is the same as above.
[0034] In summary, the present invention includes at least one of the following beneficial technical effects:
[0035] (1) The laser micro-displacement ranging device of the present invention has a simple structure. During the measurement process, only the object to be measured needs to be fine-tuned. The device is easy to operate and can accurately obtain high-resolution images of the micro-displacement of the object to be measured, thereby further improving the accuracy of visual measurement.
[0036] (2) The micro-displacement ranging device of the present invention uses red and green lasers to illuminate the surface of the object being measured at the same time, and the parameters in the image are richer.
[0037] (3) The improved ResNet50V2 model of the present invention has fewer parameters, and due to the improvement of Stem Block, the feature information extracted is richer, and the measurement accuracy is further improved compared with the original ResNet50V2. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 It is a schematic diagram of the first angle structure of the dual laser micro-displacement ranging device of the present invention.
[0039] Figure 2 It is a schematic diagram of the second angle structure of the dual laser micro-displacement ranging device of the present invention.
[0040] Figure 3 It is a schematic diagram of the three-dimensional structure of the lifting and horizontal displacement adjustment mechanism in the present invention.
[0041] Figure 4 It is a schematic diagram of the main structure of the lifting and horizontal displacement adjustment mechanism in the present invention.
[0042] Figure 5 It is a schematic diagram of the exploded structure of the lifting displacement adjustment mechanism in the present invention.
[0043] Figure 6 It is a schematic diagram of the exploded structure of the horizontal displacement adjustment mechanism in the present invention.
[0044] Figure 7 It is a schematic diagram of the connecting bracket structure in the present invention.
[0045] Figure 8 It is a schematic diagram of the process of the dual laser micro-displacement ranging device of the present invention.
[0046] Fig. 9 This is a schematic diagram of the improved ResNet50V2 model of the present invention.
[0047] Fig.10 It is a schematic diagram of the optical path of the dual laser micro-displacement ranging method of the present invention.
[0048] In the figure: bottom plate 1, CCD imaging component 2, CCD camera 2-1, spectroscope component 3, sample block component 4, laser 5, handle 6, connecting bracket 7, base 7-1, first locking nut 7-2, column 7-3, second locking nut 7-4, sleeve 7-5, third locking nut 7-6, fourth locking nut 7-7, connecting rod 7-8, lifting displacement adjustment mechanism 8, lifting base 8-1, push rod 8-2, rotating rod 8-3, rotating shaft 8-4, lifting top plate 8-5, first micrometer 8-6, fixed plate 8-7, lifting guide rail 8-8, lifting slide 8-9, horizontal displacement adjustment mechanism 9, horizontal base 9-1, second micrometer 9-2, horizontal guide rail 9-3, fixed block 9-4, horizontal slide 9-5, horizontal top plate 9-6, locking screw 9-7, push rod 9-8. DETAILED DESCRIPTION
[0049] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent forms of modifications to the present invention by those skilled in the art fall within the scope defined by the claims attached to this application.
[0050] Example 1
[0051] like Figure 1-3 As shown, a dual laser micro-displacement ranging device based on the improved ResNet50V2 is composed of a base plate 1, a lifting displacement adjustment mechanism 8 and a horizontal displacement adjustment mechanism 9 fixed to the base plate 1, a CCD imaging component 2 composed of a CCD camera 2-1, a horizontal displacement adjustment mechanism 9 and a lifting displacement adjustment mechanism 8, a sample block component 4 composed of a sample block and a horizontal displacement adjustment mechanism 9, a spectroscope component 3 composed of a spectroscope and a lifting displacement adjustment mechanism 8 and placed between the CCD imaging component 2 and the sample block component 3; and a laser 5 placed on the side of the spectroscope component 3 for emitting a laser beam.
[0052] In order to make the optical device on the top of the lifting top plate form lifting micro-displacement adjustment, the lifting displacement adjustment mechanism is used in this embodiment. Specifically, Figure 3-5 As shown, the lifting displacement adjustment mechanism 8 is composed of a lifting base 8-1, a lifting top plate 8-5 and a first micrometer 8-6; the lifting top plate 8-5 is slidably connected to the lifting base 8-1 through a lifting slide groove 8-9 and a lifting guide rail 8-8, a top rod 8-2 is fixedly provided at the bottom of the lifting top plate 8-5, a fixed block 8-7 is connected to one side of the lifting base 8-1, the first micrometer 8-6 is clamped and fixed in the fixed block 8-7, and a V-shaped rotating rod 8-3 is provided on the lifting base 8-1 through a rotating shaft 8-4, one end of the rotating rod 8-3 is in contact with the first micrometer 8-6, and the other end is in contact with the top rod 8-2.
[0053] In order to make the optical device above the horizontal top plate form a micro-displacement adjustment in the horizontal direction, a horizontal displacement adjustment mechanism is used in this embodiment. Specifically, Figure 6 As shown, the horizontal displacement adjustment mechanism 9 is composed of a horizontal base 9-1, a horizontal top plate 9-6 and a second micrometer 9-2; the horizontal top plate 9-6 and the horizontal base 9-1 are slidably connected by a horizontal guide rail 9-3 and a horizontal slide groove 9-5, a fixed block 9-4 is connected to one side of the horizontal base 9-1, and the second micrometer 9-2 is fixedly clamped in the fixed block 9-4, a push rod 9-8 is fixedly provided on one side of the horizontal top plate 9-6, and a locking screw 9-7 is provided on the other side, and the push rod 9-8 is in contact with the second micrometer 9-2.
[0054] In order to avoid the micro-displacement information being limited by its wavelength, more abundant micro-displacement information can be obtained, so that the ranging accuracy is higher; this embodiment is achieved by using two monochromatic lasers, specifically, Figure 1-2 As shown, the laser 5 is composed of a red laser 5-1 and a green laser 5-2. The green laser 5-2 is set at an angle of 45° with the spectroscope assembly 3. The red laser 5-2 is aimed at the sample block to be measured. The setting of the angle of the green laser can ensure that the reflected laser beam is in the center of the camera phase surface, which is suitable as a direct correlation relationship for ranging; the red laser, as an oblique light, has a refraction relationship with the camera lens, which is used to assist in determining the position of the sample block to be measured relative to the camera focus.
[0055] Example 2
[0056] In order to facilitate the adjustment of the incident light beam angle, the incident light beam is adjusted to correctly incident on the sample block to be tested, so that the CCD camera can capture a clear image. In this embodiment, the installation of the laser is completed by connecting the bracket, specifically, Figure 2 , Figure 7 As shown, the connecting bracket 7 is composed of a base 7-1, a column 7-3, a sleeve 7-5, a connecting rod 7-8 and a locking nut. The base 7-1 is connected and fixed to the base plate 1, the column 7-3 is fixedly set in the base 7-1, the connecting rod 7-8 is sleeved on the column 7-3 and fixed thereto by a locking nut, the sleeve 7-5 is inserted into the connecting rod 7-8 and fixed thereto by a locking nut, and the laser 5 is fixed in the sleeve 7-5 by a locking nut.
[0057] like Figure 8-10 As shown, based on the improved ResNet50V2 dual laser micro-displacement ranging device, the steps are as follows:
[0058] S1. Use red and green lasers to collect original images and obtain data sets;
[0059] S2. Perform noise reduction on the collected data set;
[0060] S3. Divide the denoised data set into a training set, a validation set, and a test set;
[0061] S4. Use the improved ResNet50V2 to train the training set and obtain the weight file;
[0062] S5. Load the trained weight file and use the validation set to evaluate and adjust the improved ResNet50V2.
[0063] S6. Calculate the weighted coefficients according to the prediction results of the red and green laser validation sets, and fit the true values;
[0064] S7. Use the improved ResNet50V2 on the test set to obtain the prediction results of the red and green lasers, multiply them by their corresponding weighting coefficients, and finally weightedly predict the micro-displacement distance.
[0065] Specifically, the operation method is as follows: first install the spectroscope on the bottom plate, the distance between the base of the spectroscope and the CCD camera is 14cm, the green laser is placed on the edge of the bottom plate, 10cm away from the base of the spectroscope, the red laser is 4cm to the right of the green laser, and the base of the sample to be tested is 8cm away from the base of the spectroscope. After connecting the power supply of the CCD camera, turn on the red laser 5-2, turn off the green laser 5-1, and irradiate the red laser beam on the sample to be tested. A red spot appears on the sample. At this time, the computer issues a shooting command to the CCD camera to obtain the initial image before micro-displacement; then rotate the micrometer by one scale, the workbench drives the sample to be tested to displace 0.01mm, and the computer takes the red spot image on the sample again. Repeat the process until 800 images are taken. The shooting process of the green laser is the same as above. After the installation is completed according to the above requirements, follow steps S1 to S7 to predict micro-displacement.
[0066] like Fig. 9 As shown, the principle of ResNet50V2 in this embodiment is as follows:
[0067] The single 7×7 convolution layer of the Stem Block is replaced with three 3×3 convolution layers. Because it is composed of only one 7×7 convolution layer, the receptive field is relatively large, which will cause the details and high-frequency information of the image to be lost during feature extraction. Moreover, the 7×7 convolution kernel needs to process a larger receptive field during calculation, which will increase the computational complexity and memory usage of the computer, especially when processing high-resolution images, resulting in slower model training and inference. To address the above problems, three 3×3 convolution layers are used to replace the original 7×7 convolution layer, because the 3×3 convolution kernel can capture local features and detail information more finely. By stacking multiple 3×3 convolution layers, the network can learn more complex feature representations. In terms of parameter quantity, the total number of parameters P of three 3×3 convolution layers is:
[0068] P = 3 × 3 × 3 × C in ×C out
[0069] Where C in Represents the number of input channels, C out Represents the number of output channels.
[0070] The total number of parameters P of a 7×7 convolutional layer is:
[0071] P=7×7×C in ×C out
[0072] It can be clearly seen that the number of parameters is reduced. In addition, multiple convolutional layers introduce multiple nonlinear activation functions, which improves the expressiveness of the model.
[0073] In the Backbone module, replace the 3×3 convolutional layers of the Bottleneck module in layer1 and layer2 with depthwise separable convolution. Depthwise separable convolution can be decomposed into depthwise convolution and pointwise convolution. Depthwise convolution performs independent convolution on each input channel, while pointwise convolution linearly combines the output channels of depthwise convolution through 1×1 convolution. This change can significantly reduce the computational complexity. For H×W×C in The total number of parameters for the standard convolution is:
[0074] P=H×W×C in ×C out ×k 2
[0075] In the formula, H represents the height of the image, W represents the width of the image, and C in Represents the number of input channels, C out Represents the number of output channels, k 2 Represents the convolution kernel size.
[0076] The total number of parameters for depthwise separable convolution is:
[0077] P=H×W×C in ×k 2 +H×W×C in ×C out
[0078] It can be seen that the total number of parameters of depthwise separable convolution is much smaller than that of ordinary convolution. Therefore, after using depthwise separable convolution instead of ordinary convolution, the memory usage will be further reduced and the model will be more lightweight.
[0079] After training the data set using the improved ResNet50V2 model, the micro-displacement distances of the red and green lasers were predicted on the validation set respectively, and the optimal weights of the red and green lasers were calculated using the least squares method based on the true values, preparing for the final prediction.
[0080] It can be concluded from Table 1 that compared with other mainstream algorithms, the improved ResNet50V2 algorithm has good performance in terms of parameter quantity Params, determination coefficient R 2 , mean square error (MSE) and maximum absolute error (MAE), all of which are in the leading position, reflecting the superiority of the method proposed in this paper.
[0081] Table 1
[0082]
Claims
1. A dual laser micro-displacement ranging device based on an improved ResNet50V2, comprising a base plate (1); characterized in that: The device also includes: A lifting displacement adjustment mechanism (8) is connected and fixed to the base plate (1) and is used for adjusting the displacement of the optical device in the vertical direction; A horizontal displacement adjustment mechanism (9) is connected and fixed to the base plate (1) and is used for adjusting the displacement of the optical device in the horizontal direction; A CCD imaging component (2), which is composed of a CCD camera (2-1), a horizontal displacement adjustment mechanism (9) and a lifting displacement adjustment mechanism (8), and is placed as a whole on one side of the bottom plate (1) and is used for collecting light beam spots; The tested sample block assembly (4) is composed of the tested sample block and the horizontal displacement adjustment mechanism (9), and is placed as a whole on the other side of the bottom plate (1); A beam splitter assembly (3), which is composed of a beam splitter and the lifting displacement adjustment mechanism (8), and is placed as a whole between the CCD imaging assembly (2) and the sample block assembly (3) to be tested, and is used for reflecting the laser beam; The laser (5) is placed on the side of the beam splitter assembly (3) and is used to emit a laser beam.
2. According to claim 1, a dual laser micro-displacement ranging device based on an improved ResNet50V2, characterized in that: The lifting displacement adjustment mechanism (8) is composed of a lifting base (8-1), a lifting top plate (8-5) and a first micrometer (8-6) connected together; the lifting top plate (8-5) and the lifting base (8-1) are slidably connected via a lifting slide groove (8-9) and a lifting guide rail (8-8); a top rod (8-2) is fixedly provided at the bottom of the lifting top plate (8-5); a fixed block (8-7) is connected to one side of the lifting base (8-1); the first micrometer (8-6) is clamped and fixed in the fixed block (8-7); a V-shaped rotating rod (8-3) is provided on the lifting base (8-1) via a rotating shaft (8-4); one end of the rotating rod (8-3) contacts the first micrometer (8-6) and the other end contacts the top rod (8-2).
3. The dual laser micro-displacement ranging device based on the improved ResNet50V2 according to claim 1 is characterized in that: The horizontal displacement adjustment mechanism (9) is composed of a horizontal base (9-1), a horizontal top plate (9-6) and a second micrometer (9-2) connected together; the horizontal top plate (9-6) and the horizontal base (9-1) are slidably connected via a horizontal guide rail (9-3) and a horizontal slide groove (9-5); a fixed block (9-4) is connected to one side of the horizontal base (9-1); the second micrometer (9-2) is fixedly clamped in the fixed block (9-4); a push rod (9-8) is fixedly provided on one side of the horizontal top plate (9-6); a locking screw (9-7) is provided on the other side; and the push rod (9-8) is in contact with the second micrometer (9-2).
4. The dual laser micro-displacement ranging device based on the improved ResNet50V2 according to claim 1 is characterized in that: The laser (5) is composed of a red laser (5-1) and a green laser (5-2), both lasers are connected and fixed to the base plate (1) via a connecting bracket (7), the green laser (5-2) and the beam splitter assembly (3) are arranged at an angle of 45°, and the red laser (5-2) is arranged to be aligned with the sample block to be tested.
5. The dual laser micro-displacement ranging device based on the improved ResNet50V2 according to claim 4 is characterized in that: The connecting bracket (7) is composed of a base (7-1), a column (7-3), a sleeve (7-5), a connecting rod (7-8) and a locking nut. The base (7-1) is connected and fixed to the bottom plate (1). The column (7-3) is fixedly arranged in the base (7-1). The connecting rod (7-8) is sleeved on the column (7-3) and fixed thereto by a locking nut. The sleeve (7-5) is inserted into the connecting rod (7-8) and fixed thereto by a locking nut. The laser (5) is fixed in the sleeve (7-5) by a locking nut.
6. A dual laser micro-displacement ranging method based on improved ResNet50V2, characterized in that: Using the dual laser micro-displacement ranging device according to any one of claims 1 to 5, the steps are as follows: S1. Use red and green lasers to collect original images and obtain data sets; S2. Perform noise reduction on the collected data set; S3. Divide the denoised data set into a training set, a validation set, and a test set; S4. Use the improved ResNet50V2 to train the training set and obtain the weight file; S5. Load the trained weight file and use the validation set to evaluate and adjust the improved ResNet50V2. S6. Calculate the weighted coefficients according to the prediction results of the red and green laser validation sets, and fit the true values; S7. Use the improved ResNet50V2 on the test set to obtain the prediction results of the red and green lasers, multiply them by their corresponding weighting coefficients, and finally weightedly predict the micro-displacement distance.
7. The dual laser micro-displacement ranging method based on improved ResNet50V2 according to claim 6 is characterized in that: The specific steps of the original image acquisition in step S1 are as follows: after the power of the CCD camera is connected, the red laser (5-2) is turned on, and the green laser (5-1) is turned off. The red laser beam is irradiated on the sample block to be tested, and a red light spot appears on the sample block. At this time, the computer issues a shooting command to the CCD camera to obtain the initial image before the micro-displacement; then the micrometer is rotated one scale, and the workbench drives the sample block to be tested to move 0.01mm. The computer takes the red light spot image on the sample block again, and the process is repeated until 800 images are taken. The shooting process of the green laser is the same as above.