Intelligent sensor for surveying ground height in highway design

Abstract

The invention discloses an intelligent sensor for surveying the ground height in highway design, and belongs to the technical field of intelligent sensors. The system specifically comprises a data acquisition and preprocessing module, a self-adaptive compensation module, a data optimization module, an elevation calculation module and an optimization output module, the data acquisition and preprocessing module receives and preprocesses original data; the adaptive compensation module compensates the reflection characteristic difference of different materials by using an algorithm; the data optimization module performs fine correction on the compensated data by using an algorithm; the elevation calculation module converts the corrected distance value into absolute elevation; the optimization output module is used for integrating and outputting the processing of the modules; compared with a traditional highway ground height surveying means, the method has the technical advantage of precision optimization, the problem that traditional height surveying is poor in precision is solved, and the technical effect that surveying precision is highly optimized is achieved.

Description

Technical Field

[0001] This invention relates to the field of intelligent sensor technology, and in particular to an intelligent sensor for surveying ground elevation in highway design. Background Technology

[0002] In the process of modern urbanization, highway design, as the cornerstone of urban planning and construction, hinges on the accurate surveying of topography. Among these, the measurement of ground elevation directly affects the layout and implementation of infrastructure such as roads, bridges, drainage systems, and green landscapes. If there are deviations in ground elevation data, it will lead to distorted design schemes, resulting in project delays, safety hazards, and even economic losses. For example, in slope development projects, if the elevation survey error exceeds 5 centimeters, it may cause uneven drainage slopes, inducing water accumulation or landslide risks. When excavating foundation pits for high-rise buildings, even micron-level elevation deviations can be amplified into structural stress problems at the meter level. Therefore, the development of efficient and accurate ground elevation surveying tools has become an urgent need in the field of highway pavement engineering.

[0003] Traditional highway design and surveying mainly rely on optical surveying instruments, such as levels, total stations, and theodolites. These devices calculate height differences by manually aligning the target point and reading parallax or angle data. As a classic tool, the level uses the direction of gravity as a reference plane to measure relative height. With the increase in terrain complexity, traditional methods require surveyors to manually focus and record data. When surveyors are fatigued or lack experience, this can lead to large errors in survey accuracy. In addition, the limitations of traditional surveying methods are further amplified in complex terrain such as hills or urban canyons. Summary of the Invention

[0004] The purpose of this invention is to disclose an intelligent sensor for surveying ground elevation in highway design. It receives raw distance signal data and performs preliminary processing using data preprocessing techniques. Further, it utilizes neural network structures and algorithms to identify and compensate for differences in the reflectivity of different materials. The compensated data is then refined using multimodal algorithms. The corrected data is converted into absolute elevation using further algorithmic techniques. Finally, the optimized elevation data is output as the final height survey data. This invention addresses the technical problems of traditional surveying methods that rely too heavily on manual readings and are unsuitable for complex terrain surveys, achieving a highly optimized surveying accuracy.

[0005] The specific plan is as follows: The present invention discloses an intelligent sensor for surveying ground elevation in highway design, specifically comprising: a data acquisition and preprocessing module, an adaptive compensation module, a data optimization module, an elevation calculation module, and an optimization output module, wherein the above modules are connected in sequence; The data acquisition and preprocessing module receives raw distance signal data and performs preprocessing to obtain an initial distance sequence. The adaptive compensation module uses a lightweight convolutional neural network model to identify the ground material at the light spot illumination point based on the initial distance sequence, and compensates for the differences in reflectivity of different materials to obtain compensation data. The data optimization module, based on the compensation data, introduces environmental data and, based on a multi-factor coupled real-time error model, performs refined error correction on the ranging value to obtain corrected data. The elevation calculation module is based on the Kalman filter-based adaptive datum calibration algorithm, which converts the correction data into absolute elevation, obtains the elevation calculation results, and calibrates the datum drift in real time. The optimization output module performs elevation post-processing optimization and data integration on the elevation calculation results and outputs the final elevation data through the interface.

[0006] Furthermore, the data acquisition and preprocessing module includes the raw distance signal from the laser rangefinder sensor, which is preprocessed to perform filtering and outlier removal.

[0007] Furthermore, the adaptive compensation module includes a material identification submodule, a material reflection coefficient table construction submodule, and a reflection compensation submodule; The material recognition submodule uses a lightweight one-dimensional convolutional neural network (1D-CNN) to process the echo waveform. The processing layer structure of the model includes an input layer → convolutional layer 1 → pooling layer 1 → convolutional layer 2 → pooling layer 2 → flattening layer → fully connected layer 1 → output layer. When training the model, a large amount of laser echo waveform data under known materials is collected in advance and labeled. The trained network model is embedded into the module to output the probability of each category. The category corresponding to the highest probability is taken as the material type M, and the highest probability value is recorded as the confidence level P. The labels include [0 concrete, 1 asphalt, 2 grass, 3 soil]. The material reflection coefficient table construction submodule extracts the initial distance measurement value of the current measuring point. Based on the material type M and confidence level P, which is in the range of [0, 1], a material reflection coefficient table is established in advance through experimental measurement. This table stores the reflection coefficient R of each material under standard conditions. The reflection compensation submodule, based on the material type M, looks up the standard reflection coefficient R in the material reflection coefficient table and uses linear adjustment compensation based on confidence level. The compensation strategy is as follows: compensation is performed when P ≥ 0.8, the default material 0 is used when P < 0.8, and no compensation is performed when P = 0. The compensation formula is as follows: D 补偿 Indicates the precise distance value after compensation; D 初始Indicates the initial distance measurement; k represents the calibration coefficient; P represents the confidence level; R represents the standard reflectance coefficient; [1+(...)*P] represents the core compensation factor; [1+(...)*P] represents the final linear scaling factor; the accurate distance value after compensation is calculated using this formula and then output.

[0008] Furthermore, the data optimization module includes an environmental data retrieval submodule, a light speed change correction submodule, and a thermal deformation correction submodule; The environmental data retrieval submodule retrieves environmental data from the actual survey through a data interface. This data includes temperature, humidity, and air pressure data, with dimensions [T, H, P], where T represents temperature data, H represents humidity data, and P represents air pressure data. The light speed change correction submodule calculates the current light speed and corrects for the distance based on the environmental data during the actual survey. Its light speed correction formula is as follows: D 修 This is the initially corrected distance; D 补偿 The above-mentioned accurate distance value after compensation; C 实 C represents the actual speed of light under current environmental conditions. 参 This is the reference speed of light under standard conditions; the formula is initially corrected based on the speed of light under actual conditions. The thermal deformation correction submodule fits the relationship between temperature and deformation error using experimental data. The formula for calculating the thermal deformation error relationship is: ∆D 热 =k1×(T-T0)+k2×(T-T0) 2 , where ∆D 热 This represents the distance error caused by thermal deformation; T represents the temperature of the current survey environment; T0 represents the reference temperature, set to 25℃; k1 represents the linear thermal expansion coefficient; k2 represents the nonlinear thermal expansion coefficient; k1×(T-T0) represents the linear term, describing the linear expansion / contraction of the material with temperature changes; k2×(T-T0) 2 This represents a quadratic term that describes a nonlinear thermal effect.

[0009] Furthermore, the data optimization module also includes an air refraction change correction submodule, a multi-factor coupling error model submodule, and a result output submodule; The air refractive index correction submodule uses Edlen to calculate the current air refractive index n, and further uses the refractive index error correction formula to calculate the optical path difference caused by the change in air refractive index, converting the refractive index change into an actual distance correction amount. The refractive index error correction formula is: ∆D 折 =D 修 ×(n-n0); where ∆D 折 D represents the amount of refractive index error correction; 修This represents the distance after correction for the speed of light; n represents the refractive index of the air in the current survey environment; n0 represents the refractive index of the air under standard conditions, taken as 1.000237; The multi-factor coupled error model submodule integrates the correction results of the light speed change correction submodule, thermal deformation correction submodule and air refraction change correction submodule and considers the coupling effect between the correction results to establish a comprehensive error model to calculate the total error correction distance. The result output submodule performs a reasonableness check on the total error correction distance, including determining the range of distance changes and detecting outliers. When an anomaly occurs, its response strategy is to use the effective correction value at time step t-1. Finally, the total correction distance and the confidence level E of the environmental parameter are calculated. 环 Output the results.

[0010] Furthermore, the elevation calculation module includes a coordinate transformation submodule, a datum surface drift detection submodule, and an elevation post-processing submodule; The coordinate transformation submodule, based on the total correction distance and combined with the attitude and absolute position data of the intelligent sensor (dimensions [X, Y, Z]), converts the data into the absolute elevation of the measurement point to obtain the original elevation sequence. The reference surface drift detection submodule detects reference surface changes based on the statistical characteristics of continuous measurement points according to the original elevation sequence. It further uses a Kalman filter to estimate the current drift amount and then calibrates the current value, finally obtaining the calibrated elevation data denoted as H. 漂 ; The elevation post-processing submodule is based on the elevation data H 漂 The data is smoothed using a confidence-based weighted moving average filter, where the weight of the current measurement is calculated using the following formula: Where W represents the weight of the current measurement; E 环 H represents the confidence level of the introduced environmental data, ranging from [0, 1]; 漂 H represents the estimated deviation from the current elevation. 漂max The maximum elevation deviation threshold is set to 0.1m; the weighted moving average is then calculated using the following formula: H 平滑 This indicates the final elevation output after smoothing; W i H represents the adaptive weight of the i-th sampling point; 漂i Let represent the calibrated elevation value of the i-th sampling point; the smoothed elevation data is obtained through calculation using the formula, and its calibration confidence is further calculated using the following formula: E 校 The value represents the calibration confidence level, ranging from [0, 1]; the rest are expressed in the same way as the weight calculation formula above.

[0011] Furthermore, the elevation calculation module also includes a confidence calculation submodule; The confidence calculation submodule will calculate the environmental confidence E. 环 and calibration confidence level E 校 These are combined to form a comprehensive confidence score, used to assess the overall reliability of the current measurement point. The weighting formula is: C 综 =E 环 *0.4+E 校 *0.6, where C 综 E represents the overall confidence level. 环 This indicates the confidence level of the introduced environmental data, with a weight set to 0.4; E 校 This represents the calibration confidence level, with a weight set to 0.6 and a total of 1.

[0012] Furthermore, the optimized output module includes a consistency optimization submodule, a consistency confidence submodule, and a final output submodule; The local consistency optimization submodule is based on the calibrated elevation data H 漂 The LOESS local weighted regression algorithm based on adaptive weights is used to smooth out abnormal undulations while preserving the real terrain features, and finally the optimized elevation data is calculated. The consistency confidence submodule extracts the confidence calculation submodule mentioned above and calculates the comprehensive confidence C. 综 Further calculation of local consistency confidence is performed, and this is compared with the overall confidence level C. 综 Further weighted fusion, with the weights allocated sequentially as follows: overall confidence level C 综 The local consistency confidence level is 0.7, which leads to a local consistency confidence level of 0.3, resulting in the final overall confidence level C. 总 ; The final output submodule uses the elevation data optimized for local consistency as the final ground survey height data H. 终 And compare it with the overall confidence level C 总 Output is provided through the data interface of the smart sensor.

[0013] Compared with the prior art, the present invention has at least one of the following technical effects: 1. This invention solves to some extent the technical problems of traditional surveying methods relying too much on manual readings and being unable to adapt to complex terrain surveying, thus improving the accuracy of ground elevation surveying.

[0014] 2. The adaptive compensation module of this invention utilizes a lightweight convolutional neural network model to identify the ground material mapped by the light spot of the laser emitted by the intelligent sensor onto the survey point. It further constructs a material reflectance coefficient table and uses an algorithm to compensate for the differences in reflectance characteristics of different materials. Finally, it obtains accurate distance values ​​and material labels, providing stable data support for subsequent modules. Compared with traditional ground height surveying, it has the technical effect of adaptive compensation for complex terrain, solving the problem that traditional surveying methods cannot adapt to complex terrain, resulting in poor surveying accuracy.

[0015] 3. The data optimization module of this invention introduces environmental factors and uses a multimodal correction algorithm to establish an error model based on multi-factor coupling in real time. This model is used to further refine the compensated distance data, ultimately obtaining highly refined distance data. This provides data support for further optimization. Compared with traditional height surveying, it has the technical effect of accuracy optimization and solves the problem that traditional ground surveying methods rely too much on manual calculation and correction, resulting in poor surveying accuracy.

[0016] 4. The elevation calculation module of this invention uses an adaptive datum calibration algorithm based on Kalman filtering to convert the corrected distance value into absolute elevation and to calibrate in real time the slow drift of the datum caused by factors such as deformation of the sensor mounting platform and temperature drift. Compared with traditional ground height surveying, it has the technical effect of real-time calibration of datum drift and solves the problem of excessive datum drift caused by the lack of real-time calibration technology in traditional surveying methods.

[0017] 5. The optimized output module of this invention utilizes the LOESS algorithm, a locally weighted regression algorithm based on adaptive weights, to further optimize the elevation data and calculate the local consistency confidence level. Compared with traditional surveying, it has the technical effect of deep optimization of output data, solving the problem of low data confidence level caused by the lack of output data re-optimization technology in traditional surveying methods. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments or examples of the present invention, the drawings used in the embodiments or examples will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained according to these drawings without creative effort.

[0019] Figure 1 This is a module architecture diagram of an intelligent sensor for surveying ground elevation in highway design according to the present invention; Figure 2 This is a diagram of the adaptive compensation module architecture of the present invention; Figure 3 This is a diagram of the data optimization module architecture of the present invention; Figure 4 This is a diagram of the elevation calculation module architecture of the present invention; Figure 5 The output module architecture diagram is optimized for this invention. Detailed Implementation

[0020] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0021] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0022] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0023] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0024] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0025] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0026] The specific implementation method is as follows: See appendix Figure 1 A smart sensor for surveying ground elevation in highway design includes: a data acquisition and preprocessing module, an adaptive compensation module, a data optimization module, an elevation calculation module, and an optimization output module, which are connected in sequence. The data acquisition and preprocessing module receives raw distance signal data and performs preprocessing to obtain an initial distance sequence. The adaptive compensation module uses a lightweight convolutional neural network model to identify the ground material at the light spot illumination point based on the initial distance sequence, and compensates for the differences in reflectivity of different materials to obtain compensation data. The data optimization module, based on the compensation data, introduces environmental data and, based on a multi-factor coupled real-time error model, performs refined error correction on the ranging value to obtain corrected data. The elevation calculation module is based on the Kalman filter-based adaptive datum calibration algorithm, which converts the correction data into absolute elevation, obtains the elevation calculation results, and calibrates the datum drift in real time. The optimization output module performs elevation post-processing optimization and data integration on the elevation calculation results and outputs the final elevation data through the interface.

[0027] The aforementioned intelligent sensor for surveying ground elevation in highway design integrates multiple modules. This modular design breaks down each function into sub-modules, each with specific functions. These interconnected modules form a complete intelligent sensor network for surveying ground elevation. The sensor incorporates deep learning algorithms and neural network technologies, along with various data processing techniques. Its surveying accuracy is deeply optimized to meet the requirements of highway design.

[0028] In specific implementation, the data acquisition and preprocessing module includes the original distance signal from the laser rangefinder sensor, which performs data preprocessing on the signal. The preprocessing process includes filtering and outlier removal.

[0029] The data preprocessing described above includes filtering and outlier removal. The filtering utilizes an adaptive sliding window filtering algorithm, which, unlike filters with fixed parameters, can analyze the variance of the input signal in real time and dynamically adjust the size of the filtering window. When the signal changes drastically, the window is automatically narrowed to preserve details, and when the signal is stable, the window is widened to improve smoothness. At the same time, the trend outlier removal algorithm can identify and isolate outliers caused by instantaneous strong interference (birds or dust), ultimately providing high-quality data support for subsequent modules.

[0030] For specific implementation details, please refer to the appendix. Figure 2 The adaptive compensation module includes a material identification submodule, a material reflection coefficient table construction submodule, and a reflection compensation submodule; The material recognition submodule uses a lightweight one-dimensional convolutional neural network (1D-CNN) to process the echo waveform. The processing layer structure of the model includes an input layer → convolutional layer 1 → pooling layer 1 → convolutional layer 2 → pooling layer 2 → flattening layer → fully connected layer 1 → output layer. When training the model, a large amount of laser echo waveform data under known materials is collected in advance and labeled. The trained network model is embedded into the module to output the probability of each category. The category corresponding to the highest probability is taken as the material type M, and the highest probability value is recorded as the confidence level P. The labels include [0 concrete, 1 asphalt, 2 grass, 3 soil]. The material reflection coefficient table construction submodule extracts the initial distance measurement value of the current measuring point. Based on the material type M and confidence level P, which is in the range of [0, 1], a material reflection coefficient table is established in advance through experimental measurement. This table stores the reflection coefficient R of each material under standard conditions. The reflection compensation submodule looks up the standard reflection coefficient R in the material reflection coefficient table according to the material type M, and uses a linear adjustment compensation strategy based on confidence level. The compensation strategy is as follows: compensation is performed when P ≥ 0.8, the default material 0 is used when P < 0.8, and no compensation is performed when P = 0. The compensation formula is as follows: D 补偿 Indicates the precise distance value after compensation; D 初始 Indicates the initial distance measurement; k represents the calibration coefficient; P represents the confidence level; R represents the standard reflectance coefficient; [1+(...)*P] represents the core compensation factor; [1+(...)*P] represents the final linear scaling factor; the accurate distance value after compensation is calculated using this formula and then output.

[0031] The processing layer structure of the above model includes an input layer → convolutional layer 1 → pooling layer 1 → convolutional layer 2 → pooling layer 2 → flattening layer → fully connected layer 1 → output layer. The specific parameter settings and calculation process for each layer are as follows: Input layer: The input is echo waveform data (obtained from the initial distance sequence). Assuming L = 128 sampling points, the waveform data preprocessing formula is as follows: , where X 标准 [i] represents the value of the i-th sample point after standardization; X 原始 [i] represents the value of the original echo waveform at the i-th sampling point; μ 均 The mean of 0 is used for centering; σ 差 The standard deviation of 1 is used for scaling; the purpose is to improve the stability of model training and accelerate the convergence speed, and finally obtain the standardized waveform tensor X[b, 128, 1]. Convolutional Layer 1: Uses Conv1D convolution with the following parameters: 16 filters → 5 kernel size → 1 stride, padding "same" (keeping length constant) → activated using ReLU. The formula for Convolutional Layer 1 is: Z 1 [j] represents the linear output of convolutional layer 1 at position j; W 1 [i] represents the i-th weight coefficient of convolutional layer 1; X 波 [t+1] represents the value of the output waveform processed at time step t+1; b 1 This represents the bias term of convolutional layer 1; the ReLU function is used to apply the bias term to Z. 1 [j] performs the activation operation to obtain A. 1 [j] Linear output of convolutional layer 1; Pooling layer 1: Pooling is performed using MaxPooling1D, and the pooling formula is as follows: A 1 [2j+k] represents the value of the activation function output of convolutional layer 1 at position 2j+k; P 1 [j] represents the max pooling result of pooling layer 1 at position j; the aim is to reduce the sequence length from 128 to 64, enhance robustness to waveform time shifts, and retain the most significant feature responses. Its output dimension is P. 1 [b, 64, 32]; Convolutional Layer 2: Uses Conv1D convolution with the following parameters: 32 filters → 5 kernels → 1 stride, padding "same" (keeping length constant) → activated using ReLU. The formula for Convolutional Layer 2 is as follows: Z 2 [j] The linear output of convolutional layer 2 at position j; W2 [i] represents the i-th weight coefficient of convolutional layer 2; P 1 [t+1] represents the output of pooling layer 1 at position t+i (time step); b 2 This represents the bias term of convolutional layer 2; the ReLU function is used to apply the bias term to Z. 2 [j] performs the activation operation to obtain A. 2 [j] Linear output of convolutional layer 2; Pooling layer 2: Pooling operations are performed using MaxPooling1D, and the pooling formula is as follows: A 2 [2j+k] represents the value of the activation function output of convolutional layer 2 at position 2j+k; P 2 [j] represents the max pooling result at position j of pooling layer 1; the purpose is to further reduce the dimensionality of the sequence length, and its output dimension is P. 2 [b, 32, 64]; Flattening layer: Flattens the 3D output into 1D for input into the fully connected layer. The flattening formula is: F = Flatten(P 2 ), where F represents the flattened eigenvector; Flatten represents the flattening function; P 2 This represents the output of pooling layer 2; Fully connected layer 1: The number of neurons is set to 64, and its forward propagation formula is: Z 线 =W 权1 ·F+b 偏 Z 线 The linear output of the fully connected layer is represented by F (length 128); F represents the flattened feature vector (length 32*64=2048 obtained from the flattening formula above); b 偏 This represents the bias term of the fully connected layer (length 128); its purpose is to non-linearly combine features to learn complex relationships between them; the ReLU function is used on Z. 线 Perform the activation operation to obtain A. 线 Output; Output layer: The number of neurons is set to 4 (corresponding to the four ground material labels: concrete, asphalt, grass, and soil). Its forward propagation formula is: Z 输 =W 输 ·A drop +b 偏1 Z 输 This indicates a linear output from the output layer, with a length of 4 (corresponding to 4 material types); W 输 This represents the weight matrix of the output layer, with a shape of [128, 4]; A drop This represents the output of the fully connected layer 1 after the Dropout layer (a portion of neurons are randomly dropped during training and not used during inference); b 偏1This represents the bias term of the output layer (length 4), which aims to map the 128-dimensional features to 4 dimensions, corresponding to the scores of the four materials; the SoftMax function is used for activation, and its activation formula is:

[0032] Where y represents the output after the Softmax function (representing the predicted probabilities of the four categories, with the probability of each category between [0, 1], and the sum of the probabilities of all categories being 1); Z 输 [j] represents the score of the i-th category in the linear output of the output layer; exp represents the exponential function; the purpose is to convert the linear output into a probability distribution for multi-class classification problems.

[0033] The loss function (classification cross-entropy) is further calculated to measure the difference between the model's predicted probability distribution and the true probability distribution. Its formula is as follows:

[0034] Where L represents the average cross-entropy loss of the batch samples; N represents the number of samples; y true [i,j] represents the j-th category of the true label of the i-th sample; y[i,j] represents the probability that the i-th sample is predicted to be of the j-th category; log represents the inverse operation of the exponential function; the average cross-entropy loss of the samples is calculated using this formula, and the model is trained by minimizing the loss function.

[0035] For specific implementation details, please refer to the appendix. Figure 3 The data optimization module includes an environmental data retrieval submodule, a light speed change correction submodule, and a thermal deformation correction submodule; The environmental data retrieval submodule retrieves environmental data from the actual survey through a data interface. This data includes temperature, humidity, and air pressure data, with dimensions [T, H, P], where T represents temperature data, H represents humidity data, and P represents air pressure data. The light speed change correction submodule calculates the current light speed and corrects for the distance based on the environmental data during the actual survey. Its light speed correction formula is as follows: D 修 This is the initially corrected distance; D 补偿 The above-mentioned accurate distance value after compensation; C 实 C represents the actual speed of light under current environmental conditions. 参 This is the reference speed of light under standard conditions; the formula is initially corrected based on the speed of light under actual conditions. The thermal deformation correction submodule fits the relationship between temperature and deformation error using experimental data. The formula for calculating the thermal deformation error relationship is: ∆D 热 =k1×(T-T0)+k2×(T-T0)2 , where ∆D 热 This represents the distance error caused by thermal deformation; T represents the temperature of the current survey environment; T0 represents the reference temperature, set to 25℃; k1 represents the linear thermal expansion coefficient; k2 represents the nonlinear thermal expansion coefficient; k1×(T-T0) represents the linear term, describing the linear expansion / contraction of the material with temperature changes; k2×(T-T0) 2 This represents a quadratic term that describes a nonlinear thermal effect.

[0036] The above calculation of the current speed of light is based on the fact that the speed of light used internally by the sensor is usually unknown. Therefore, a relative correction method is used to calculate the speed of light C under standard conditions. 参 =299,792,458 m / s (speed of light in vacuum), actual speed of light C 实 =C 参 / n, where n is the refractive index of air.

[0037] In practice, the data optimization module also includes an air refraction change correction submodule, a multi-factor coupling error model submodule, and a result output submodule. The air refractive index correction submodule uses Edlen to calculate the current air refractive index n, and further uses the refractive index error correction formula to calculate the optical path difference caused by the change in air refractive index, converting the refractive index change into an actual distance correction amount. The refractive index error correction formula is: ∆D 折 =D 修 ×(n-n0); where ∆D 折 Indicates the refractive index error correction amount; n represents the current atmospheric refractive index of the survey environment; D 修 This represents the distance after correction for the speed of light; n0 represents the air refractive index under standard conditions, taken as 1.000237; The multi-factor coupled error model submodule integrates the correction results of the light speed change correction submodule, thermal deformation correction submodule and air refraction change correction submodule and considers the coupling effect between the correction results to establish a comprehensive error model to calculate the total error correction distance. The result output submodule performs a reasonableness check on the total error correction distance, including determining the range of distance changes and detecting outliers. When an anomaly occurs, its response strategy is to use the effective correction value at time step t-1. Finally, the total correction distance and the confidence level E of the environmental parameter are calculated. 环 Output the results.

[0038] The above calculation of the current air refractive index n using Edlen is based on the Edlen formula:

[0039] Where n represents the air refractive index under the current environment; P represents the current atmospheric pressure (introduced through the data interface); T represents the current ambient temperature (introduced through the data interface); (1.049-0.0157×T) represents the temperature-related air compressibility correction factor; (1+(1.049-0.0157×T)×10 -6 ×P) represents the nonlinear compression effect under high pressure; 720.883 represents the refractive index reference constant under standard conditions; (1+0.003661×T) represents the correction of air density with temperature change; 0.003661 represents the temperature expansion coefficient (1 / 273.15).

[0040] The above-mentioned results are synthesized and their coupling effects are considered. The coupling correction formula is as follows: ∆D 耦 =k1×(T-T0)×(P-P0)+k2×(T-T0) 2 ×(P-P0)+k3×(T-T0)×(P-P0) 2 Where ∆D 耦 k1, k2, and k3 represent coupling error coefficients (obtained through experimental calibration, with typical values ​​of 5 × 10⁻⁶ respectively). -7 m / (℃·hPa), 2×10 -8 m / (℃²·hPa) and 1×10 -9 m / (℃·hPa²)); T represents the current ambient temperature (introduced via the data interface); T0 represents the reference temperature (taken as 25℃); P represents the current atmospheric pressure (introduced via the data interface); P0 represents the reference pressure (taken as 1013.25hPa); This formula consists of three terms: the first is a first-order linear coupling term k1×(T-T0)×(P-P0) describing the first-order interaction between temperature and pressure; the second is a temperature-dominated nonlinear coupling term k2×(T-T0) 2 ×(P-P0) describes the coupling effect between nonlinear temperature changes and pressure; the third is the pressure-dominated nonlinear coupling term k3×(T-T0)×(P-P0). 2 The coupling effect between nonlinear pressure changes and temperature is described; finally, a total correction is performed, and the formula is: D 总 =∆D 热 +∆D 折 +∆D 耦 D 总 Indicates the total correction amount; ∆D 热 This represents the distance error caused by thermal deformation (calculated by the thermal deformation correction submodule mentioned above); ∆D 折 This represents the refractive index error correction amount (calculated by the air refraction change correction submodule mentioned above); ∆D 耦This represents the coupling error (calculated using the above correction formula); this total correction is used to calculate the final distance value.

[0041] For specific implementation details, please refer to the appendix. Figure 4 The elevation calculation module includes a coordinate transformation submodule, a datum surface drift detection submodule, and an elevation post-processing submodule; The coordinate transformation submodule, based on the total correction distance and combined with the attitude and absolute position data of the intelligent sensor (dimensions [X, Y, Z]), converts the data into the absolute elevation of the measurement point to obtain the original elevation sequence. The reference surface drift detection submodule detects reference surface changes based on the statistical characteristics of continuous measurement points according to the original elevation sequence. It further uses a Kalman filter to estimate the current drift amount and then calibrates the current value, finally obtaining the calibrated elevation data denoted as H. 漂 ; The elevation post-processing submodule is based on the elevation data H 漂 The data is smoothed using a confidence-based weighted moving average filter, where the weight of the current measurement is calculated using the following formula: Where W represents the weight of the current measurement; E 环 H represents the confidence level of the introduced environmental data, ranging from [0, 1]; 漂 H represents the estimated deviation from the current elevation. 漂max The maximum elevation deviation threshold is set to 0.1m; the weighted moving average is then calculated using the following formula: H 平滑 This indicates the final elevation output after smoothing; W i H represents the adaptive weight of the i-th sampling point; 漂i Let represent the calibrated elevation value of the i-th sampling point; the smoothed elevation data is obtained through calculation using the formula, and its calibration confidence is further calculated using the following formula: E 校 The value represents the calibration confidence level, ranging from [0, 1]; the rest are expressed in the same way as the weight calculation formula above.

[0042] In practice, the elevation calculation module also includes a confidence calculation submodule; The confidence calculation submodule will calculate the environmental confidence E. 环 and calibration confidence level E 校 These are combined to form a comprehensive confidence score, used to assess the overall reliability of the current measurement point. The weighting formula is: C 综 =E 环 *0.4+E 校 *0.6, where C 综 E represents the overall confidence level. 环This indicates the confidence level of the introduced environmental data, with a weight set to 0.4; E 校 This represents the calibration confidence level, with a weight set to 0.6 and a total of 1.

[0043] The above-mentioned process of estimating the current drift using a Kalman filter and then calibrating the current value includes reference surface drift modeling → drift estimation → reference surface calibration. The reference surface drift modeling extracts multi-scale drift features, and the extraction process includes short-term stability analysis, long-term trend analysis, and periodic component detection; the short-term stability analysis formula is: , where σ 短 [i] represents the short-term standard deviation of the i-th sampling point; H[j] represents the elevation value of the j-th sampling point; This represents the mean elevation within a 5-point window centered at point i; 5 represents the sliding window size (5 sampling points); this formula reflects the consistency between the current point and its neighboring points. A larger value indicates more drastic data fluctuations near that point, while a smaller short-term standard deviation indicates more stable sensor operation. The long-term trend analysis formula is:

[0044] Where Trend[i] represents the trend slope of the i-th sampling point; H[i+15] represents the elevation value of the (i+15)-th sampling point; H[i-15] represents the elevation value of the (i-15)-th sampling point; t[i+15]-t[i-15] represents the time interval; 15 represents the number of sampling points shifted forward and backward; the periodic component detection formula is: P[i]=FFT(H[i-60:i+60]), where P[i] represents the periodic feature of the i-th sampling point; FFT represents the Fast Fourier Transform algorithm; H[i-60:i+60] represents the elevation sequence of 120 sampling points centered at i; 60 represents taking 60 points forward and backward (total window is 120 points); the reference surface drift features are extracted through the above three formulas, and the drift sources are further classified. The classification judgment strategy is as follows: Mechanical vibration: short-term stability > 0.02 (preset threshold) and the overall trend is close to 0. Thermal expansion: The long-term trend slope data shows a clear upward trend, and the main frequency is close to 0.001 (preset threshold). Solar radiation: The data trend is related to the intensity of solar radiation (solar radiation directly affects the sensor temperature, resulting in a trend that changes with the amount of sunlight). Further, the model is constructed using the drift prediction model formula, which is: D 预 [k]=α·d[k-1]+β·ΔT+γ·Δt, where D 预[k] represents the drift prediction at time k; d[k-1] represents the drift estimate at time k-1; ∆t represents the time interval; α, β, and γ represent the model coefficients; β·∆T represents the temperature driving term; γ·∆t represents the time accumulation term; α·d[k-1] represents the history dependency term; The drift estimation utilizes an adaptive Kalman filter, the process of which includes state space definition, state transition, adaptive noise adjustment, and Kalman gain calculation; the adaptive noise adjustment formula is as follows: Q[k]=Q0·(1+λ·|u 漂 [k+1]|) Where Q[k] represents the process noise covariance at time k; Q0 represents the basic process noise; and λ represents the velocity sensitivity coefficient. u 漂 [k+1] represents the drift velocity at time k-1; the process noise is adjusted based on the absolute value of the drift velocity at the previous time. The higher the drift velocity, the higher the process noise. When the drift velocity is high, the uncertainty increases. Therefore, increasing the process noise can make the filter more confident in the observations; further, the Kalman gain is calculated, and the formula is:

[0045] Where K[k] represents the Kalman gain at time k; P[k|k-1] represents the prior state covariance matrix at time k (the uncertainty of the predicted state); H represents the observation matrix (mapping the state to the observation); R represents the observation noise covariance; H·P[k|k-1]·H T This represents the uncertainty in predicting the elevation; H·P[k|k-1]·H T +R represents the overall term that combines system model error and sensor measurement error; P[k|k-1]·H T It represents the statistical relationship between state variables and observed variables; based on the predicted state covariance and observation noise, the Kalman gain is calculated, which determines the degree to which the predicted and observed values ​​are believed during state updates; The drift compensation formula for the reference surface calibration is: H 漂 [k]=H 原始 [k]-D 预 [k], where H 漂 [k] represents the calibrated elevation at time k; H 原始 [k] represents the original high-order sequence; D 预 [k] represents the estimated drift at time k; the calibrated elevation is obtained by subtracting the estimated drift from the original elevation.

[0046] For specific implementation details, please refer to the appendix. Figure 5The optimized output module includes a consistency optimization submodule, a consistency confidence submodule, and a final output submodule; The local consistency optimization submodule is based on the calibrated elevation data H 漂 The LOESS local weighted regression algorithm based on adaptive weights is used to smooth out abnormal undulations while preserving the real terrain features, and finally the optimized elevation data is calculated. The consistency confidence submodule extracts the confidence calculation submodule mentioned above and calculates the comprehensive confidence C. 综 Further calculation of local consistency confidence is performed, and this is compared with the overall confidence level C. 综 Further weighted fusion, with the weights allocated sequentially as follows: overall confidence level C 综 The local consistency confidence level is 0.7, which leads to a local consistency confidence level of 0.3, resulting in the final overall confidence level C. 总 ; The final output submodule uses the elevation data optimized for local consistency as the final ground survey height data H. 终 And compare it with the overall confidence level C 总 Output is provided through the data interface of the smart sensor.

[0047] The aforementioned locally weighted regression algorithm LOESS based on adaptive weights smooths out abnormal fluctuations while preserving true terrain features. The implementation process includes neighborhood analysis → weight calculation → weighted smoothing → local consistency confidence calculation; the neighborhood analysis includes spatial and temporal neighborhoods. For a spatial neighborhood (i.e., when data is available at a location), the Euclidean distance between point i and point j within the neighborhood is calculated using the following formula: , where d ij Represents the Euclidean distance between points i and j; (x i -x j ) 2 and (y) i -y j ) 2 Let represent the planar coordinates of points i and j, respectively; the purpose of this formula is to measure the proximity of their spatial positions; d is chosen. ij Point j with a spatial threshold (for the survey environment, the spatial threshold values ​​are: sidewalk width 1.5-3.0 meters, lane width 3.0-3.5 meters, typical step length 0.6-0.8 meters, and sensor moving speed 1-2 meters / second) is taken as a neighborhood point; further, for the temporal neighborhood, points with indices in the time series within the range of [i-window radius, i+window radius] are directly taken (it must be ensured that the index is within the valid range), and finally the neighborhood point set N(i) of each point i is obtained; The weight calculation includes spatial, elevation, and confidence weight calculations. The formula for calculating the spatial weight is:

[0048] Among them W 空间 [i,j] represents the weight values ​​based on spatial distance (range [0,1]); exp represents the exponential function operation; d ij 2σ represents the distance between two points; 空间 The spatial scale parameter is represented (typically 1m); its exponential decay rule is that the greater the distance, the smaller the weight; the elevation weight is further calculated using the following formula:

[0049] Among them W 高程 [i,j] represents the weight values ​​based on elevation differences (range [0,1]); exp represents the exponential function operation; H 稳 [i] represents the stable elevation value of point i; H 稳 [j] represents the stable elevation value of point j; 2σ 高程 This represents the elevation difference scale parameter (with a typical value of 0.05m); the smaller the elevation difference, the closer the weight is to 1, and the larger the elevation difference, the closer the weight is to 0. Furthermore, its confidence weight is taken as the comprehensive confidence score W at point j. 置信 [i,j] (range [0,1]), when confidence weight ≈ 1 (i.e., high weight), confidence weight ≈ 0 (i.e., low weight); further calculate the total weight using the comprehensive weight formula, which is: W 总 [i,j]=W 空间 [i,j]*W 高程 [i,j]*W 置信 [i,j]; where W 总 [i,j] represents the total weight; W 空间 [i,j] represents the spatial weights; W 高程 [i,j] represents the elevation weight; W 置信 [i,j] represents the confidence weight; when both conditions are met, the space is close to (W). 空间 (High), similar in elevation (W) 高程 (for high quality) and data reliability (W 置信 Only by having a high score can one obtain a high total weight. The weighted smoothing process utilizes the aforementioned weight calculation steps to perform weighted smoothing on the elevation values ​​of neighboring points, resulting in optimized elevation values. The weighted smoothing formula is as follows:

[0050] Where H 高[i] represents the optimized elevation value of point i; N(i) represents the set of neighboring points of point i (obtained from the neighborhood analysis steps above); ∑ represents the summation result; W_total[i,j] represents the total weight (calculated from the total weight above); H 稳 [j] represents the stable elevation value of point j (obtained from the elevation weight calculation steps described above); This represents the weighted sum of the elevations of all points j in the neighborhood; This represents the sum of the comprehensive weights of all points within the neighborhood; the optimized elevation data is obtained through calculation using a formula. The formula for calculating the local consistency confidence level is as follows:

[0051] Where C 一致 [i] represents the consistency confidence level; |H 稳 [j]-H 高 [i]| represents the absolute value of the change before and after optimization; td 阀 This represents the consistency threshold (typically 0.02m); its judgment logic is as follows: when the change is 0, the confidence level is 1; when the change is td... 阀 →Confidence level 0, when the change > threshold → confidence level 0; the consistency confidence assessment result is obtained by calculating using this formula.

[0052] Working Principle: The intelligent sensor uses laser ranging to obtain raw distance data sequences through surveying. Data preprocessing techniques are used to perform initial processing on the raw data. A convolutional network is used to analyze the ground material of the laser beam at the laser irradiation point and to perform linear compensation for the influence of different materials. Real-time survey environment data is transmitted to the sensor through an external data interface, and a multi-factor coupling error model is used to refine the error correction of the data. An adaptive datum calibration algorithm is used to calculate the elevation and confidence level of the data. Finally, the elevation data is further optimized and confidence level fused. The optimized elevation data is then used as the final height survey data, which can be output through the intelligent sensor's data interface to complete the design.

[0053] Those skilled in the art will recognize that the modules, units, and algorithm implementation steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of electronic hardware and software. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

Claims

1. An intelligent sensor for highway design surveying ground height, characterized in that, Specifically comprising: The data acquisition and preprocessing module, the adaptive compensation module, the data optimization module, the height calculation module and the optimization output module are sequentially connected; The data acquisition and preprocessing module receives original distance signal data and performs preprocessing to obtain an initial distance sequence; The adaptive compensation module identifies the ground material of the light spot irradiation point according to the initial distance sequence, and compensates for the difference in reflection characteristics of different materials by using a lightweight convolutional neural network model, to obtain compensation data; The data optimization module introduces environmental data and performs fine error correction on the ranging value based on a multi-factor coupled real-time error model according to the compensation data, to obtain corrected data; The height calculation module converts the corrected data into absolute height based on an adaptive reference surface calibration algorithm of Kalman filtering, to obtain height calculation results and calibrate the reference surface drift in real time; The optimization output module performs height post-processing optimization and data integration on the height calculation results, and outputs the final height data through an interface.

2. The intelligent sensor for highway design surveying ground height according to claim 1, wherein, The data acquisition and preprocessing module includes the original distance signal of the laser ranging sensor, and performs data preprocessing on the signal, including filtering and abnormal point elimination.

3. The intelligent sensor for highway design surveying ground height according to claim 2, characterized in that, The adaptive compensation module includes a material identification submodule, a material reflection coefficient table construction submodule, and a reflection compensation submodule; The material identification submodule processes the echo waveform by using a lightweight one-dimensional convolutional neural network 1D-CNN, and the processing layer structure of the model includes an input layer→a convolutional layer 1→a pooling layer 1→a convolutional layer 2→a pooling layer 2→a flattening layer→a fully connected layer 1→an output layer. When training the model, a large amount of laser echo waveform data under known materials is collected in advance and labeled. The trained network model is embedded into the module to output the probability of each category. The category corresponding to the maximum probability is taken as the material type M, and the maximum probability value is recorded as the confidence P. The label includes [0 concrete, 1 asphalt, 2 grass, 3 soil]; The material reflection coefficient table construction submodule extracts the initial distance measurement value of the current measurement point, and according to the material type M and the confidence P, which is in the range of [0, 1], a material reflection coefficient table is established through experimental measurement in advance. The table stores the reflection coefficient R of each material under standard conditions; The reflection compensation submodule looks up the standard reflection coefficient R in the material reflection coefficient table according to the material type M, and uses a confidence-based linear adjustment compensation strategy. When P≧0.8, compensation is performed, P<0.8, the default material 0 is used, and when P=0, no compensation is performed. The compensation formula is: where D 补偿 represents the compensated accurate distance value; D 初始 represents the initial distance measurement value; k represents the calibration coefficient; P represents the confidence; R represents the standard reflection coefficient; represents the core compensation factor; [1 + (...)*P] represents the final linear scaling coefficient; the compensated accurate distance value is calculated through the formula and output.

4. The intelligent sensor for highway design surveying ground height according to claim 3, characterized in that, The data optimization module includes an environmental data calling submodule, a light speed change correction submodule, and a thermal deformation correction submodule; The environmental data calling submodule calls the environmental data during actual surveying through a data interface. The data includes temperature and humidity data and air pressure data, with dimensions of [T, H, P], where T represents temperature data, H represents humidity data, and P represents air pressure data; The light speed change correction sub-module calculates the current light speed according to the environment data in the actual survey and corrects the distance, and the light speed correction formula is: Wherein D 修 is the preliminary corrected distance; D 补偿 is the above-mentioned compensated accurate distance value; C 实 is the actual light speed under the current environment condition; C 参 is the reference light speed under the standard condition. The formula performs preliminary correction according to the actual environmental light speed; The thermal deformation correction submodule fits the relationship between temperature and deformation error through experimental data, and a thermal deformation error relationship calculation formula is: ΔD 热 =k1×(T-T0)+k2×(T-T0) 2 , wherein ΔD 热 represents a distance error caused by thermal deformation; T represents a temperature value of a current survey environment; T0 represents a reference temperature, which is set to 25 DEG C; k1 represents a linear thermal expansion coefficient; k2 represents a nonlinear thermal expansion coefficient; k1×(T-T0) represents a linear term, which describes linear expansion / contraction of a material with temperature change; and k2×(T-T0) 2 represents a quadratic term, which describes a nonlinear thermal effect.

5. The intelligent sensor for highway design surveying ground height according to claim 4, wherein, The data optimization module further comprises an air refraction change correction sub-module, a multi-factor coupling error model sub-module and a result output sub-module; The air refraction change correction sub-module utilizes Edlen to calculate the current air refraction n, and further utilizes a refraction error correction formula to calculate the optical path difference caused by the air refraction change, and converts the refraction change into an actual distance correction amount, and the refraction error correction formula is: ΔD 折 =D 修 ×(n-n0); wherein ΔD 折 represents the refraction error correction amount; D 修 represents the distance after the light speed correction; n represents the current survey environment air refraction; n0 represents the air refraction under the standard condition, and is taken as 1.000237. The multi-factor coupling error model sub-module synthesizes the correction results of the light speed change correction sub-module, the thermal deformation correction sub-module and the air refraction change correction sub-module, and considers the coupling effect between the correction results, to establish a comprehensive error model for calculating a total error correction distance; The result output sub-module checks the total error correction distance for reasonableness, including distance change range determination and abnormal value detection. When an abnormality occurs, the strategy is to use the effective correction value at time step t-1. Finally, the total correction distance and the environmental parameter confidence E 环 is output.

6. The intelligent sensor for highway design surveying ground height according to claim 5, wherein, The height calculation module comprises a coordinate conversion sub-module, a datum plane drift detection sub-module and a height post-processing sub-module; The coordinate conversion sub-module converts the data into an absolute height of the measuring point according to the total correction distance, the attitude and the absolute position data of the intelligent sensor, and obtains an original height sequence, wherein the dimension is [X, Y, Z]; The base surface drift detection submodule detects the base surface change based on the statistical characteristics of the continuous measurement points according to the original high program sequence, further calibrates the current value after estimating the current drift amount by using a Kalman filter, and finally obtains the calibrated height data recorded as H 漂 ; The elevation post-processing submodule smoothes its data according to elevation data H 漂 , using a weighted moving average filter based on confidence, where the weight of the current measurement is calculated as: , where W represents the weight of the current measurement; E 环 represents the confidence of the introduced environmental data, ranging from [0, 1]; H 漂 represents the bias estimate of the current elevation; H 漂max is the maximum elevation bias threshold, set to 0.1m; further weighted moving average, the formula is: , where H 平滑 represents the final elevation output after smoothing; W i represents the adaptive weight of the ith sampling point; H 漂i represents the calibrated elevation value for the i-th sampling point; The smooth elevation data is obtained by formula calculation, and the calibration confidence is further calculated, and the formula is: wherein E 校 represents the calibration confidence, ranging between [0, 1]; the rest are consistent with the weight calculation formula.

7. The intelligent sensor for highway design surveying ground height according to claim 6, wherein, The height calculation module further comprises a confidence calculation sub-module; The confidence calculation sub-module combines the environment confidence E 环 and the calibration confidence E 校 into an integrated confidence for overall evaluation of the reliability of the current measurement point, and the weighted formula is: C 综 =E 环 *0.4+E 校 *0.6, wherein C 综 represents the integrated confidence; E 环 represents the introduced environment data confidence, and the weight is set to 0.4; E 校 represents the calibration confidence, and the weight is set to 0.6, and the total is 1.

8. The intelligent sensor for highway design surveying ground height according to claim 7, characterized in that, The optimization output module comprises a consistency optimization sub-module, a consistency confidence sub-module and a final output sub-module; The local consistency optimization submodule smoothes the abnormal undulations while retaining true terrain features by using a local weighted regression algorithm LOESS based on adaptive weights according to the calibrated elevation data H 漂 to calculate the optimized elevation data. The consistency confidence submodule extracts the above confidence calculation submodule, and calculates the comprehensive confidence C 综 Further local consistency confidence calculation is performed, which is combined with the comprehensive confidence C 综 Further weighted fusion, and the weight distribution is comprehensive confidence C 综 0.7→local consistency confidence is 0.3, and the final overall confidence C 总 ; The final output sub-module takes the height data after the local consistency optimization as the final ground survey height data H 终 and the integrated confidence C 总 Output through the data interface of the intelligent sensor.

Images