A laser terrain intelligent matching method based on deep features and bayesian optimization

Abstract

The application provides a laser terrain intelligent matching method based on deep features and Bayesian optimization, comprising: acquiring measured terrain profile data, constructing multi-channel structured terrain data according to the elevation information of the terrain profile, and obtaining measured multi-channel terrain data; wherein the measured terrain profile data comprises initial trajectory information, elevation information of the measured terrain profile and geographic position information; inputting the measured multi-channel terrain data into a feature coding model to output a measured terrain feature vector; based on the measured terrain feature vector, performing a cyclic matching update based on a Bayesian optimizer, and outputting optimal transformation parameters of the measured terrain profile data and a reference digital elevation model when the cycle ends, and correcting the measured terrain profile data according to the optimal transformation parameters.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing data processing or terrain matching technology, specifically involving a laser terrain intelligent matching method based on depth features and Bayesian optimization. Background Technology

[0002] In applications such as terrain-aided navigation, high-precision positioning, and multi-source elevation data fusion, the core element lies in accurately matching measured terrain profile data with existing digital elevation models (DEMs). Traditional methods often employ template matching algorithms based on correlation analysis, such as the typical TERCOM or SITAN methods. These methods estimate the optimal matching location by calculating the correlation coefficient or mean square error between the measured profile and the reference terrain through a sliding search within an uncertainty region. However, with the development of high-precision sensor technologies such as lidar and the increasing complexity of application scenarios, the inherent technical limitations of these traditional methods are becoming increasingly apparent when processing modern high-dimensional, high-precision data.

[0003] Traditional methods rely on cost functions such as mean squared error (MSE) or normalized cross-correlation (NCC), which are essentially point-to-point numerical comparisons based on the original elevation values. This approach makes the matching process highly sensitive to various errors, resulting in poor robustness. Whether it's measurement noise from the sensor itself, errors in DEM data caused by land cover (such as vegetation), or nonlinear scale distortion of the trajectory profile caused by the accumulation of velocity errors in the inertial navigation system (INS), all directly affect the calculation of the cost function, leading to decreased matching accuracy, and even matching failure when terrain features are weak or interference is present.

[0004] A deeper problem lies in the fact that traditional matching models have a limited understanding of terrain, making them prone to getting trapped in local optima. These methods treat terrain merely as a one- or two-dimensional sequence of high-order parameters, lacking a deep understanding of the terrain's inherent geometric structure (such as slope and curvature variations) and global contextual relationships. Therefore, when the vehicle traverses areas with repetitive or ambiguous terrain features (such as continuous hills or ridges along a single direction), the cost function surface is prone to exhibiting multiple similar minima. In this case, the algorithm's results may converge to a mathematically plausible but geographically incorrect local optimum, leading to serious positioning errors.

[0005] Meanwhile, the exhaustive grid search strategy used in traditional methods suffers from severe efficiency bottlenecks in solving unknown parameters such as location and scale. The computational complexity of this strategy increases exponentially with the increase of the search range and parameter dimensions. Even with parallel computing, its brute-force search nature makes it difficult to cope with the need for rapid relocation under large-scale uncertainty, greatly limiting the engineering practicality of the algorithm. Summary of the Invention

[0006] To overcome the problems of poor robustness, susceptibility to local optima, and low matching search efficiency caused by sensitivity to noise and nonlinear distortion in existing terrain matching methods, this invention provides a laser terrain intelligent matching method based on deep features and Bayesian optimization. By introducing slope and curvature to construct multi-channel inputs and combining them with a feature encoding model, the method can learn the deep structure and semantic features of the terrain from the data. Simultaneously, a Bayesian optimization algorithm is used to intelligently search the multi-dimensional parameter space of location, scale, etc., and sampling is guided by a probabilistic model, enabling the method to quickly approximate the optimal matching result with fewer evaluation attempts than exhaustive search.

[0007] According to one aspect of the present invention, a laser terrain intelligent matching method based on depth features and Bayesian optimization is provided, comprising:

[0008] Acquire measured topographic profile data, construct multi-channel structured topographic data based on the elevation information of the topographic profile, and obtain measured multi-channel topographic data; wherein, the measured topographic profile data includes the initial trajectory, the elevation information of the measured topographic profile, and the geographical location information;

[0009] The measured multi-channel terrain data is input into the feature encoding model, and the measured terrain feature vector is output.

[0010] Based on the measured terrain feature vector, a loop matching update based on a Bayesian optimizer is performed. At the end of the loop, the optimal transformation parameters of the measured terrain profile data and the reference digital elevation model are output. The measured terrain profile data is then corrected based on the optimal transformation parameters.

[0011] The cyclic matching update based on the Bayesian optimizer includes the following steps:

[0012] Candidate parameter points are generated by a Bayesian optimizer in a preset parameter space; wherein the parameter space includes at least two-dimensional translation parameters and one-dimensional scaling parameters.

[0013] Perform an initial trajectory transformation based on the candidate parameter points to obtain a sequence of candidate geographic reference point coordinates;

[0014] Elevation information of candidate geographic reference point coordinate sequences is extracted from the reference digital elevation model, candidate multi-channel structured terrain data is constructed, and candidate terrain feature vectors are generated based on the feature coding model.

[0015] The measured terrain feature vector and the candidate terrain feature vector are input into the matching model, and the matching score is output.

[0016] The Bayesian optimizer is updated based on the candidate parameter points and their matching scores. Then, the next candidate parameter point is generated to start a new round of cyclic matching and updating. This process is repeated until the preset number of iterations or convergence conditions are met. The candidate parameter point with the highest matching score is then output as the optimal transformation parameter.

[0017] As a further technical solution, multi-channel structured terrain data includes at least three channels: elevation channel, slope channel, and curvature channel. The construction steps include:

[0018] For any elevation sampling point in the topographic profile, the slope of the sampling point is obtained by calculating the first derivative of the elevation using the numerical difference method, and the curvature of the sampling point is obtained by calculating the second derivative of the elevation.

[0019] The elevation, slope, and curvature components of all elevation sampling points on the terrain profile are stacked and normalized along the channel dimension to construct measured multi-channel structured terrain data.

[0020] As a further technical solution, the feature encoding model processes multi-channel terrain data and outputs terrain feature vectors, including a first feature extraction module for extracting local features and a second feature extraction module for modeling global features; wherein, the first feature extraction module is a convolutional neural network structure and the second feature extraction module is a Transformer structure based on a self-attention mechanism.

[0021] As a further technical solution, the Bayesian optimizer includes a Gaussian process surrogate model and a collection function: the Gaussian process surrogate model is used to receive all candidate parameter points and their corresponding matching scores in the experience pool, construct and output the posterior probability distribution of the matching scores in the preset parameter space; the collection function is used to calculate and recommend the next candidate parameter point to be evaluated based on the posterior probability distribution of the matching scores.

[0022] As a further technical solution, the matching model is a regression network based on a multilayer perceptron, which takes the fusion representation of the measured terrain feature vector and the candidate terrain feature vector as input, and outputs the matching score through nonlinear mapping.

[0023] As a further technical solution, an initial trajectory transformation is performed based on the candidate parameter points to obtain a sequence of candidate geographic reference point coordinates. The steps include:

[0024] The planar coordinates of each point on the initial trajectory are transformed using the parameter combinations corresponding to the candidate parameter points to obtain the candidate reference point coordinate sequence. Then, the candidate reference point coordinate sequence is transformed to the spatial geographic coordinate system to obtain the candidate reference point geographic coordinate sequence.

[0025] As a further technical solution, the steps for outputting the optimal transformation parameters include:

[0026] The candidate parameter points and their corresponding matching scores are added to the experience pool as a new set of observation data and fed back to the Bayesian optimizer. The Gaussian process surrogate model inside the Bayesian optimizer is updated, and then it is determined whether the preset maximum number of iterations or convergence conditions are met. If not, the next candidate parameter point is calculated based on the updated surrogate model and the acquisition function, and a new round of iteration is started. If the conditions are met, the candidate parameter point with the highest matching score is selected from the experience pool to obtain the optimal transformation parameters.

[0027] According to another aspect of this specification, a laser terrain intelligent matching system based on depth features and Bayesian optimization is provided, comprising:

[0028] The input and preprocessing module is used to acquire measured terrain profile data, construct multi-channel structured terrain data based on the elevation information of the terrain profile, and obtain measured multi-channel terrain data; wherein, the measured terrain profile data includes the initial trajectory, the elevation information of the measured terrain profile, and the geographical location information;

[0029] The feature extraction module is used to input measured multi-channel terrain data into the feature encoding model and output measured terrain feature vectors.

[0030] The intelligent matching module is used to perform a loop matching update based on the measured terrain feature vector and a Bayesian optimizer. After the loop ends, it outputs the optimal transformation parameters between the measured terrain profile data and the reference digital elevation model, and corrects the measured terrain profile data according to the optimal transformation parameters.

[0031] The cyclic matching update based on the Bayesian optimizer includes the following steps:

[0032] Candidate parameter points are generated by a Bayesian optimizer in a preset parameter space; wherein the parameter space includes at least two-dimensional translation parameters and one-dimensional scaling parameters.

[0033] Perform an initial trajectory transformation based on the candidate parameter points to obtain a sequence of candidate geographic reference point coordinates;

[0034] Elevation information of candidate geographic reference point coordinate sequences is extracted from the reference digital elevation model, candidate multi-channel structured terrain data is constructed, and candidate terrain feature vectors are generated based on the feature coding model.

[0035] The measured terrain feature vector and the candidate terrain feature vector are input into the matching model, the matching score is output, and the Bayesian optimizer is updated based on the candidate parameter points and their matching scores.

[0036] Repeat the iteration until the preset number of iterations or convergence condition is met, and output the candidate parameter point with the highest matching score as the optimal transformation parameter.

[0037] According to another aspect of this specification, an electronic device is provided, including a memory and a processor, the memory storing program instructions executed by the processor, the processor invoking the program instructions to perform a laser terrain intelligent matching method based on depth features and Bayesian optimization.

[0038] According to another aspect of this specification, a non-transitory computer-readable storage medium is provided, the non-transitory computer-readable storage medium storing computer instructions that cause the computer to execute a laser terrain intelligent matching method based on depth features and Bayesian optimization.

[0039] Compared with existing technologies, the advantages of this invention are as follows: By introducing slope and curvature to construct multi-channel input and employing a deep feature encoding model with local and global feature modeling capabilities, the deep structure and semantic features of the terrain can be learned from the data. This feature-learning-based matching cost function exhibits excellent robustness to sensor measurement noise, DEM errors, and nonlinear scale distortion caused by the accumulation of errors in the inertial navigation system. It can effectively avoid matching failures and local optima problems caused by blurred or repeated terrain features, thereby achieving higher positioning accuracy and reliability. Simultaneously, this invention uses a Bayesian optimization algorithm to intelligently search the multi-dimensional parameter space of position, scale, etc., and guides sampling with a probabilistic model. This allows for a rapid approximation of the optimal matching result with fewer evaluation attempts than exhaustive search, significantly improving search efficiency while maintaining accuracy. Furthermore, the matching model of this invention can also output an uncertainty estimate of the matching result. This information can be used to determine the reliability of the current matching result, providing a crucial dynamic weighting basis for the upper-level navigation information fusion system, further enhancing the stability and reliability of the entire system. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying 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.

[0041] Figure 1 A schematic diagram of a laser terrain intelligent matching process based on depth features and Bayesian optimization is provided for an embodiment of the present invention;

[0042] Figure 2 A schematic diagram of a laser terrain intelligent matching system based on depth features and Bayesian optimization is provided in an embodiment of the present invention.

[0043] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0044] It should be noted that:

[0045] The terms “comprising” and “having”, and any variations thereof, in the specification, claims, and accompanying drawings of this invention are intended to cover a non-exclusive inclusion, such as a process, method, system, product, or apparatus that includes a series of steps or units, not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0046] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices. The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be decomposed, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0047] 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, not all embodiments. 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. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined to form new technical solutions. Such combinations are not bound by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0048] like Figure 1 As shown, a laser terrain intelligent matching method based on deep features and Bayesian optimization includes the following steps:

[0049] Step 1: Obtain measured terrain profile data (measured laser terrain profile data), and construct multi-channel structured terrain data based on the elevation information of the terrain profile; wherein, the measured terrain profile data includes the initial trajectory and the elevation information and geographical location information of the measured terrain profile;

[0050] Step 2: Input the measured multi-channel terrain data into the feature coding model and output the measured terrain feature vector (measured feature vector).

[0051] Step 3: Based on the measured terrain feature vector, perform a loop matching update based on the Bayesian optimizer. After the loop ends, output the optimal transformation parameters of the measured terrain profile data and the reference digital elevation model, and correct the measured terrain profile data according to the optimal transformation parameters.

[0052] In step 1, optionally, the measured terrain profile data is collected by a lidar carried by a UAV or other carrier, and the initial trajectory is the original coordinate sequence with accumulated sensor error actually recorded when the sensor moves during the collection process.

[0053] The terrain profile consists of a series of discrete elevation points sampled along the initial trajectory. The terrain profile data includes, but is not limited to, the elevation data and spatial coordinate data of the discrete elevation points.

[0054] Furthermore, in step 1, the multi-channel structured terrain data includes at least elevation, slope, and curvature components, that is, the multi-channel structured terrain data includes at least three channels, namely the elevation channel, the slope channel, and the curvature channel.

[0055] The steps for constructing multi-channel structured terrain data based on the elevation information of terrain profiles include:

[0056] For any elevation sampling point in the topographic profile, the slope of the sampling point is obtained by calculating the first derivative of the elevation using the numerical difference method, and the curvature of the sampling point is obtained by calculating the second derivative of the elevation.

[0057] The elevation, slope, and curvature components of all elevation sampling points on the terrain profile are stacked and normalized along the channel dimension to construct measured multi-channel structured terrain data, which includes three channel features: elevation, slope, and curvature.

[0058] Specifically, for a terrain profile consisting of N sampling points, its mathematical representation is as follows:

[0059]

[0060] in, Function representing elevation values; This represents the set of elevation sampling point indices along the initial trajectory direction; Indicates the first Elevation values ​​of each elevation sampling point.

[0061] To further enhance the structural feature representation of the terrain profile, multi-channel transformation is performed, including:

[0062] Its first derivative (slope) and second derivative (curvature) are calculated using the numerical difference method, specifically as follows:

[0063]

[0064]

[0065] in, This represents the slope calculation function; Represents the curvature calculation function; This indicates the distance between adjacent sampling points.

[0066] The elevation, slope, and curvature components are stacked and normalized along the channel dimension to construct the measured multi-channel structured terrain data. Its dimensions are [C,L], where C=3 is the number of channels, L is the profile length, and M is the number of lines.

[0067]

[0068] In step 2, the feature encoding model processes multi-channel terrain data and outputs terrain feature vectors, including a first feature extraction module for extracting local features and a second feature extraction module for modeling global features; wherein, the first feature extraction module is a convolutional neural network (CNN) structure and the second feature extraction module is a Transformer structure based on a self-attention mechanism.

[0069] The feature encoding model is a nonlinear mapping function whose model parameters are obtained during the offline training phase. It performs deep feature extraction on multi-channel structured terrain data and outputs measured terrain feature vectors.

[0070] Specifically, the measured multi-channel structured terrain data obtained in step 1 is input into the feature coding model, resulting in a measured terrain feature vector. This vector can be considered as a global representation of the measured terrain profile structure and semantics, mathematically expressed as:

[0071]

[0072] In the formula, Represents a feature encoding model; Represents the measured terrain feature vector; This represents the parameters of the feature encoding model.

[0073] In step 3, the cyclic matching update based on the Bayesian optimizer includes the following steps:

[0074] Step 3-1: The Bayesian optimizer generates candidate parameter points in a preset parameter space; wherein the parameter space includes at least two-dimensional translation parameters and one-dimensional scaling parameters.

[0075] Step 3-2: Perform initial trajectory transformation based on the candidate parameter points to obtain the coordinate sequence of candidate geographic reference points;

[0076] Step 3-3: Extract elevation information of candidate geographic reference point coordinate sequences from the reference digital elevation model (DEM), construct candidate multi-channel structured terrain data, and generate candidate terrain feature vectors (candidate feature vectors) based on the feature coding model.

[0077] Steps 3-4: Input the measured terrain feature vector and the candidate terrain feature vector into the matching model, output the matching score, and update the Bayesian optimizer based on the candidate parameter points and their matching scores.

[0078] Step 3-5: Repeat steps 3-1 to 3-4 for iteration until the preset number of iterations or convergence conditions are met. Then, output the candidate parameter point with the highest matching score as the optimal transformation parameter.

[0079] In step 3, the Bayesian optimizer includes a Gaussian process surrogate model and a collection function: the Gaussian process surrogate model is used to receive all candidate parameter points in the experience pool and their corresponding matching scores, construct and output the posterior probability distribution of the matching scores in the preset parameter space; the collection function is used to calculate and recommend the next candidate parameter point to be evaluated based on the posterior probability distribution of the matching scores.

[0080] The posterior probability distribution of the matching score includes the posterior mean and the posterior variance. The posterior mean represents the best estimate of the expected value of the matching score by the surrogate model, and the posterior variance quantifies the confidence of the surrogate model in the expected value.

[0081] In step 3, in the preset parameter space, the two-dimensional translation parameter and the one-dimensional scale parameter are used to describe the geometric transformation of the reference trajectory: the incremental displacement vector between adjacent points on the reference trajectory is scaled based on the scale parameter, and the scaled trajectory is translated based on the translation parameter.

[0082] Specifically, in step 3-1, at the start of the cyclic matching update, a Bayesian optimizer is used to perform an intelligent search on the preset parameter space. Within the function, the optimizer generates a candidate parameter point based on the acquisition function. Mathematically, this is expressed as:

[0083]

[0084]

[0085] In the formula, For the preset parameter space, This represents the translation along the horizontal and vertical axes of a two-dimensional plane. This represents a one-dimensional scaling factor; This represents a candidate parameter point, which is an element in a predefined parameter space. Index it.

[0086] In step 3-2, an initial trajectory transformation is performed based on the candidate parameter points to obtain the coordinate sequence of candidate geographic reference points. The steps include:

[0087] The planar coordinates of each point on the initial trajectory are transformed using the parameter combinations corresponding to the candidate parameter points to obtain the candidate reference point coordinate sequence. Then, the candidate reference point coordinate sequence is transformed to the spatial geographic coordinate system to obtain the candidate reference point geographic coordinate sequence.

[0088] Specifically, step 3-2 uses the candidate parameter points generated in step 3-1. The initial trajectory is subjected to spatial geometric transformation to obtain a sequence of candidate reference point coordinates. The steps include:

[0089] Let the initial inaccurate planar coordinates of any point k on the original trajectory be... ,in Represent the initial plane coordinates of point k, using candidate parameter points. Coordinates of the transformed candidate reference point Mathematically, this is expressed as:

[0090]

[0091] The coordinate transformation results of all candidate reference points on the original initial trajectory are converted back to spatial geographic coordinates to obtain a set of candidate geographic reference point coordinate sequences.

[0092] In step 3-3, based on the coordinate sequence of candidate geographic reference points, the elevation information (i.e., elevation value) of the corresponding initial trajectory point position is extracted from the reference digital elevation model by interpolation, forming a candidate elevation profile.

[0093] Specifically, for any point in the geographic reference point coordinate sequence The mathematical representation of the steps for extracting its elevation value is as follows:

[0094]

[0095] In the formula, This represents the elevation value of any point k on the extracted original initial trajectory; This indicates a reference digital elevation model.

[0096] Subsequently, using the same method as in step 1, slope and curvature were calculated, and multi-channel transformation was performed on the candidate elevation profiles to construct candidate multi-channel structured terrain data. .

[0097] Candidate multichannel data Input the same feature encoding model as in step 2. In the process, candidate terrain feature vectors are obtained. :

[0098]

[0099] In steps 3-4, the matching model is a regression network based on multilayer perceptron (MLP), which takes the fused representation of the measured terrain feature vector and the candidate terrain feature vector as input and outputs the matching score through nonlinear mapping.

[0100] Preferably, in steps 3-4, the matching model also outputs an uncertainty estimate of the matching score, which is used to determine the reliability of the current matching result.

[0101] The uncertainty estimation is the standard deviation of the prediction results obtained by applying Monte Carlo Dropout (MCDropout) technology to the matching model and performing multiple forward propagation calculations during the inference phase.

[0102] Specifically, the fused representation of the measured terrain feature vector and the candidate terrain feature vector is input into the matching model with Dropout enabled, and several forward propagations are performed. Each forward propagation outputs a matching score. The matching scores from each forward propagation are combined to obtain the matching score set, and the standard deviation of the matching score set is calculated as the uncertainty estimate of the matching score.

[0103] Optionally, the fusion representation of the measured terrain feature vector and the candidate terrain feature vector can be obtained by means of feature concatenation or feature subtraction.

[0104] Specifically, the measured terrain feature vector With candidate terrain feature vectors Common inputs to a matching model The function outputs a similarity score in the interval [0,1]. And optional uncertainty estimates for this forecast. Mathematically, this is expressed as:

[0105]

[0106] In steps 3-5, the candidate parameter points obtained in step 3-4 are... and their corresponding matching scores As a new set of observation data The parameter is added to the experience pool and fed back to the Bayesian optimizer. The Gaussian process surrogate model inside the Bayesian optimizer is updated. Then, it is determined whether the preset maximum number of iterations or convergence conditions are met. If not, the next candidate parameter point is calculated based on the updated surrogate model and the acquisition function, and the process returns to step 3-1 to start a new round of iteration. Compared with the traditional exhaustive search, this intelligent sampling strategy can explore the parameter space with fewer evaluations. If the conditions are met, the candidate parameter point with the highest matching score is selected from the experience pool to obtain the optimal transformation parameter.

[0107] Specifically, in the experience pool, the subsequent parameter point that maximizes the matching score S is selected to obtain the optimal transformation parameters, which can be mathematically represented as:

[0108]

[0109] In the formula, This represents the optimal parameter point, which is an element in the preset parameter space; This represents the optimal translation along the horizontal and vertical axes of a two-dimensional plane. This represents the optimal one-dimensional scaling factor.

[0110] Based on the optimal transformation parameters, the geographic location information of the original initial trajectory is finally translated and scaled, outputting the corrected high-precision geographic location information. For any point k on the original initial trajectory, the corrected high-precision geographic location information is... Mathematically represented as:

[0111]

[0112] In addition, the framework of the laser terrain intelligent matching method based on deep features and Bayesian optimization proposed in this invention also includes an offline training stage. Optionally, a self-supervised contrastive learning approach is adopted, which constructs positive and negative sample pairs and uses loss functions such as binary cross-entropy loss or triplet loss to perform end-to-end optimization training on the feature encoding model and the matching model until the model performance converges on the validation set.

[0113] The implementation of the various embodiments of the present invention is based on programmed processing through a system with processor functionality. Therefore, in practical engineering, the technical solutions and functions of the various embodiments of the present invention are encapsulated into various modules. Based on this reality, and building upon the above embodiments, the embodiments of the present invention provide a laser terrain intelligent matching system based on depth features and Bayesian optimization. This system is used to execute a laser terrain intelligent matching method based on depth features and Bayesian optimization from the above method embodiments.

[0114] See Figure 2 The system includes:

[0115] The input and preprocessing module is used to acquire measured terrain profile data, construct multi-channel structured terrain data based on the elevation information of the terrain profile, and obtain measured multi-channel terrain data; wherein, the measured terrain profile data includes the initial trajectory, the elevation information of the measured terrain profile, and the geographical location information;

[0116] The feature extraction module is used to input measured multi-channel terrain data into the feature encoding model and output measured terrain feature vectors.

[0117] The intelligent matching module is used to perform a loop matching update based on the measured terrain feature vector and a Bayesian optimizer. After the loop ends, it outputs the optimal transformation parameters between the measured terrain profile data and the reference digital elevation model, and corrects the measured terrain profile data according to the optimal transformation parameters.

[0118] The cyclic matching update based on the Bayesian optimizer includes the following steps:

[0119] Candidate parameter points are generated by a Bayesian optimizer in a preset parameter space; wherein the parameter space includes at least two-dimensional translation parameters and one-dimensional scaling parameters.

[0120] Perform an initial trajectory transformation based on the candidate parameter points to obtain a sequence of candidate geographic reference point coordinates;

[0121] Elevation information of candidate geographic reference point coordinate sequences is extracted from the reference digital elevation model, candidate multi-channel structured terrain data is constructed, and candidate terrain feature vectors are generated based on the feature coding model.

[0122] The measured terrain feature vector and the candidate terrain feature vector are input into the matching model, the matching is output, and the Bayesian optimizer is updated based on the candidate parameter points and their matching scores.

[0123] Repeat the iteration until the preset number of iterations or convergence condition is met, and output the candidate parameter point with the highest matching score as the optimal transformation parameter.

[0124] It should be noted that the system embodiments provided by the present invention are used not only to implement the methods in the above method embodiments, but also to implement the methods in other method embodiments provided by the present invention. The only difference is that corresponding functional modules are set. The principle is basically the same as that of the above system embodiments provided by the present invention. As long as those skilled in the art can improve the modules in the above system embodiments by referring to the specific technical solutions in other method embodiments and combining technical features to obtain corresponding technical means and technical solutions composed of these technical means, on the basis of the above system embodiments, and on the premise of ensuring the practicality of the technical solutions, they can obtain corresponding system-class embodiments for implementing the methods in other method-class embodiments.

[0125] The method in this embodiment of the invention is implemented using an electronic device; therefore, it is necessary to introduce the relevant electronic device. For this purpose, embodiments of the present invention provide an electronic device, such as... Figure 3 As shown, the electronic device includes: at least one processor, a communications interface, at least one memory, and a communications bus, wherein the at least one processor, the communications interface, and the at least one memory communicate with each other via the communications bus. The at least one processor invokes logical instructions stored in the at least one memory to execute all or part of the steps of the methods provided in the foregoing method embodiments.

[0126] Furthermore, when the logical instructions in at least one of the aforementioned memories are implemented as software functional units and sold or used as independent products, they are stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, is embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (a personal computer, server, or network device) to execute all or part of the steps of the methods described in the various method embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks—various media for storing program code.

[0127] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, located in one place, or distributed across multiple network units. The purpose of this embodiment is achieved by selecting some or all of the modules according to actual needs. Those skilled in the art will understand and implement this without any inventive effort.

[0128] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0129] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0130] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0131] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0132] Based on the same technical concept as the foregoing embodiments, the present invention provides a non-transitory computer-readable storage medium that stores computer instructions that cause the computer to execute a laser terrain intelligent matching method based on depth features and Bayesian optimization.

[0133] 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 therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A laser terrain intelligent matching method based on deep features and Bayesian optimization, characterized in that, include: Acquire measured topographic profile data, and construct multi-channel structured topographic data based on the elevation information of the topographic profile to obtain measured multi-channel topographic data. The measured topographic profile data includes an initial trajectory, elevation information, and geographic location information of the measured topographic profile. The multi-channel structured topographic data includes at least three channels: an elevation channel, a slope channel, and a curvature channel. The construction steps include: for any elevation sampling point in the topographic profile, calculating the first derivative of the elevation to obtain the slope of the sampling point using the numerical difference method, and calculating the second derivative of the elevation to obtain the curvature of the sampling point; stacking and normalizing the elevation, slope, and curvature components of all elevation sampling points on the topographic profile along the channel dimension to construct measured multi-channel structured topographic data. The measured multi-channel terrain data is input into the feature encoding model, and the measured terrain feature vector is output. The feature encoding model processes the multi-channel terrain data and outputs the terrain feature vector, including a first feature extraction module for extracting local features and a second feature extraction module for modeling global features. The first feature extraction module is a convolutional neural network structure, and the second feature extraction module is a Transformer structure based on a self-attention mechanism. Based on the measured terrain feature vectors, a loop matching update based on a Bayesian optimizer is performed. At the end of the loop, the optimal transformation parameters between the measured terrain profile data and the reference digital elevation model are output. The measured terrain profile data is then corrected based on these optimal transformation parameters. The Bayesian optimizer includes a Gaussian process surrogate model and a data acquisition function: the Gaussian process surrogate model receives all candidate parameter points and their corresponding matching scores from the experience pool, constructs and outputs the posterior probability distribution of the matching scores in the preset parameter space; the data acquisition function calculates and recommends the next candidate parameter point to be evaluated based on the posterior probability distribution of the matching scores. The cyclic matching update based on the Bayesian optimizer includes the following steps: Candidate parameter points are generated by a Bayesian optimizer in a preset parameter space; wherein the parameter space includes at least two-dimensional translation parameters and one-dimensional scaling parameters. An initial trajectory transformation is performed based on the candidate parameter points to obtain a sequence of candidate geographic reference point coordinates; Elevation information of candidate geographic reference point coordinate sequences is extracted from the reference digital elevation model, candidate multi-channel structured terrain data is constructed, and candidate terrain feature vectors are generated based on the feature coding model. The measured terrain feature vector and the candidate terrain feature vector are input into the matching model, the matching score is output, and the Bayesian optimizer is updated based on the candidate parameter points and their matching scores. Repeat the iteration until the preset number of iterations or convergence condition is met, and output the candidate parameter point with the highest matching score as the optimal transformation parameter.

2. The laser terrain intelligent matching method based on depth features and Bayesian optimization as described in claim 1, characterized in that, The matching model is a regression network based on a multilayer perceptron. It takes the fusion representation of the measured terrain feature vector and the candidate terrain feature vector as input and outputs the matching score through nonlinear mapping.

3. The laser terrain intelligent matching method based on depth features and Bayesian optimization as described in claim 1, characterized in that, The initial trajectory transformation is performed based on the candidate parameter points to obtain the coordinate sequence of candidate geographic reference points. The steps include: The planar coordinates of each point on the initial trajectory are transformed using the parameter combinations corresponding to the candidate parameter points to obtain the candidate reference point coordinate sequence. Then, the candidate reference point coordinate sequence is transformed to the spatial geographic coordinate system to obtain the candidate reference point geographic coordinate sequence.

4. The laser terrain intelligent matching method based on depth features and Bayesian optimization as described in claim 1, characterized in that, The steps for outputting the optimal transformation parameters include: The candidate parameter points and their corresponding matching scores are added to the experience pool as a new set of observation data and fed back to the Bayesian optimizer. The Gaussian process surrogate model inside the Bayesian optimizer is updated, and then it is determined whether the preset maximum number of iterations or convergence conditions are met. If not, the next candidate parameter point is calculated based on the updated surrogate model and the acquisition function, and a new round of iteration is started. If the conditions are met, the candidate parameter point with the highest matching score is selected from the experience pool to obtain the optimal transformation parameters.

5. A laser terrain intelligent matching system based on deep features and Bayesian optimization, characterized in that, include: The input and preprocessing module is used to acquire measured topographic profile data, construct multi-channel structured topographic data based on the elevation information of the topographic profile, and obtain measured multi-channel topographic data. The measured topographic profile data includes an initial trajectory, elevation information, and geographic location information of the measured topographic profile. The multi-channel structured topographic data includes at least three channels: an elevation channel, a slope channel, and a curvature channel. The construction steps include: for any elevation sampling point in the topographic profile, calculating the first derivative of the elevation to obtain the slope of the sampling point using the numerical difference method, and calculating the second derivative of the elevation to obtain the curvature of the sampling point; stacking and normalizing the elevation, slope, and curvature components of all elevation sampling points on the topographic profile along the channel dimension to construct measured multi-channel structured topographic data. The feature extraction module is used to input measured multi-channel terrain data into the feature encoding model and output measured terrain feature vectors. The feature encoding model processes multi-channel terrain data and outputs terrain feature vectors, including a first feature extraction module for extracting local features and a second feature extraction module for modeling global features. The first feature extraction module is a convolutional neural network structure, and the second feature extraction module is a Transformer structure based on a self-attention mechanism. The intelligent matching module is used to perform a loop matching update based on the measured terrain feature vector and a Bayesian optimizer. At the end of the loop, it outputs the optimal transformation parameters between the measured terrain profile data and the reference digital elevation model, and corrects the measured terrain profile data based on the optimal transformation parameters. The Bayesian optimizer includes a Gaussian process surrogate model and a data acquisition function: the Gaussian process surrogate model receives all candidate parameter points and their corresponding matching scores from the experience pool, constructs and outputs the posterior probability distribution of the matching scores in the preset parameter space; the data acquisition function calculates and recommends the next candidate parameter point to be evaluated based on the posterior probability distribution of the matching scores. The cyclic matching update based on the Bayesian optimizer includes the following steps: Candidate parameter points are generated by a Bayesian optimizer in a preset parameter space; wherein the parameter space includes at least two-dimensional translation parameters and one-dimensional scaling parameters. An initial trajectory transformation is performed based on the candidate parameter points to obtain a sequence of candidate geographic reference point coordinates; Elevation information of candidate geographic reference point coordinate sequences is extracted from the reference digital elevation model, candidate multi-channel structured terrain data is constructed, and candidate terrain feature vectors are generated based on the feature coding model. The measured terrain feature vector and the candidate terrain feature vector are input into the matching model, the matching score is output, and the Bayesian optimizer is updated based on the candidate parameter points and their matching scores. Repeat the iteration until the preset number of iterations or convergence condition is met, and output the candidate parameter point with the highest matching score as the optimal transformation parameter.

6. An electronic device, characterized in that, The system includes a memory and a processor, wherein the memory stores program instructions that are executed by the processor, and the processor invokes the program instructions to execute a laser terrain intelligent matching method based on depth features and Bayesian optimization as described in any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions that cause the computer to execute the laser terrain intelligent matching method based on depth features and Bayesian optimization as described in any one of claims 1 to 4.

Images