Intelligent laser triangulation method

Abstract

The application discloses an intelligent laser triangulation method, comprising the following steps: acquiring an imaging light spot image of a measured object, pre-processing the imaging light spot image to form a data set of light spot image coordinates; constructing an improved neural network model capable of effectively capturing the nonlinear relationship between the light spot position and the displacement of the measured object; training the improved neural network model based on the coordinate difference data between the light spots in the training set to obtain a trained improved neural network model; inputting the coordinate difference data between the light spots in the test set into the trained improved neural network model to output a prediction result; inputting the normalized imaging light spot position information, and then performing reverse normalization processing on the prediction result output by the trained improved neural network model to obtain the actual displacement distance of the measured object. The application realizes accurate measurement of complex surface topography and displacement parameters of fast-moving objects, and meets the application requirements of product quality control, process monitoring and the like under high-precision requirements.

Description

Technical Field

[0001] This invention belongs to the field of measurement technology and instruments, and relates to an intelligent laser triangulation method. Background Technology

[0002] Currently, laser triangulation systems still rely on traditional digital image processing algorithms (grayscale centroid method, Gaussian fitting method, and ellipse fitting method) for imaging spot positioning to complete the ranging function. In actual measurements, traditional algorithms are quite sensitive to environmental noise and struggle to maintain stable measurement accuracy under complex lighting conditions, especially when interference signals are present or the reflective surface is non-uniform, significantly affecting the accuracy and reliability of the measurement results. Secondly, traditional methods rely on manually adjusting system structural parameters multiple times to optimize the measurement effect, a cumbersome process that is difficult to adapt to different industrial measurement scenarios, reducing the system's versatility and operational efficiency. These shortcomings make traditional laser triangulation systems unable to meet the demands of modern industry for real-time, high-precision, and intelligent measurement. Summary of the Invention

[0003] To solve the above problems, the technical solution adopted by the present invention is: an intelligent laser triangulation ranging method, comprising the following steps:

[0004] Acquire an image of the light spot on the object under test;

[0005] The imaging spot image is preprocessed to form a dataset of spot image coordinates;

[0006] An improved neural network model can effectively capture the nonlinear relationship between the position of the light spot and the displacement of the object being measured.

[0007] Based on the coordinate difference data between light spots in the training set, the improved neural network model is trained to obtain the trained improved neural network model.

[0008] The coordinate difference data between light spots in the test set is input into the trained improved neural network model, and the output prediction results are obtained.

[0009] Input the normalized imaging spot position information, and then perform inverse normalization on the prediction results output by the trained improved neural network model to obtain the actual displacement distance of the measured object.

[0010] Furthermore, the imaging spot image is acquired using a direct-fire laser triangulation system for imaging spot images.

[0011] Furthermore, the preprocessing of the imaging spot image employs the gray-scale centroid method, Gaussian fitting method, and ellipse fitting method to locate the imaging spot.

[0012] Furthermore, the improved neural network includes a structure of: input layer - fully connected layer - output layer, and is a BP neural network using stochastic gradient descent momentum method.

[0013] Furthermore, the training of the improved neural network model corresponds to the upward and downward movement of the test object relative to the calibration reference plane. Specifically, the test object is placed on the reference plane, first it is moved upward and an imaging spot is collected, then it is moved downward and an imaging spot is collected, and then the upward and downward movement training is performed in sequence.

[0014] Furthermore, the fully connected layer consists of 12 layers, with 550 nodes in each of layers 1-11, and only one neuron in layer 12. The output layer consists of one node, corresponding to the displacement information of the object being measured.

[0015] A smart laser triangulation ranging device, comprising:

[0016] Acquisition module: used to acquire the imaging spot image of the object under test, preprocess the imaging spot image to form a dataset of spot image coordinates;

[0017] Building blocks: Used to build improved neural network models that can effectively capture the nonlinear relationship between the position of the light spot and the displacement of the object being measured;

[0018] Training module: Used to train the improved neural network model based on the coordinate difference data between light spots in the training set, and obtain the trained improved neural network model;

[0019] Prediction module: Used to input the coordinate difference data between light spots in the test set into the trained improved neural network model, and output the prediction results;

[0020] The module is used to input the normalized imaging spot position information, and then perform inverse normalization on the prediction results output by the trained improved neural network model to obtain the actual displacement distance of the measured object.

[0021] A computer device includes: a processor and a memory, the memory storing a program module, characterized in that the program module runs on the processor to implement the method as described in any one of the claims.

[0022] A readable storage medium that stores a program module, which, when executed in a processor, can implement the method as described in any one of the claims.

[0023] An intelligent laser triangulation system includes:

[0024] Linear displacement stage: used to fix the object being measured;

[0025] Image acquisition sensor: used to acquire images and imaging spot; when the object being measured is displaced, the position of the imaging spot on the CMOS photosensitive surface also changes accordingly; each position of the object being measured within the measurement range corresponds to the position of the imaging spot on the CMOS.

[0026] Intelligent laser triangulation distance measuring device: used to measure the actual displacement distance of the object being measured based on the image acquired by the image acquisition sensor and the method described above.

[0027] This invention provides an intelligent laser triangulation method to improve the measurement accuracy and computational efficiency of laser triangulation systems. A nonlinear programming genetic algorithm is used to determine the optimal structural parameters of the system, constructing a direct-fire laser triangulation system to achieve high-resolution distance measurement. Three digital image processing algorithms—grayscale centroid method, Gaussian fitting method, and ellipse fitting method—are used to locate the imaging spot, and a BP neural network is further used to fuse the spot positions to directly determine the displacement of the measured object. This invention's laser triangulation system avoids measurement errors and uncertainties caused by structural parameters by eliminating the need for distance measurement formulas based on system parameters. It also significantly improves measurement accuracy, stability, robustness, and intelligence, making it suitable for industrial scenarios requiring high-precision, non-contact, real-time measurement.

[0028] In industrial manufacturing scenarios, it can be used for high-precision, non-contact, real-time inspection of metal and other material surfaces. By integrating deep learning and laser triangulation technology, this invention can achieve accurate measurement of parameters such as complex surface morphology and displacement of rapidly moving objects, meeting the application needs of product quality control and process monitoring under high-precision requirements. It is widely applicable to industrial scenarios requiring precision measurement, such as semiconductor manufacturing, automobile production, and precision machining.

[0029] Beneficial effects of the present invention

[0030] 1. Reduced Measurement Error and Relative Uncertainty: Using spot positioning data from gray-scale centroid method, Gaussian fitting method, and ellipse fitting method as feature inputs, the trained BP neural network can effectively learn the complex nonlinear relationship between the imaging spot position and the displacement of the measured object. This eliminates the need for the measurement system to use distance measurement formulas based on structural parameters, thus avoiding the measurement errors and uncertainties caused by these parameters.

[0031] 2. Enhanced System Robustness and Adaptability: Compared to traditional spot localization methods, this invention significantly enhances the measurement system's resistance to noise and environmental changes through the adaptive capability of the BP neural network. The spot is preprocessed using various traditional algorithms before being input into the neural network model, further improving the system's stability and robustness, making the laser triangulation measurement system applicable to various complex industrial measurement environments.

[0032] 3. High degree of intelligence and automation: This invention integrates a BP neural network with a traditional spot localization algorithm, significantly improving the automation level of the system's measurement process. The system can automatically adjust and optimize its measurement strategy based on the input spot data, possessing self-learning capabilities, reducing reliance on manual intervention, and enhancing the level of intelligence in industrial production. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 Flowchart of the method;

[0035] Figure 2 It is based on the principle of laser triangulation.

[0036] Figure 3 It is an elliptical speckle image;

[0037] Figure 4 This is a diagram of the BP neural network model structure;

[0038] Figure 5 This is a flowchart of the BP neural network training process. Detailed Implementation

[0039] It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0040] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the present invention or its application or use. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] This invention is a method for non-contact, high-precision, and real-time measurement of the displacement of an object based on the principle of direct laser triangulation and by introducing a BP neural network to integrate three traditional imaging spot positioning algorithms.

[0042] Figure 1This is a flowchart of the method;

[0043] A smart laser triangulation method includes the following steps:

[0044] S1: Acquire the image of the light spot on the object under test;

[0045] S2: Preprocess the imaging spot image to form a dataset of spot image coordinates;

[0046] S3: Construct an improved neural network model that can effectively capture the nonlinear relationship between the position of the light spot and the displacement of the object being measured;

[0047] S4: Based on the coordinate difference data between light spots in the training set, train the improved neural network model to obtain the trained improved neural network model;

[0048] S5: Input the coordinate difference data between the light spots in the test set into the trained improved neural network model, and output the prediction results;

[0049] S6: Input the normalized imaging spot position information, and then perform inverse normalization on the prediction results output by the trained improved neural network model to obtain the actual displacement distance of the measured object.

[0050] The steps S1 / S2 / S3 / S4 / S5 are executed sequentially;

[0051] Figure 2 It is based on the principle of laser triangulation.

[0052] Figure 3 It is an elliptical speckle image;

[0053] A measurement system for the imaging spot is constructed. Under the constraints of Scheimpflug's imaging law, the size of the imaging spot, and the geometric dimensions of the mechanical structure, the structural parameters of the laser triangulation system are determined using a nonlinear programming genetic algorithm (E-NPGA) with the system resolution as the objective function. A direct-fire laser triangulation system is constructed, in which the laser beam is collimated by a plano-convex lens and then incident perpendicularly onto the surface of the object being measured, resulting in diffuse reflection. The scattered light is then focused onto the CMOS photosensitive surface by an imaging lens.

[0054] The resolution of the direct-fire laser triangulation system or system range As the objective function of N-GA;

[0055] Where c is the pixel size of the image sensor. This refers to the object distance during the focusing process of diffuse reflected light from the object being measured through the imaging lens when the object is on the calibration reference plane. Let α be the angle between the diffuse reflected light and the image sensor, α be the angle between the diffuse reflected light from the reference plane and the incident light, H be the height of the imaging lens from the reference plane, and f be the ideal focal length of the imaging lens. It is worth noting that α, H, and f are the fundamental variables of the entire laser triangulation system; once these three parameters are determined, the entire optical system can be determined.

[0056] The three constraints of N-GA are determined based on different measurement scenarios:

[0057] (1) To ensure that the light spot can always be clearly imaged on the image sensor during the movement of the object being measured, Scheimpflug's imaging law L1*tanα=L′1*tanβ must be satisfied, where It is the image distance during the focusing process of diffuse reflected light from the reference surface through the imaging lens.

[0058] (2) To ensure that the performance of the measurement system is not affected, the size of the elliptical imaging spot on the image sensor needs to be limited, where the size in the X direction on the image sensor is... (Vertical direction of light spot movement), Y-axis dimension is (Direction of movement of the light spot), where r is the radius of the incident light spot on the reference plane. (3) Different measurement scenarios leave different spaces for the laser triangulation system, so it is very important to constrain the geometric dimensions of the system's mechanical structure.

[0059] Since the positions of the collimating lens, image sensor, and imaging lens are not fixed, the dimensional formulas of the system's mechanical structure are discussed in three cases. First, if the image sensor is lower than the collimating lens, the dimensional formulas for the measurement system are W = (L1 + L′1)cosα and R = VH. Second, if the imaging lens is higher than the collimating lens, the dimensional formulas for the measurement system are W = (L1 + L′1)cosα and R = (L1 + L′1)cosα - V. Finally, if the image sensor is higher than the collimating lens, and the imaging lens is lower than the collimating lens, the dimensional formulas are W = (L1 + L′1)cosα and R = (L1 + L′1)cosα - H. Here, W is the lateral dimension of the system, R is the longitudinal dimension of the system, and V is the height of the collimating lens from the collimating plane.

[0060] Under the premise of satisfying all constraints, the first generation population is randomly generated. Fitness is evaluated according to the objective function, and the best chromosome in the first generation is recorded. Entering the evolutionary stage, six evolutionary sequences of three operations (selection, crossover, and mutation) are randomly determined, and evolution begins.

[0061] After each evolutionary iteration, an elitist strategy is executed, replacing the worst chromosome in the current population with the best chromosome from the previous generation. The best chromosome of the current generation is compared with the best chromosome of the previous generation and recorded. After a certain number of generations, a nonlinear programming function is invoked to perform local optimization using the current best chromosome as the initial value. After optimization, the elitist strategy is executed again to improve the local search capability of the traditional genetic algorithm.

[0062] This process is repeated over and over again. After a certain number of generations of evolution, the initial structural parameters of the system that meet the system's resolution or range requirements are finally found: the angle α between the diffuse reflected light and the incident light corresponding to the reference plane, the height H of the imaging lens from the reference plane, and the focal length f of the imaging lens. The evolutionary trends of system resolution and system range during the N-GA evolution process are also shown.

[0063] The system structure parameters provided by N-GA were simulated using optical simulation software. The accuracy of the initial structure parameters was verified by calculating the relative error between the RMS radius of the light spot on the image sensor and the theoretical value of the imaging light spot size.

[0064] Based on the direct-fire laser triangulation system, the displacement formula of the measured object is determined:

[0065]

[0066] In the formula, S′ is the interval value of the imaging spot on the CMOS photosensitive surface, and S is the displacement of the object under test. When the object under test moves upward relative to the reference plane along the optical axis, the denominator of the range formula is "+". Conversely, the denominator of the range formula is "-".

[0067] The preprocessing of the imaging spot image uses the gray-scale centroid method, Gaussian fitting method and ellipse fitting method to locate the imaging spot.

[0068] The gray-scale centroid method is based on the idea of ​​treating the imaging spot as a two-dimensional gray-scale distribution map, considering the gray value of each pixel as the mass of that point, and locating the position of the imaging spot by calculating the centroid of these gray values.

[0069] (2) Gaussian fitting method

[0070] Gaussian fitting method: Under ideal conditions, the light intensity of the imaging spot acquired by the CMOS photosensitive surface follows a Gaussian distribution, meaning the light intensity of the imaging spot smoothly decreases from the center outwards. By fitting the light spot with a Gaussian function, the position of the light intensity amplitude is the position of the center of the light spot.

[0071] (3) Ellipse fitting method

[0072] Ellipse fitting is an image detection method based on the least squares method. It searches for the optimal ellipse parameters by minimizing the sum of squared distances between the fitted ellipse and the edge points of the actual spot, thereby accurately determining the center position of the imaging spot.

[0073] Preprocessing: (1) Read in the imaging spot sequentially; (2) Determine the position coordinates of the spot using the gray centroid method, Gaussian fitting method, and ellipse fitting method respectively; (3) Each imaging spot has three sets of coordinates (x1, y1), (x2, y2), and (x3, y3), calculate the difference between the three sets of coordinates and the reference spot coordinates; (4) The difference is then normalized to become the input data of the BP neural network.

[0074] Each of the three traditional processing algorithms has its own advantages and disadvantages and is suitable for different industrial measurement scenarios. Combining the three algorithms can maximize the ranging accuracy of the laser triangulation system. The imaging spot position data obtained from the three algorithms are used as features and input into the subsequent neural network training model.

[0075] Figure 4 This is a diagram of the BP neural network model structure;

[0076] Figure 5 This is a flowchart of the BP neural network training process.

[0077] The improved neural network comprises an input layer, a fully connected layer, and an output layer, employing a backpropagation (BP) neural network with stochastic gradient descent momentum (SGDM). The activation function is ReLU (Rectified Linear Unit), and the loss function is mean squared error (MSE).

[0078] The imaging spot position data after Min-Max normalization is used as the input of the BP neural network for training. The training of the improved neural network model corresponds to the upward and downward movement of the test object relative to the calibration reference plane.

[0079] This is a BP neural network training method specifically for laser triangulation; (1) the relative distance measured by laser triangulation, i.e. the movement of the object under test relative to the reference plane; (2) when the object under test is on the calibration reference plane, the corresponding imaging spot is located at the center of the CMOS photosensitive surface; (3) when collecting the training spot: ensure that the object under test is on the reference plane, first move it up and collect the imaging spot, then move it down and collect the imaging spot. Then perform the upward and downward training in sequence.

[0080] The fully connected layer consists of 12 layers, with 550 nodes in each of layers 1-11, and only one neuron in layer 12. The output layer consists of one node, corresponding to the displacement information of the measured object.

[0081] Furthermore, by combining the actual displacement data of the object under test with the acquisition frame rate of the CMOS of the measurement system, real-time monitoring of the motion of the object under test or restoration of its surface morphology can be achieved.

[0082] During the continuous movement of the object under test, the CMOS will continuously acquire imaging spots: (1) Knowing the acquisition frame rate of the CMOS, the acquisition interval between two spots can be determined. The distance the object under test moves divided by the acquisition interval is the speed of the object under test; (2) Or, during the lateral movement of the object under test with surface undulations, as long as its lateral movement speed is known, the lateral distance between its surface undulations can be determined, thus realizing the restoration of the surface morphology.

[0083] A smart laser triangulation ranging device, comprising:

[0084] Acquisition module: used to acquire the imaging spot image of the object under test, preprocess the imaging spot image to form a dataset of spot image coordinates;

[0085] Building blocks: Used to build improved neural network models that can effectively capture the nonlinear relationship between the position of the light spot and the displacement of the object being measured;

[0086] Training module: Used to train the improved neural network model based on the coordinate difference data between light spots in the training set, and obtain the trained improved neural network model;

[0087] Prediction module: Used to input the coordinate difference data between light spots in the test set into the trained improved neural network model, and output the prediction results;

[0088] The module is used to input the normalized imaging spot position information, and then perform inverse normalization on the prediction results output by the trained improved neural network model to obtain the actual displacement distance of the measured object.

[0089] An intelligent laser triangulation system includes:

[0090] Linear displacement stage: used to fix the object under test; (1) by accurately moving the displacement stage back and forth, the distance of movement of the object under test relative to the calibration reference surface is determined; (2) it is also the source of label data in the training process of BP neural network.

[0091] CMOS image acquisition sensor: used to acquire images and imaging spot; when the object being measured is displaced, the position of the imaging spot on the CMOS photosensitive surface also changes accordingly; each position of the object being measured within the measurement range corresponds to the position of the imaging spot on the CMOS.

[0092] Intelligent laser triangulation distance measuring device: used to measure the actual displacement distance of the object being measured based on the image acquired by the image acquisition sensor and the method described above.

[0093] A computer device includes: a processor and a memory, the memory storing a program module that runs on the processor to implement the method as described in any one of the claims.

[0094] A readable storage medium that stores a program module, which, when executed in a processor, can implement any of the methods described herein.

[0095] Example 1: This example is based on the laser triangulation principle and deep neural network. The method for measuring the distance of the object under test is as follows: a collimated Gaussian beam is vertically projected onto the surface of the object under test; a reflected light image acquisition module is used to acquire light spots on the surface of the object under test to obtain diffuse reflected light images corresponding to each calibration surface of the object under test; after preprocessing the imaging light spots, the images are input into a BP neural network for model training to achieve accurate measurement of the distance of the object under test.

[0096] I. System Setup

[0097] 1. Construct a laser triangulation system with a resolution of 5µm. The system's structural parameters are as follows: the ideal focal length of the collimating lens is f = 2mm, the distance from the laser fiber port is 2mm, the distance from the calibration reference plane of the object being measured is 70mm, the angle between the diffuse reflected light and the incident light of the reference plane is α = 31.14°, the height of the imaging lens from the reference plane is H = 59.27mm, the ideal focal length of the imaging lens is f = 25mm, the tilt angle of the CMOS photosensitive surface is β = 46.92°, and the distance from the imaging lens is 35.93mm.

[0098] 2. The metal object under test is fixed on a linear displacement stage (resolution 0.018µm) for precise control of its displacement. The acquisition frequency of the CMOS image sensor is set to 100Hz to ensure that the position of the imaging spot corresponds to the displacement information of the object under test at all times during the acquisition process.

[0099] II. BP Neural Network Training Process

[0100] 3. A backpropagation (BP) neural network model was constructed, employing an input layer—fully connected layer—output layer structure. The input layer contains 3 nodes, corresponding to the position information of the imaging spot; the fully connected layer consists of 12 layers, with layers 1-11 each containing 550 nodes; the 12th layer has 1 neuron node; and the output layer has 1 node, corresponding to the displacement information of the measured object. The activation functions chosen are ReLU and Sigmoid, the loss function is mean squared error (MSE), and the optimization algorithm is stochastic gradient descent momentum method (SGDM).

[0101] 4. The imaging spot acquired by CMOS is pre-positioned using the gray-scale centroid method, Gaussian fitting method, and ellipse fitting method. The spot position data is then processed by Min-Max normalization to eliminate the influence between different units, so that the feature values ​​are within the same numerical range, thereby improving the training speed and generalization ability of the BP neural network model.

[0102] 5. The normalized spot positioning data is input into a BP neural network for training. The training process is divided into two parts, corresponding to the upward and downward movement of the object under test based on the calibration reference plane. The training parameters for upward movement are: a learning rate of 0.003, training for 80 rounds; then incremental learning, reducing the learning rate to 0.002, with a learning rate decay period of 10 and a decay factor of 0.5, training for 100 rounds. The training parameters for downward movement are: a learning rate of 0.002, training for 100 rounds; then incremental learning, further reducing the learning rate to 0.0015, training for 200 rounds. After each training round, the data is randomly rearranged.

[0103] 6. The BP neural network model was tested using the training and validation sets to evaluate its performance in real-world measurements. The results are as follows:

[0104]

[0105] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent laser triangulation method, characterized in that: The method comprises the following steps: An imaging light spot image of the measured object is acquired; The imaging light spot image is preprocessed to form a data set of light spot image coordinates; An improved neural network model is constructed, which can effectively capture the nonlinear relationship between the light spot position and the displacement of the measured object; The improved neural network model is trained based on the coordinate difference data between the light spots in the training set, and a trained improved neural network model is obtained; The coordinate difference data between the light spots in the test set is input into the trained improved neural network model, and a prediction result is output; The normalized imaging light spot position information is input, and the prediction result output by the trained improved neural network model is de-normalized to obtain the actual displacement distance of the measured object; The imaging light spot image is collected by a direct laser triangulation system of the imaging light spot image; The resolution of the direct laser triangulation system Or system range As an objective function of N-GA The overall displacement of the imaged spot on CMOS wherein: is the pixel size of the image sensor, is the object distance during the focusing of the diffuse reflection light of the reference surface by the imaging lens, is the angle between the diffuse reflection light and the image sensor, is the angle between the diffuse reflection light and the incident light of the reference surface, is the height of the imaging lens from the reference surface, is the ideal focal length of the imaging lens; Three constraint conditions of N-GA are determined according to different measurement scenes: To ensure that the measured object in the moving process, the light spot can be always clear imaging on the image sensor, need to meet the Scheimpflug imaging law Wherein Is the image distance of the diffuse reflection light of the reference plane in the focusing process of the imaging lens. To ensure that the performance of the measurement system is not affected, the size of the elliptical imaging spot on the image sensor needs to be limited, wherein the size in the X direction on the image sensor, i.e. perpendicular to the direction of movement of the spot, is and the size in the Y direction on the image sensor, i.e. in the direction of movement of the spot, is wherein is the radius of the incident light spot on the reference surface. The space left for the laser triangulation system is different in different measurement scenes, so the geometric dimensions of the system mechanical structure are constrained.

2. The intelligent laser triangulation method according to claim 1, wherein: The imaging light spot image is preprocessed by using the gray centroid method, the Gaussian fitting method and the ellipse fitting method to collect the imaging light spot.

3. The intelligent laser triangulation method according to claim 1, wherein: The improved neural network includes an input layer, a full connection layer and an output layer, and adopts a BP neural network with a momentum method of random gradient descent.

4. The intelligent laser triangulation method of claim 1, wherein: The training of the improved neural network model corresponds to the upward movement and downward movement of the measured object relative to the calibration reference surface, specifically, the measured object is on the reference surface, is first moved upward and collects the imaging light spot, is then moved downward and collects the imaging light spot, and the upward movement and downward movement training are sequentially performed.

5. The intelligent laser triangulation method according to claim 3, wherein: The full connection layer has 12 layers, 550 nodes in each of the first 11 layers, only one neuron in the 12th layer, and the output layer has one node corresponding to the displacement information of the measured object.

6. A device for intelligent laser triangulation method, applying the intelligent laser triangulation method according to any one of claims 1-5, characterized in that: It comprises: An acquisition module for acquiring the imaging light spot image of the measured object, preprocessing the imaging light spot image, and forming a data set of light spot image coordinates; A construction module for constructing an improved neural network model capable of effectively capturing the nonlinear relationship between the light spot position and the displacement of the measured object; A training module for training the improved neural network model based on the coordinate difference data between the light spots in the training set, and obtaining a trained improved neural network model; A prediction module for inputting the coordinate difference data between the light spots in the test set into the trained improved neural network model, and outputting a prediction result; A obtaining module for inputting normalized imaging light spot position information, de-normalizing the prediction result output by the trained improved neural network model, and obtaining the actual displacement distance of the measured object.

7. A computer device comprising: A processor and a memory, wherein the memory stores program modules, and the program modules are run on the processor to implement the method of any one of claims 1-5.

8. A readable storage medium, storing program modules, characterized in that, The program modules run on the processor to implement the method of any one of claims 1-5.

9. An intelligent laser triangulation system characterized in that: It comprises: A linear displacement table for fixing the measured object; Image acquisition sensor: used for acquiring image acquisition imaging spot; when the measured object is displaced, the position of the imaging spot on the CMOS photosensitive surface also changes accordingly; each position of the measured object within the range corresponds to the position of the imaging spot on the CMOS; Intelligent laser triangulation device: used for measuring the actual displacement distance of the measured object based on the images collected by the image acquisition sensor, using the method of any one of claims 1-5.

Images