Method for three-dimensional reconstruction of material surface of mine shovel based on gabor wave periodic activation function neural network

Abstract

This invention relates to a method for 3D reconstruction of the material surface in electric shovels based on Gabor wavelet periodic activation function neural networks, belonging to the field of intelligent operation and industrial 3D vision technology in mining machinery. Multimodal data of the material pile is simultaneously acquired by a binocular camera and a LiDAR mounted on the shovel boom. After point cloud denoising and coordinate normalization preprocessing, an initial 3D point cloud is formed. Subsequently, a Gabor wavelet network with periodic activation functions is used as input to learn a continuous, differentiable implicit symbolic distance field. This invention can faithfully fit complex geometric surfaces of the material surface containing rich high-frequency details. It can quickly and faithfully reconstruct a 3D model of the material surface with rich details from sparse and noisy point cloud data collected on-site by the electric shovel, directly serving the key problem of optimal excavation trajectory planning. This achieves a closed-loop process from perception to planning for the electric shovel, significantly improving the intelligence level and operational efficiency of electric shovel operations.

Description

Technical Field

[0001] This invention belongs to the field of intelligent operation of mining machinery and industrial three-dimensional vision technology, and relates to a method for three-dimensional reconstruction of the material surface of a mining electric shovel based on Gabor wavelet periodic activation function neural network. Specifically, it relates to a method for three-dimensional reconstruction of the material surface of a mining electric shovel by introducing a Gabor wavelet periodic activation function neural network. Background Technology

[0002] As a core piece of equipment in open-pit mining, the level of intelligence in electric shovels directly affects the mine's production efficiency and safety. In an automated electric shovel system, rapid and high-precision 3D reconstruction of the material pile (i.e., the material surface) is a crucial prerequisite for trajectory planning, bucket capacity control, and energy consumption optimization. However, the harsh operating environment of electric shovels (high dust levels and strong vibrations) often results in noisy, unevenly dense, or even severely missing point cloud data of the material surface collected by sensors such as lidar. This poses a significant challenge to traditional material surface modeling methods.

[0003] In recent years, Implicit Neural Representation (INR) has shown potential in the field of 3D reconstruction as a continuous function approximation framework. INR learns the continuous representation of signals through a multilayer perceptron (MLP), avoiding gridding constraints. However, existing INR activation functions (such as ReLU and sine functions) have a trade-off between accuracy and robustness: for example, while sine functions (such as SIREN) have high representation capacity, they are prone to overfitting noise; Gaussian functions are robust to noise but have limited approximation capabilities. This leads to artifacts, blurring, or loss of detail in the reconstructed material surface in the complex environment of electric shovel operations (such as changes in lighting and dust interference).

[0004] Wavelet analysis has attracted widespread attention in signal processing due to its optimal space-frequency localization properties, but it has not yet been applied to the reconstruction of material surfaces in electric shovels. Gabor wavelet-based INR performs excellently in tasks such as image reconstruction and denoising, providing the technical basis for this invention.

[0005] In industrial settings such as metallurgy and mining, 3D reconstruction of material surfaces (e.g., ore piles, furnace charge surfaces) is a crucial step in optimizing operational efficiency and achieving automated control. However, existing material surface reconstruction methods (such as polynomial response surface models, radial basis function interpolation, or traditional triangular meshing techniques) have significant limitations when dealing with sparse, noisy point cloud data collected in industrial settings: First, traditional methods are insufficient in capturing high-frequency geometric features (e.g., steep edges, local concavities and convexities), resulting in overly smooth reconstructed surfaces and severe loss of detail; second, they are highly dependent on data quality and integrity, easily exhibiting distortions such as holes and breaks under point cloud defects or noise interference; third, it is difficult to balance computational efficiency and reconstruction accuracy, making it hard to meet real-time control requirements. Especially for automated operations of heavy equipment such as electric shovels, real-time and accurate perception of the material surface morphology is required to plan the excavation trajectory, but existing technologies cannot achieve high-fidelity reconstruction under harsh working conditions (point cloud defects caused by dust and vibration).

[0006] Mine surface reconstruction has been extensively studied. Liu Yu et al. proposed a method based on semantic segmentation networks to determine semantic masks from information data and to perform cross-modal interaction from images to point clouds (Chinese Invention Patent CN121353264A). However, the core drawback of this method lies in the limitations of feature extraction and alignment processes, and it places extremely high demands on sensor accuracy, computational efficiency, environmental robustness, and architectural flexibility. For irregular excavation faces, it is difficult to achieve accurate 3D reconstruction using this method. This invention overcomes the bottlenecks of traditional methods in detail preservation, noise resistance, and real-time performance by introducing a SIREN periodic activation function neural network to a 3D reconstruction method for mine electric shovels, thus solving the problem of high-precision and high-efficiency surface reconstruction under sparse noisy point clouds. Summary of the Invention

[0007] To address the problems of existing technologies, this invention provides a method for 3D reconstruction of the material surface in mining electric shovels based on Gabor wavelet periodic activation function neural networks. Specifically, it is a real-time, high-precision reconstruction method for the material surface in electric shovel operations based on Gabor wavelet neural networks. This invention simultaneously acquires multimodal data of the material pile using a binocular camera and a lidar mounted on the shovel boom. After point cloud denoising and coordinate normalization preprocessing, an initial 3D point cloud is formed. Subsequently, using a Gabor wavelet network with periodic activation functions, the sparse point cloud is used as input to learn a continuous, differentiable implicit symbolic distance field. This allows for high-fidelity fitting of the complex geometric surface of the material surface, containing rich high-frequency details. This method can quickly and accurately reconstruct a richly detailed 3D model of the material surface from sparse, noisy point cloud data collected on-site by the electric shovel, directly serving the key problem of optimal excavation trajectory planning. This achieves a closed-loop process from perception to planning for the electric shovel, significantly improving the intelligence level and operational efficiency of electric shovel operations.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The method for 3D reconstruction of the material surface of a mining electric shovel based on Gabor wavelet periodic activation function neural network is a high-fidelity parameterized representation of the geometric morphology of the material surface based on the unique periodic activation function of Gabor wavelet neural network, including the following steps: The first step is environmental perception, which involves collecting geographic information as raw point cloud data. Using binocular cameras and LiDAR, comprehensive geographic information of the excavation site is collected, including topographic and geomorphological environmental data, as raw data point cloud.

[0009] The second step is to preprocess the raw point cloud data collected by the binocular camera and LiDAR. The preprocessing includes point cloud denoising based on statistical filtering and radius filtering methods, and finally obtains clean point cloud data G with significantly reduced noise, preserved key geometric features and optimized data volume.

[0010] Step 2.1, for each point in the raw point cloud data collected by the binocular camera and LiDAR... Calculate the nearest Euclidean distance. The average distance of each neighboring point , The nearest neighboring point is the target point. One point. For The value of is first determined from empirical values. =50 and observe the filtering effect, then perform adaptive fine-tuning, with the optimal fine-tuning range being [30, 70]. Combined with the visualization effect, determine the best value. The average distance is obtained using formula (1): (1) in, This means The nth "nearest neighbor" point There are points, among which =1, 2, ..., traverse all neighboring points; Then, using the average distance Calculate the mean of the average distances of all points in the raw point cloud data acquired by the binocular camera and LiDAR. and standard deviation As shown in formula (2): (2) Where N represents the total number of original data point clouds involved in the "average distance" calculation; i is the "index variable" for the summation process, used to iterate through each of these N data points. μ represents all... The arithmetic mean of the point cloud reflects the average local density of the point cloud. for The standard deviation measures the degree of local density dispersion. Calculate the standard deviation for each point in the original point cloud data. To its The average distance between the nearest neighbors Then seek all The mean μ and standard deviation .

[0011] Subsequently, if satisfy If a point is identified as an outlier and is removed, then it is considered an outlier and is removed. Let be a positive scaling factor. =2, used to adjust the strictness of noise judgment.

[0012] After removing outliers, the intermediate point cloud is obtained. The total number of points in the intermediate point cloud at this point is denoted as . As shown in formula (3): The denoising method provided in this step can effectively filter out random "flying points" in the original data point cloud, but it may not be effective for point clouds with non-uniform distribution, so further denoising is required.

[0013] Step 2.2, after removing outlier noise points in step 2.1, adjust the intermediate point cloud based on local density. Each point in Count the number of neighborhood points within a sphere centered at the given point and with a radius of r = 0.5 m. , including This helps avoid misjudgments of empty neighborhoods. Neighborhood points within the sphere Neighborhood points within each sphere Then, construct a sphere with radius r = 0.5m and count the number of its secondary neighbor points. ,common indivual value.

[0014] Calculate the local mean of the number of secondary neighbors and standard deviation As shown in formula (4): (4) in, This is the local mean of the number of secondary neighbors within the neighborhood. The standard deviation of the number of secondary neighbors within the neighborhood. This is an index variable. It represents the index at point... Within the neighborhood of a sphere centered at r = 0.5m, the first... Neighboring points, For the point The total number of points contained within a sphere centered at a radius of r = 0.5m is given by... Each specific point within the neighborhood (i.e., the first) After taking the neighborhood points as the center, draw a sphere with a radius of r = 0.5m. This represents the number of points inside the sphere.

[0015] Set dynamic threshold ,in This is the adjustment coefficient, let it be here. =2. If Less than the set threshold Then determine Noise points are removed; if Less than the set threshold Then determine These are noise points and should be removed. Not less than the set threshold If a point is found to be a neighboring point, then that point is retained as a neighboring point. The point cloud after retaining the neighboring points is... , Point cloud after preserving neighboring points Points are recorded as ,and .

[0016] Step 2.3: Process the point cloud obtained in Step 2.2 after preserving the neighborhood points. Coordinate normalization is performed to map the data to a normalized space Ω, generating pure point cloud data G. This pure point cloud data provides stable and efficient input for the Gabor wavelet neural network. The specific steps are as follows: First, the point cloud obtained in step 2.2 after retaining neighborhood points is transformed using translation centering. The center of the network is moved to the origin to eliminate the influence of the absolute position on network training, as shown in formulas (5) and (6): (5) in, To preserve the point cloud of neighboring points The coordinates of the centroid; for The Middle The original coordinates of each point; The coordinates are those of the centroid. for Points.

[0017] Centralized coordinates obtained by translating a point cloud for: (6) in, The point cloud obtained in step 2.2 after preserving neighborhood points No. The coordinates of each point after centering; These are the centered coordinate components.

[0018] Then, the point cloud obtained in step 2.2 after retaining the neighborhood points is calculated using formula (7). The maximum distance from all points to the origin is the maximum norm. The point cloud obtained in step 2.2 after preserving neighborhood points is then processed. Divide all point coordinates by this maximum value to obtain the point cloud after retaining neighborhood points obtained in step 2.2. The entire scale is adjusted to the range [-1, 1], as shown in formula (8): (7) (8) in, Point cloud with maximum norm and neighborhood points preserved. The maximum distance from all points in the set to the origin; for Coordinate components; For the first The normalized coordinates of each point. These are the normalized coordinate components, ranging from [-1, 1].

[0019] In step 2.3, the point cloud obtained in step 2.2 after retaining the neighborhood points... All are scaled to [-1, 1], therefore the canonical space is defined as Ω = Ω represents the gauge space. All coordinate vectors obtained after coordinate normalization... The set is the pure point cloud data G, which not only conforms to the periodicity of the Gabor wavelet activation function, but also avoids training difficulties caused by the input value being too large or too small.

[0020] The third step involves training a Gabor wavelet network to transform the preprocessed, clean point cloud data from the second step into a continuous, differentiable, and high-fidelity geometric model of the material surface. This invention employs the signed distance function (SDF) as an implicit representation of the scene and leverages the superior characteristics of the Gabor wavelet network to learn and parameterize this SDF field. Specifically: Step 3.1: Use the Gabor network model as the activation function to replace the traditional ReLU or TanH to accurately model the high-frequency details of the material surface.

[0021] Step 3.1.1: Location encoding; The clean point cloud data obtained after normalization in step 2.3 is mapped to the high-frequency feature space by the position encoding formula shown in formula (9); (9) in, B is the three-dimensional coordinate vector of a single point in the clean point cloud data obtained after normalization in step 2.3; B is a 3×m frequency band matrix, with 3 rows corresponding to the x, y, and z axes and 8 columns corresponding to 8 frequency bands. Typically, m=8 is set to cover the geometric frequency range from 0.1 to 10Hz. It is the encoded high-dimensional feature vector, representing the input coordinates. The high-dimensional feature representation after trigonometric function transformation has a dimension of 2×3×m. It is composed of the sine and cosine function values ​​of the product of each coordinate component and the frequency band matrix B. It is used as the input of Gabor neural network to enhance the perception of high-frequency details (such as cracks and textures) on the material surface and improve the accuracy of geometric reconstruction.

[0022] Step 3.1.2: Construct the network forward propagation and Gabor activation function; The network structure is a fully connected feedforward network (8 layers), with a 48-dimensional input layer (corresponding to...). The number of neurons in the hidden layer decreases (256→128→...→32), and the output layer is 1-dimensional. A Gabor activation function is constructed to replace the traditional ReLU or TanH function. Input is applied to the neurons in the l-th layer. As shown in formula (10): (10) Activate the output, as shown in formula (11): (11) in, Use the Gabor wavelet activation function; and These are weights and biases, respectively. This is the output feature vector of the (l-1)th layer (i.e., the input data of the lth layer); It is the center frequency, initially sampled randomly from a uniform distribution U(0, π). This is the bandwidth, initialized to a fixed constant, which is 1.0 here. This represents the phase shift and is initialized to 0. These parameters are given reasonable initial values ​​and then automatically learned and updated by the optimizer during training. This represents the output vector of the l-th layer, and the final output layer. That is, prediction .

[0023] Step 3.2: Construct a pseudo-SDF-based loss function to supervise the SDF values ​​predicted in step 3.1.

[0024] Since accurate ground truth SDF values ​​cannot be obtained in advance at the excavation site, this embodiment adopts a view-dependent pseudo-SDF supervision strategy. Using the real-time pose of the electric shovel bucket as a spatial reference frame, and combining it with point cloud normal information, a self-supervised loss function is constructed to guide the Gabor network in learning a physically intuitive signed distance field.

[0025] Step 3.2.1, Spatial Sampling and Nearest Neighbor Distance Calculation The gauge space Ω obtained in step 2.3 (i.e. Within the cubic region, a set of detached query points Q is randomly sampled using a stratified sampling strategy. For each detached query point q∈Q, the nearest Euclidean distance d(q,G) to the pure point cloud G is calculated, as shown in formula (12): (12) Where q is any three-dimensional free query point sampled in the canonical space Ω defined in step 2.3, G is the set of clean point clouds output in step 2.3, and p is any three-dimensional point in the point cloud set G. It is the shortest straight-line distance from the free point q to the point cloud G.

[0026] Step 3.2.2, estimation of the normal vector based on the local covariance matrix; The excavator's operating status is introduced as prior knowledge. An RTK-GPS positioning terminal is deployed on the bucket to directly measure the bucket's absolute three-dimensional coordinates in the physical world coordinate system. .

[0027] The real-time acquired position of the electric shovel bucket In physical space, the three-dimensional coordinates are used to obtain the point cloud after preserving neighboring points in step 2.2. By applying the same normalization parameters for translation and scaling, we obtain the normalized electric shovel bucket position vector located in the normal space Ω. .

[0028] (13) in, This represents the real-time three-dimensional coordinates of the electric shovel bucket in a standard spatial coordinate system. This represents the line-of-sight vector pointing from the free point q towards the bucket.

[0029] Find the nearest point of the 3D free query point q in the point cloud G. and with Using a circle as the center, search for its 20 nearest neighbors to form a local neighborhood set. .

[0030] Calculate the local neighborhood point set center of mass And construct a 3×3 covariance matrix C, as shown in formulas (14) and (15): (14) (15) Where C is a symmetric positive semi-definite matrix describing the degree of dispersion of the point cloud distribution in the local neighborhood. For local neighborhood point set The center of mass.

[0031] The covariance matrix C is decomposed into eigenvalues, and the eigenvector corresponding to the smallest eigenvalue is taken as the original normal vector and normalized, as shown in formula (16): (16) in It is the eigenvector corresponding to the smallest eigenvalue of matrix C. It is the original unit normal vector without direction correction. Define from the nearest point Pointing to the bucket position The vector is denoted as As shown in formula (17).

[0032] (17) like This indicates that the normal vector points inwards from the material pile. By reversing it so that it always points outwards from the material pile (towards the air), we obtain the oriented unit normal vector. As shown in formula (18).

[0033] (18) in, It is the final output unit normal vector with a consistent outward orientation.

[0034] Step 3.2.3 Symbol determination and pseudo-truth value construction based on bucket viewpoint; Using the normal after orientation With line of sight vector The sign attribute of the free point q is determined by the angle relationship between the two points, as shown in formula (19): Here, sign(q) represents the sign property of the free point q.

[0035] By combining distance and sign, pseudo-true values ​​are generated for supervised training. As shown in formula (20): (20) in It is the pseudo-SDF ground truth constructed from the free point q, which serves as the target output of the Gabor network.

[0036] Step 3.3 Construction of the composite loss function; Step 3.3.1 Construct the data item loss ; Constructing data item loss Measuring network predictions With false truth value The difference is shown in formula (21): (twenty one) in, It refers to the Gabor network at the coordinate point in step 3.2.1. Predicted SDF value; Coordinates The corresponding actual SDF value; It is the set of spatial points sampled during training; This is the total number of sampling points. Minimizing this means making the network's output value as close as possible to the true SDF value, which is fundamental to reconstructing the accurate surface shape.

[0037] Step 3.3.2, apply the Eikonal constraint term. Physical constraints are introduced to ensure that the network learns a truly signed distance function, as shown in equation (22): (twenty two) in, It is the Gabor network at coordinate points Predicted SDF value; It is the gradient operator, representing the function Regarding spatial coordinates The vector of partial derivatives.

[0038] Step 3.3.3, the loss function adopted. It is the weighted sum of steps 3.3.1 and 3.3.2, as shown in formula (23): (twenty three) In the parameter initialization settings, =1, =0.1; Optimizer uses Adam, learning rate... The training run consists of 1000 epochs. The goal is to achieve a trained network that can generalize from sparse, noisy point clouds to a complete, continuous, and physically reliable high-precision surface model.

[0039] Step 3.4: Use formulas (24) and (25) to calculate the shortest distance from the spatial point cloud to the material surface point cloud using the signed distance function SDF, and define the outer side of the material surface at the excavation site as positive and the inner side as negative.

[0040] (twenty four) (25) in, Represents the point The signed distance value at the location; Represents the point The closest distance to point cloud S; The three-dimensional coordinates of any point in space; This represents the set of point clouds on the material surface obtained after preprocessing. Point The negative of the nearest distance to point cloud S.

[0041] The fourth step is to convert the implicit field into an explicit mesh; Step 4.1, SDF field prediction: The specification space obtained in step 2.3 includes the entire effective area where the material pile surface is located. Create 256 [units / areas] within the specification space. 3 For each grid point, the Gabor network model trained in step 3 retrieves the SDF value of each clean point cloud generated in the normalized space after coordinate normalization in step 2.3. For each coordinate point (x, y, z), the Gabor network model outputs a scalar value, which serves as the predicted SDF value for that coordinate point. This process is highly parallelized on the GPU. The predicted SDF values ​​obtained from all clean point cloud data are collected and arranged according to their positions in the normalized space, ultimately resulting in a discrete, regular three-dimensional scalar field. Each predicted SDF value represents the shortest distance from its corresponding input clean point cloud data to the surface of the material pile. The sign of the predicted SDF value is used to distinguish whether it is inside or outside the surface of the material pile.

[0042] Step 4.2, extract isosurfaces with predicted SDF values ​​of 0: Using the Marching Cubes algorithm, isosurfaces with SDF = 0 are extracted from the SDF field predicted in step 4.1. These isosurfaces are output as a standard triangular mesh model to complete the three-dimensional reconstruction and obtain the triangular mesh model of the material pile surface.

[0043] Furthermore, the isosurface refers to the surface formed by all points in this SDF field where the function value is equal to a specific value. Therefore, the isosurface with SDF = 0 is the precise surface of the material pile. Thus, in this three-dimensional space filled with numbers, all these points with a value of 0 are identified and correctly connected to form a continuous surface. The three-dimensional space is divided into dense small cubes (voxels), and the relationship between the values ​​of the eight vertices of each cube and a given threshold (such as SDF = 0) is determined one by one, generating an 8-bit state index. The algorithm uses a pre-generated lookup table to immediately determine how the isosurface intersects with the cube and how to connect the intersection points to form triangular patches to approximate the local surface.

[0044] The beneficial effects of this invention are as follows: The core advantage of this invention in the technical field lies in its introduction of the Gabor wavelet activation function, which fundamentally overcomes the "accuracy-robustness" trade-off bottleneck faced by traditional implicit neural representations in the reconstruction of material surfaces in mining electric shovels. Compared to the inherent defects of ReLU and its variants, which easily lead to excessive smoothing of geometric details, or the inherent defects of sinusoidal activation functions (such as SIREN), which easily fit sensor noise and produce artifacts, this invention utilizes the unique mathematical properties of Gabor wavelets, which combine the high-frequency representation capability of sinusoidal carrier waves with the spatial locality of Gaussian envelopes, to endow neural networks with accurate analytical capabilities for complex geometric features of material surfaces. Specifically, the Gaussian term in the Gabor activation function can effectively suppress noise interference from dust, vibration, and missing point cloud regions, significantly improving the generalization ability and stability of the model; while its inherent periodic oscillation characteristics can capture high-dimensional geometric information such as steep edges, local concavities and convexities, and fine textures on the surface of the material pile with high fidelity, avoiding the loss of details caused by traditional interpolation methods. Furthermore, the Gabor network combined with Eikonal physical constraints can construct a signed distance field with continuous gradients and clear physical meaning, thereby achieving a seamless closed loop from perception to planning under sparse and noisy working conditions, providing a three-dimensional vision foundation for electric shovel operations that combines high precision, high robustness and real-time performance. Attached Figure Description

[0045] Figure 1 To implement the flowchart; Figure 2 The graph shows the Gabor activation function. Figure 3 This is a graph of the SIREN activation function. Detailed Implementation

[0046] The present invention will be further described below with reference to specific implementation examples.

[0047] This embodiment provides a method for three-dimensional reconstruction of the material surface in a mining electric shovel by introducing a Gabor periodic activation function neural network. Figure 1 As shown, this is a high-fidelity parameterized representation scheme for the geometry of a material surface based on the periodic activation function unique to Gabor neural networks. Specifically, it includes the following steps: The first step is environmental perception, which involves collecting geographic information as raw point cloud data. Utilizing the binocular camera and lidar in the sensing hardware system of a mining electric shovel, such as Figure 1 As shown, comprehensive geographic information of the excavation site was collected, including topographic and geomorphological environmental data, as the raw data point cloud.

[0048] The second step involves preprocessing the raw point cloud data acquired by the binocular camera and LiDAR. This preprocessing includes point cloud denoising based on statistical filtering and radius filtering methods, ultimately yielding clean point cloud data G with significantly reduced noise, preserved key geometric features, and optimized data volume. This completes the data processing. Figure 1 The process is shown below.

[0049] Step 2.1, for each point in the raw point cloud data collected by the binocular camera and LiDAR... Calculate the nearest Euclidean distance. The average distance of each neighboring point , The nearest neighboring point is the target point. One point. For The value of is first determined from empirical values. =50 and observe the filtering effect, then perform adaptive fine-tuning, with the optimal fine-tuning range being [30, 70]. Combined with the visualization effect, determine the best value. The average distance is obtained using formula (1): (1) in, This means The nth "nearest neighbor" point There are points, among which =1, 2, ..., traverse all neighboring points; Then, using the average distance Calculate the mean of the average distances of all points in the raw point cloud data acquired by the binocular camera and LiDAR. and standard deviation As shown in formula (2): (2) Where N represents the total number of original data point clouds involved in the "average distance" calculation; i is the "index variable" for the summation process, used to iterate through each of these N data points. μ represents all... The arithmetic mean of the point cloud reflects the average local density of the point cloud. for The standard deviation measures the degree of local density dispersion. Calculate the standard deviation for each point in the original point cloud data. To its The average distance between the nearest neighbors Then seek all The mean μ and standard deviation .

[0050] Subsequently, if satisfy If a point is identified as an outlier and is removed, then it is considered an outlier and is removed. Let be a positive scaling factor. =2, used to adjust the strictness of noise judgment.

[0051] After removing outliers, the intermediate point cloud is obtained. The total number of points in the intermediate point cloud at this point is denoted as . As shown in formula (3): The denoising method provided in this step can effectively filter out random "flying points" in the original data point cloud, but it may not be effective for point clouds with non-uniform distribution, so further denoising is required.

[0052] Step 2.2, after removing outlier noise points in step 2.1, adjust the intermediate point cloud based on local density. Each point in Count the number of neighborhood points within a sphere centered at the given point and with a radius of r = 0.5 m. , including This helps avoid misjudgments of empty neighborhoods. Neighborhood points within the sphere Neighborhood points within each sphere Then, construct a sphere with radius r = 0.5m and count the number of its secondary neighbor points. ,common indivual value.

[0053] Calculate the local mean of the number of secondary neighbors and standard deviation As shown in formula (4): (4) in, This is the local mean of the number of secondary neighbors within the neighborhood. The standard deviation of the number of secondary neighbors within the neighborhood. This is an index variable. It represents the index at point... Within the neighborhood of a sphere centered at r = 0.5m, the first... Neighboring points, For the point The total number of points contained within a sphere centered at a radius of r = 0.5m is given by... Each specific point within the neighborhood (i.e., the first) After taking the neighborhood points as the center, draw a sphere with a radius of r = 0.5m. This represents the number of points inside the sphere.

[0054] Set dynamic threshold ,in This is the adjustment coefficient, let it be here. =2. If Less than the set threshold Then determine Noise points are removed; if Less than the set threshold Then determine These are noise points and should be removed. Not less than the set threshold If a point is found to be a neighboring point, then that point is retained as a neighboring point. The point cloud after retaining the neighboring points is... , Point cloud after preserving neighboring points Points are recorded as ,and .

[0055] Step 2.3: Process the point cloud obtained in Step 2.2 after preserving the neighborhood points. Coordinate normalization is performed to map the data to a normalized space Ω, generating pure point cloud data G. This pure point cloud data provides stable and efficient input for the Gabor wavelet neural network. The specific steps are as follows: First, the point cloud obtained in step 2.2 after retaining neighborhood points is transformed using translation centering. The center of the network is moved to the origin to eliminate the influence of the absolute position on network training, as shown in formulas (5) and (6): (5) in, To preserve the point cloud of neighboring points The coordinates of the centroid; for The Middle The original coordinates of each point; The coordinates are those of the centroid. for Points.

[0056] Centralized coordinates obtained by translating a point cloud for: (6) in, The point cloud obtained in step 2.2 after preserving neighborhood points No. The coordinates of each point after centering; These are the centered coordinate components.

[0057] Then, the point cloud obtained in step 2.2 after retaining the neighborhood points is calculated using formula (7). The maximum distance from all points to the origin is the maximum norm. The point cloud obtained in step 2.2 after preserving neighborhood points is then processed. Divide all point coordinates by this maximum value to obtain the point cloud after retaining neighborhood points obtained in step 2.2. The entire scale is adjusted to the range [-1, 1], as shown in formula (8): (7) (8) in, Point cloud with maximum norm and neighborhood points preserved. The maximum distance from all points in the set to the origin; for Coordinate components; For the first The normalized coordinates of each point. These are the normalized coordinate components, ranging from [-1, 1].

[0058] In step 2.3, the point cloud obtained in step 2.2 after retaining the neighborhood points... All are scaled to [-1, 1], therefore the canonical space is defined as Ω = Ω represents the gauge space. All coordinate vectors obtained after coordinate normalization... The set is the pure point cloud data G, which not only conforms to the periodicity of the Gabor wavelet activation function, but also avoids training difficulties caused by the input value being too large or too small.

[0059] The third step involves training a Gabor wavelet network to transform the preprocessed, clean point cloud data from the second step into a continuous, differentiable, and high-fidelity geometric model of the material surface. This invention employs the signed distance function (SDF) as an implicit representation of the scene and leverages the superior characteristics of the Gabor wavelet network to learn and parameterize this SDF field. The Gabor activation function image is shown below. Figure 2 As shown, specifically: Step 3.1: Use the Gabor network model as the activation function to replace the traditional ReLU or TanH to accurately model the high-frequency details of the material surface.

[0060] Step 3.1.1: Location encoding; The clean point cloud data obtained after normalization in step 2.3 is mapped to the high-frequency feature space by the position encoding formula shown in formula (9); (9) in, B is the three-dimensional coordinate vector of a single point in the clean point cloud data obtained after normalization in step 2.3; B is a 3×m frequency band matrix, with 3 rows corresponding to the x, y, and z axes and 8 columns corresponding to 8 frequency bands. Typically, m=8 is set to cover the geometric frequency range from 0.1 to 10Hz. It is the encoded high-dimensional feature vector, representing the input coordinates. The high-dimensional feature representation after trigonometric function transformation has a dimension of 2×3×m. It is composed of the sine and cosine function values ​​of the product of each coordinate component and the frequency band matrix B. It is used as the input of Gabor neural network to enhance the perception of high-frequency details (such as cracks and textures) on the material surface and improve the accuracy of geometric reconstruction.

[0061] Step 3.1.2: Construct the network forward propagation and Gabor activation function; The network structure is a fully connected feedforward network (8 layers), with a 48-dimensional input layer (corresponding to...). The number of neurons in the hidden layer decreases (256→128→...→32), and the output layer is 1-dimensional. A Gabor activation function is constructed to replace the traditional ReLU or TanH functions and the SIREN function, although the SIREN network, based on the sine function, performs excellently in implicit representations, such as... Figure 3 As shown. However, it has the drawbacks of global periodic interference and difficulty in taking into account both gentle material surfaces and sharp rocks when dealing with large-scale mining scenes with a single frequency. Therefore, the Gabor wavelet activation function is introduced, as shown in formula (11). The Gabor function introduces a Gaussian envelope on the basis of the sine function. This endows the network with local perception capabilities, while also enabling the learning of center frequencies. With bandwidth This enables the network to have multi-scale adaptive characteristics, and to reproduce flat soil and steep rock corners with high fidelity.

[0062] Input to neurons in layer l As shown in formula (10): (10) Activate the output, as shown in formula (11): (11) in, Use the Gabor wavelet activation function; and These are weights and biases, respectively. This is the output feature vector of the (l-1)th layer (i.e., the input data of the lth layer); It is the center frequency, initially sampled randomly from a uniform distribution U(0, π). This is the bandwidth, initialized to a fixed constant, which is 1.0 here. This represents the phase shift and is initialized to 0. These parameters are given reasonable initial values ​​and then automatically learned and updated by the optimizer during training. This represents the output vector of the l-th layer, and the final output layer. That is, prediction .

[0063] Step 3.2: Construct a pseudo-SDF-based loss function to supervise the SDF values ​​predicted in step 3.1.

[0064] Since accurate ground truth SDF values ​​cannot be obtained in advance at the excavation site, this embodiment adopts a view-dependent pseudo-SDF supervision strategy. Using the real-time pose of the electric shovel bucket as a spatial reference frame, and combining it with point cloud normal information, a self-supervised loss function is constructed to guide the Gabor network in learning a physically intuitive signed distance field.

[0065] Step 3.2.1, Spatial Sampling and Nearest Neighbor Distance Calculation The gauge space Ω obtained in step 2.3 (i.e. Within the cubic region, a set of detached query points Q is randomly sampled using a stratified sampling strategy. For each detached query point q∈Q, the nearest Euclidean distance d(q,G) to the pure point cloud G is calculated, as shown in formula (12): (12) Where q is any three-dimensional free query point sampled in the canonical space Ω defined in step 2.3, G is the set of clean point clouds output in step 2.3, and p is any three-dimensional point in the point cloud set G. It is the shortest straight-line distance from the free point q to the point cloud G.

[0066] Step 3.2.2, Normal vector estimation based on local covariance matrix The excavator's operating status is introduced as prior knowledge. An RTK-GPS positioning terminal is deployed on the bucket to directly measure the bucket's absolute three-dimensional coordinates in the physical world coordinate system. .

[0067] The real-time acquired position of the electric shovel bucket In physical space, the three-dimensional coordinates are used to obtain the point cloud after preserving neighboring points in step 2.2. By applying the same normalization parameters for translation and scaling, we obtain the normalized electric shovel bucket position vector located in the normal space Ω. .

[0068] (13) in, This represents the real-time three-dimensional coordinates of the electric shovel bucket in a standard spatial coordinate system. This represents the line-of-sight vector pointing from the free point q towards the bucket.

[0069] Find the nearest point of the 3D free query point q in the point cloud G. and with Using a circle as the center, search for its 20 nearest neighbors to form a local neighborhood set. .

[0070] Calculate the local neighborhood point set center of mass And construct a 3×3 covariance matrix C, as shown in formulas (14) and (15): (14) (15) Where C is a symmetric positive semi-definite matrix describing the degree of dispersion of the point cloud distribution in the local neighborhood. For local neighborhood point set The center of mass.

[0071] The covariance matrix C is decomposed into eigenvalues, and the eigenvector corresponding to the smallest eigenvalue is taken as the original normal vector and normalized, as shown in formula (16): (16) in It is the eigenvector corresponding to the smallest eigenvalue of matrix C. It is the original unit normal vector without direction correction. Define from the nearest point Pointing to the bucket position The vector is denoted as As shown in formula (17).

[0072] (17) like This indicates that the normal vector points inwards from the material pile. By reversing it so that it always points outwards from the material pile (towards the air), we obtain the oriented unit normal vector. As shown in formula (18).

[0073] (18) in, It is the final output unit normal vector with a consistent outward orientation.

[0074] 3.2.3 Symbol determination and pseudo-truth value construction based on bucket viewpoint Using the normal after orientation With line of sight vector The sign attribute of the free point q is determined by the angle relationship between the two points, as shown in formula (19): Here, sign(q) represents the sign property of the free point q.

[0075] By combining distance and sign, pseudo-true values ​​are generated for supervised training. As shown in formula (20): (20) in It is the pseudo-SDF ground truth constructed from the free point q, which serves as the target output of the Gabor network.

[0076] 3.3 Construction of Composite Loss Function 3.3.1 Constructing the loss of data items Constructing data item loss Measuring network predictions With false truth value The difference is shown in formula (21): (twenty one) in, It refers to the Gabor network at the coordinate point in step 3.2.1. Predicted SDF value; Coordinates The corresponding actual SDF value; It is the set of spatial points sampled during training; This is the total number of sampling points. Minimizing this means making the network's output value as close as possible to the true SDF value, which is fundamental to reconstructing the accurate surface shape.

[0077] Step 3.3.2, apply the Eikonal constraint term. Physical constraints are introduced to ensure that the network learns a truly signed distance function, as shown in equation (22): (twenty two) in, It is the Gabor network at coordinate points Predicted SDF value; It is the gradient operator, representing the function Regarding spatial coordinates The vector of partial derivatives.

[0078] Step 3.3.3, the loss function adopted. It is the weighted sum of steps 3.3.1 and 3.3.2, as shown in formula (23): (twenty three) In the parameter initialization settings, =1, =0.1; Optimizer uses Adam, learning rate... The training run consists of 1000 epochs. The goal is to achieve a trained network that can generalize from sparse, noisy point clouds to a complete, continuous, and physically reliable high-precision surface model.

[0079] Step 3.4: Use formulas (24) and (25) to calculate the shortest distance from the spatial point cloud to the material surface point cloud using the signed distance function SDF, and define the outer side of the material surface at the excavation site as positive and the inner side as negative.

[0080] (twenty four) (25) in, Represents the point The signed distance value at the location; Represents the point The closest distance to point cloud S; The three-dimensional coordinates of any point in space; This represents the set of point clouds on the material surface obtained after preprocessing. Point The negative of the nearest distance to point cloud S.

[0081] The fourth step is to convert the implicit field into an explicit mesh; Step 4.1, SDF field prediction: The specification space obtained in step 2.3 includes the entire effective area where the material pile surface is located. Create 256 [units / areas] within the specification space. 3 For each grid point, the Gabor network model trained in step 3 retrieves the SDF value of each clean point cloud generated in the normalized space after coordinate normalization in step 2.3. For each coordinate point (x, y, z), the Gabor network model outputs a scalar value, which serves as the predicted SDF value for that coordinate point. This process is highly parallelized on the GPU. The predicted SDF values ​​obtained from all clean point cloud data are collected and arranged according to their positions in the normalized space, ultimately resulting in a discrete, regular three-dimensional scalar field. Each predicted SDF value represents the shortest distance from its corresponding input clean point cloud data to the surface of the material pile. The sign of the predicted SDF value is used to distinguish whether it is inside or outside the surface of the material pile.

[0082] Step 4.2, extract isosurfaces with predicted SDF values ​​of 0: Using the Marching Cubes algorithm, isosurfaces with SDF = 0 are extracted from the SDF field predicted in step 4.1. These isosurfaces are output as a standard triangular mesh model to complete the three-dimensional reconstruction and obtain the triangular mesh model of the material pile surface.

[0083] Furthermore, the isosurface refers to the surface formed by all points in this SDF field where the function value is equal to a specific value. Therefore, the isosurface with SDF = 0 is the precise surface of the material pile. Thus, in this three-dimensional space filled with numbers, all these points with a value of 0 are identified and correctly connected to form a continuous surface. The three-dimensional space is divided into dense small cubes (voxels), and the relationship between the values ​​of the eight vertices of each cube and a given threshold (such as SDF = 0) is determined one by one, generating an 8-bit state index. The algorithm uses a pre-generated lookup table to immediately determine how the isosurface intersects with the cube and how to connect the intersection points to form triangular patches to approximate the local surface.

[0084] The embodiments described above are merely illustrative of the implementation of the invention, but should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements are all within the scope of protection of the present invention.

Claims

1. A method for 3D reconstruction of a material surface of a mining shovel based on a neural network with a Gabor wave periodic activation function, characterized in that, The material surface three-dimensional reconstruction method includes the following steps: The first step is environmental perception, which involves collecting geographic information as raw point cloud data. Using binocular cameras and LiDAR, comprehensive geographic information of the excavation site is collected, including topographic and geomorphological environmental data, as raw data point cloud. Step 2: Preprocess the raw point cloud data collected by the binocular camera and LiDAR. The preprocessing includes point cloud denoising based on statistical filtering and radius filtering methods to obtain clean point cloud data G. The third step involves training a Gabor wavelet network to transform the preprocessed clean point cloud data from the second step into a continuous, differentiable, and high-fidelity surface geometry model. A signed distance function (SDF) is used as the implicit representation of the scene, and the superior characteristics of the Gabor wavelet network are leveraged to learn and parameterize this SDF field. Specifically: Step 3.1: Use the Gabor network model as the activation function to accurately model the high-frequency details of the material surface; Step 3.2: Construct a pseudo-SDF-based loss function to supervise the SDF values ​​predicted in step 3.1; Since the accurate SDF ground truth cannot be obtained in advance at the excavation site, a viewpoint-dependent pseudo-SDF supervision strategy is adopted. The real-time pose of the electric shovel bucket is used as a spatial reference frame. Combined with the point cloud normal information, a self-supervised loss function is constructed to guide the Gabor network to learn a signed distance field that conforms to physical intuition. Step 3.3 Construction of the composite loss function; Step 3.4: Calculate the shortest distance from the spatial point cloud to the material surface point cloud using the signed distance function SDF, and define the outer side of the material surface at the excavation site as positive and the inner side as negative. The fourth step is to convert the implicit field into an explicit mesh; Step 4.1, SDF field prediction: Step 4.2: Extract isosurfaces with predicted SDF values ​​of 0.

2. The method according to claim 1, wherein, The second step is as follows: Step 2.1, for each point in the raw data point cloud collected by binocular camera and lidar , calculate the average distance of its nearest neighbor points , The nearest points to the target point are the nearest neighbor points to the target point; use the average distance to calculate the mean and standard deviation of the average distance of all points in the raw data point cloud collected by binocular camera and lidar like satisfy If a point is identified as an outlier and is removed, then it is considered an outlier and is removed. Let be a positive scaling factor. =2, used to adjust the strictness of noise judgment; After removing outliers, we obtain the intermediate point cloud. The total number of points in the intermediate point cloud at this point is denoted as . ; As shown in formula (3): Step 2.2, after removing outlier noise points in step 2.1, adjust the intermediate point cloud based on local density. Each point in Count the number of neighborhood points within a sphere centered at the given point and with a radius of r = 0.5 m. , including To avoid misjudgment of empty neighborhoods; Neighborhood points within the sphere Neighborhood points within each sphere Then, construct a sphere with radius r = 0.5m and count the number of its secondary neighbor points. ,common indivual value; Calculate the local mean of the number of secondary neighbors and standard deviation ; Set dynamic threshold ,in This is the adjustment coefficient, let it be here. =2; if Less than the set threshold Then determine Noise points are removed; if Less than the set threshold Then determine These are noise points and should be removed. Not less than the set threshold If a point is not found in the neighborhood, then that point is retained as a neighboring point; the point cloud after retaining the neighboring points is... , Point cloud after preserving neighboring points Points are recorded as ,and ; Step 2.3: Process the point cloud obtained in Step 2.2 after preserving the neighborhood points. The coordinates are normalized and mapped to a normalized space Ω to generate clean point cloud data G; the specific steps are as follows: First, the point cloud obtained in step 2.2 after retaining neighborhood points is transformed using translation centering. The center of the network is moved to the origin to eliminate the influence of the absolute position on network training, as shown in formulas (5) and (6): （5） in, To preserve the point cloud of neighboring points The coordinates of the centroid; for The Middle The original coordinates of each point; The coordinates are those of the centroid. for points; Centralized coordinates obtained by translating a point cloud for: （6） in, The point cloud obtained in step 2.2 after preserving neighborhood points No. The coordinates of each point after centering; These are the centered coordinate components; Then, the point cloud obtained in step 2.2 after retaining the neighborhood points is calculated using formula (7). The maximum distance from all points to the origin is the maximum norm; the point cloud obtained in step 2.2 after retaining neighborhood points is... Divide all point coordinates by this maximum value, and then use the point cloud obtained in step 2.2 after retaining the neighborhood points. The entire range is scaled to [-1, 1], as shown in formula (8): （7） （8） in, Point cloud with maximum norm and neighborhood points preserved. The maximum distance from all points in the set to the origin; for Coordinate components; For the first Normalized coordinates of each point; These are normalized coordinate components, ranging from [-1, 1]. The point cloud obtained in step 2.2 after preserving neighborhood points If all are scaled to [-1, 1], then the canonical space is defined as Ω = Ω represents the normalized space; all coordinate vectors obtained after coordinate normalization. The set is the pure point cloud data G.

3. The method for three-dimensional reconstruction of the material surface of a mining electric shovel based on a Gabor wavelet periodic activation function neural network according to claim 2, characterized in that, In step 2.1: The average distance and mean and standard deviation The following formula is used to obtain it: （1） in, This means The nth "nearest neighbor" point There are points, among which =1, 2, ..., traverse all neighboring points; （2） Where N represents the total number of all original data point clouds involved in the "average distance" calculation; i is the "index variable" for the summation process, used to iterate through each of these N data points; μ is the summation value for all points in the original data cloud. The arithmetic mean of the point cloud reflects the average local density of the point cloud. for The standard deviation measures the degree of local density dispersion; it is used to calculate the standard deviation of each point in the original point cloud. To its The average distance between the nearest neighbors Then seek all Mean μ and standard deviation .

4. The method for three-dimensional reconstruction of the material surface of a mining electric shovel based on a Gabor wavelet periodic activation function neural network according to claim 3, characterized in that, In step 2.2: The local mean and standard deviation As shown in formula (4): （4） in, This is the local mean of the number of secondary neighbors within the neighborhood. The standard deviation of the number of secondary neighbors within the neighborhood. It is an index variable.

5. The method for three-dimensional reconstruction of the material surface of a mining electric shovel based on a Gabor wavelet periodic activation function neural network according to claim 4, characterized in that, Step 3.1 specifically involves: Step 3.1.1: Location encoding; The clean point cloud data obtained after normalization in step 2.3 is mapped to the high-frequency feature space by the position encoding formula shown in formula (9); （9） in, B is the three-dimensional coordinate vector of a single point in the clean point cloud data obtained after normalization in step 2.3; B is a 3×m frequency band matrix, with 3 rows corresponding to the x, y, and z axes and 8 columns corresponding to 8 frequency bands, covering the geometric frequency range from 0.1 to 10Hz. It is the encoded high-dimensional feature vector, representing the input coordinates. The high-dimensional feature representation after trigonometric function transformation has a dimension of 2×3×m. It is composed of the sine and cosine function values ​​of the product of each coordinate component and the frequency band matrix B, and is used as the input of the Gabor neural network. Step 3.1.2: Construct the network forward propagation and Gabor activation function; The network structure is a fully connected feedforward network with a 48-dimensional input layer. The number of neurons in the hidden layer decreases, and the output layer is 1-dimensional; a Gabor activation function is constructed; the input to the l-th layer neuron is... As shown in formula (10): （10） Activate the output, as shown in formula (11): （11） in, Use the Gabor wavelet activation function; and These are weights and biases, respectively. This is the output feature vector of the (l-1)th layer; It is the center frequency, initially sampled randomly from a uniform distribution U(0, π); It is the bandwidth, initialized to a fixed constant; This represents the phase shift; it is given an initial value, and then the optimizer automatically learns and updates it during training. This represents the output vector of the l-th layer, and the final output layer. That is, prediction .

6. The method for three-dimensional reconstruction of the material surface of a mining electric shovel based on a Gabor wavelet periodic activation function neural network according to claim 5, characterized in that, Step 3.2 specifically involves: Step 3.2.1, Spatial Sampling and Nearest Neighbor Distance Calculation Within the canonical space Ω obtained in step 2.3, a set of free query points Q is randomly sampled using a stratified sampling strategy; for each free query point q∈Q, the nearest Euclidean distance d(q,G) to the pure point cloud G is calculated, as shown in formula (12): （12） Where q is any three-dimensional free query point sampled in the canonical space Ω defined in step 2.3, G is the set of clean point clouds output in step 2.3, and p is any three-dimensional point in the point cloud set G; It is the shortest straight-line distance from the free point q to the point cloud G; Step 3.2.2, estimation of the normal vector based on the local covariance matrix; The excavator's operating status is introduced as prior knowledge; an RTK-GPS positioning terminal is deployed on the bucket to directly measure the bucket's absolute three-dimensional coordinates in the physical world coordinate system. ; The real-time acquired position of the electric shovel bucket In physical space, the three-dimensional coordinates are used to obtain the point cloud after preserving neighboring points in step 2.

2. By applying the same normalization parameters for translation and scaling, we obtain the normalized electric shovel bucket position vector located in the normal space Ω. ; （13） in, This represents the real-time three-dimensional coordinates of the electric shovel bucket in a standard spatial coordinate system. This represents the line-of-sight vector from the free point q towards the bucket; Find the nearest point of the 3D free query point q in the point cloud G. and with Using a circle as the center, search for its 20 nearest neighbors to form a local neighborhood set. ; Calculate the local neighborhood point set center of mass And construct a 3×3 covariance matrix C, as shown in formulas (14) and (15): （14） （15） Where C is a symmetric positive semi-definite matrix describing the degree of dispersion of the point cloud distribution in the local neighborhood. For local neighborhood point set The center of mass; The covariance matrix C is decomposed into eigenvalues, and the eigenvector corresponding to the smallest eigenvalue is taken as the original normal vector and normalized, as shown in formula (16): （16） in It is the eigenvector corresponding to the smallest eigenvalue of matrix C; It is the original unit normal vector without direction correction. Define from the nearest point Pointing to the bucket position The vector is denoted as As shown in formula (17); （17） like This indicates that the normal vector points inwards from the material pile. By reversing it so that it always points outwards from the material pile (towards the air), we obtain the oriented unit normal vector. As shown in formula (18); （18） in, It is the final output unit normal vector with a consistent outward orientation; Step 3.2.3 Symbol determination and pseudo-truth value construction based on bucket viewpoint; Using the normal after orientation With line of sight vector The sign attribute of the free point q is determined by the angle relationship between the two points, as shown in formula (19): Where sign(q) represents the sign property of the free point q; By combining distance and sign, pseudo-true values ​​are generated for supervised training. As shown in formula (20): （20） in It is the pseudo-SDF ground truth constructed from the free point q, which serves as the target output of the Gabor network.

7. The method for three-dimensional reconstruction of the material surface of a mining electric shovel based on a Gabor wavelet periodic activation function neural network according to claim 6, characterized in that, Step 3.3 specifically involves: Step 3.3.1 Construct the data item loss ; Constructing data item loss Measuring network predictions With false truth value The difference is shown in formula (21): （21）。 in, It refers to the Gabor network at the coordinate point in step 3.2.

1. Predicted SDF value; Coordinates The corresponding actual SDF value; It is the set of spatial points sampled during training; This is the total number of sampling points; Step 3.3.2, apply the Eikonal constraint term. Introduce physical constraints; as shown in formula (22): （22） in, It is the Gabor network at coordinate points Predicted SDF value; It is the gradient operator, representing the function Regarding spatial coordinates The vector of partial derivatives; Step 3.3.3, the loss function adopted. It is the weighted sum of steps 3.3.1 and 3.3.2, as shown in formula (23): (twenty three).

8. The method for three-dimensional reconstruction of the material surface of a mining electric shovel based on a Gabor wavelet periodic activation function neural network according to claim 7, characterized in that, Step 3.4 specifically involves: Formulas (24) and (25) are used to calculate the shortest distance from the spatial point cloud to the material surface point cloud using the signed distance function SDF, and it is agreed that the outer side of the material surface at the excavation site is positive and the inner side is negative; （24） （25） in, Represents the point The signed distance value at the location; Represents the point The closest distance to point cloud S; The three-dimensional coordinates of any point in space; This represents the set of point clouds on the material surface obtained after preprocessing. Point The negative of the nearest distance to point cloud S.

9. The method for three-dimensional reconstruction of the material surface of a mining electric shovel based on a Gabor wavelet periodic activation function neural network according to claim 8, characterized in that, The fourth step specifically involves: Step 4.1, SDF field prediction: The canonical space obtained in step 2.3 includes the entire effective area where the material pile surface is located. Multiple grid points are created in the canonical space, and the SDF value of each clean point cloud data generated in the canonical space after coordinate normalization in step 2.3 is obtained by querying the Gabor network model trained in step 3. For each coordinate point (x, y, z), the Gabor network model will output a scalar value, which is used as the predicted SDF value of that coordinate point. The predicted SDF values ​​obtained from all clean point cloud data are collected and arranged according to their positions in the canonical space, resulting in a discrete, regular three-dimensional scalar field. Each predicted SDF value represents the shortest distance from its corresponding input clean point cloud data to the material pile surface. The sign of the predicted SDF value is used to distinguish whether it is inside or outside the material pile surface. Step 4.2, extract isosurfaces with predicted SDF values ​​of 0: Using the Marching Cubes algorithm, isosurfaces with SDF = 0 are extracted from the SDF field predicted in step 4.

1. These isosurfaces are output as a standard triangular mesh model to complete the three-dimensional reconstruction and obtain the triangular mesh model of the material pile surface.

10. The method for three-dimensional reconstruction of the material surface of a mining electric shovel based on a Gabor wavelet periodic activation function neural network according to claim 9, characterized in that, The isosurface refers to the surface formed by all points in this SDF field where the function value is equal to a certain specific value; then the isosurface with SDF = 0 is the precise surface of the stockpile.

Images