Atmospheric laser unidirectional transmission multi-channel fusion method and device and storage medium

Abstract

The application relates to the technical field of multi-channel fusion, and specifically discloses a one-way transmission multi-channel fusion method for atmospheric laser, which acquires multi-channel color point cloud data through a multispectral laser radar and carries out denoising processing; analyzes laser loss caused by atmospheric absorption and scattering, and constructs a laser loss prediction model; calculates the average point spacing, the minimum point spacing and the maximum point spacing of the multi-channel color point cloud data, constructs a noise elimination degree evaluation model, and improves the quality of the multi-channel color point cloud data; fuses the multi-channel color point cloud data, generates a three-dimensional model by using the fused color point cloud data, establishes a three-dimensional model quality evaluation model, and evaluates the quality of the generated three-dimensional model, so that the influence of noise and laser transmission loss is avoided, and the authenticity and accuracy of the generated three-dimensional model are enhanced. The application further discloses a corresponding computer device and a storage medium.

Description

Technical Field

[0001] This invention relates to the field of multi-channel fusion technology, and more specifically, to a method for multi-channel fusion of atmospheric laser unidirectional transmission, a computer device, and a storage medium. Background Technology

[0002] Currently, most lidar systems utilize unidirectional, single-wavelength lasers, offering advantages such as high precision, high resolution, and high-precision point clouds. These systems can effectively reflect the three-dimensional spatial information of objects. However, the monochromatic point cloud data they acquire lacks sufficient spectral or intensity information, failing to meet the requirements of color 3D images and hindering its visualization development. The emergence and development of multispectral lidar technology has provided a completely new data format for point clouds: multispectral lidar point clouds. These contain both multi-band radiation information about the target and three-dimensional spatial information. Compared to monochromatic lidar data, multispectral lidar point cloud data plays a significant role in target classification and recognition technologies, and its visualization capabilities in three-dimensional imaging have been explored. Based on multispectral lidar technology, multiple spectral channels in the visible light band are selected and fused to form a color three-dimensional image, obtaining a visually appealing color point cloud without the need for passive image compensation.

[0003] In the acquisition of color point cloud data by multispectral lidar, noise exists in the color point cloud data due to factors such as measurement errors and specular reflection. However, existing technologies do not evaluate the degree of noise reduction in color point cloud data or the quality of the 3D model, thus affecting the accuracy of the generated 3D model. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, the present invention provides an atmospheric laser unidirectional transmission multi-channel fusion method, a computer device, and a storage medium to solve the above-mentioned technical problems.

[0005] To achieve the above objectives, according to one aspect of the present invention, a method for fusion of unidirectional atmospheric laser transmission channels is provided, comprising the following steps:

[0006] S1, Multi-channel data acquisition: Acquire multi-channel color point cloud data through multispectral lidar and perform noise reduction processing;

[0007] S2, Laser Loss Analysis: This section analyzes laser loss caused by atmospheric absorption and scattering, and constructs a laser loss prediction model to predict the loss of laser signals of different wavelengths during transmission. The formula for the laser loss prediction model is as follows:

[0008] Z LOSS =(LOSS abp +LOSS sct )*d;

[0009] In the formula: Z LOSS LOSS is the predicted value for laser loss. abp LOSS is the amount of atmospheric absorption loss. sct Where d is the atmospheric scattering loss and d is the transmission distance;

[0010] S3, Noise Reduction Assessment: By calculating the average, minimum, and maximum point spacing of multi-channel color point cloud data, a noise reduction assessment model is constructed to evaluate the degree of noise reduction in channel color point cloud data.

[0011] S4, Multi-channel fusion and 3D model generation: Fuse multi-channel color point cloud data and use the fused color point cloud data to generate a 3D model;

[0012] S5, Model Quality Assessment: Combining S2 and S3, this step assesses the quality of the generated 3D model.

[0013] As a preferred embodiment of the present invention, the laser loss analysis in S2 analyzes the laser loss caused by atmospheric absorption and scattering, wherein the analytical formula for the amount of atmospheric absorption loss is:

[0014]

[0015] In the formula: LOSS abp d is the atmospheric absorption loss, d is the transmission distance, α is the atmospheric absorption coefficient, and dz is the integral variable;

[0016] As a preferred embodiment of the present invention, the analytical formula for atmospheric scattering loss is as follows:

[0017]

[0018] In the formula: LOSS sct I0 is the atmospheric scattering loss, I0 is the initial intensity of the laser, and λ is the laser wavelength. z The wavelength of the scattered light is denoted as .

[0019] As a preferred embodiment of the present invention, the noise cancellation evaluation in S3 is carried out by calculating the average point spacing, minimum point spacing, and maximum point spacing of the multi-channel color point cloud data to construct a noise cancellation degree evaluation model. The formula for the noise cancellation degree evaluation model is as follows:

[0020]

[0021] In the formula: CZX represents the noise reduction level, MZX represents the average point spacing of the multi-channel color point cloud data, MinZX represents the minimum point spacing of the multi-channel color point cloud data, and MaxZX represents the maximum point spacing of the multi-channel color point cloud data.

[0022] As a preferred embodiment of the present invention, the average point spacing, minimum point spacing, maximum point spacing, and standard deviation of multi-channel color point cloud data are calculated, wherein the formula for the average point spacing of multi-channel color point cloud data is:

[0023]

[0024] In the formula: MZX is the average pixel spacing of the multi-channel color point cloud data, p is the total amount of multi-channel color point cloud data, and P i For the i-th multi-channel color point cloud data, P ih h is the nearest neighbor of the i-th multi-channel color point cloud data.

[0025] As a preferred embodiment of the present invention, the formula for the minimum point spacing of multi-channel color point cloud data is:

[0026] MinZX=min i min h ||P i -P ih ||;

[0027] In the formula: MinZX is the minimum point spacing of the multi-channel color point cloud data, min i For the minimum value operation of the index channel color point cloud data, min h This is an operation to find the minimum nearest neighbor point in the index channel color point cloud data.

[0028] As a preferred embodiment of the present invention, the formula for the maximum point spacing of multi-channel color point cloud data is:

[0029] MaxZX = max i max h ||P i -P ih ||;

[0030] In the formula: MaxZX is the maximum pixel spacing of the multi-channel color point cloud data, max i The max operation is used to maximize the color point cloud data in the index channel. h This operation sets the maximum value of the nearest neighbor points in the index channel color point cloud data.

[0031] As a preferred embodiment of the present invention, in S4, multi-channel fusion and 3D model generation involves fusing multi-channel color point cloud data. During the fusion process, a duplicate data removal network is established to remove duplicate multi-channel color point cloud data. The steps for establishing the duplicate data removal network are as follows:

[0032] Step A1, Feature Representation: The feature representation of the multi-channel color point cloud data is learned through a convolutional neural network. The feature representation of the multi-channel color point cloud data is as follows:

[0033]

[0034] In the formula: G M This represents the total feature representation of multi-channel color point cloud data, where ReLU is the ReLU activation function. This is the first feature representation of multi-channel color point cloud data. This represents the p-th feature of the multi-channel color point cloud data;

[0035] Step A2 involves constructing a deduplication network to remove duplicate multi-channel color point cloud data. The deduplication network is represented as follows:

[0036]

[0037] Where: PLG MJ The results of the deduplication network are represented as follows: Q represents the total number of duplicate multi-channel color point cloud data, and w u Let G be the weight matrix. M represents the overall feature representation of multi-channel color point cloud data, and softmax is the softmax activation function;

[0038] Step A3, calculate the network loss function: calculate the loss function of the deduplication network using cross-entropy loss;

[0039] Step A4, Point Cloud Fusion: The multi-channel color point cloud data is fused. The fused multi-channel color point cloud data is represented as follows:

[0040]

[0041]

[0042] Where: RXH fused For the representation of the fused multi-channel color point cloud data, RXH n The data after removing duplicate multi-channel color point cloud data, Qp is the number of data points after removing duplicate multi-channel color point cloud data, ω n This represents the weight of the nth channel.

[0043] As a preferred embodiment of the present invention, step A3 calculates the network loss function. The loss function of the deduplication network is calculated using cross-entropy loss. The formula for the loss function of the deduplication network is as follows:

[0044] LCE=-∑ u PLG MJ*lg(p u );

[0045] In the formula: LCE is the loss function result of the deduplication network, p u This represents the probability distribution of the laser filtering network results.

[0046] As a preferred technical solution of the present invention, the step S41 of learning the feature representation of multi-channel color point cloud data through a convolutional neural network can be carried out using the following specific steps, but the present invention is not limited to the following steps:

[0047] Step B1, Data Processing: Divide the multi-channel color point cloud data into a training signal set and a test signal set;

[0048] Step B2, construct the convolutional neural network model: The convolutional neural network model includes several input layers, convolutional layers, pooling layers, and fully connected layers;

[0049] Step B3, Input layer construction: The size of the input layer depends on the data dimension of the multi-channel color point cloud data. When the multi-channel color point cloud data is a single channel with a width of W and a height of H, the shape of the input layer is W×H×1.

[0050] Step B4, Convolutional Layer Construction: The convolutional layer uses filters to perform convolution operations on the input multi-channel color point cloud data and adds bias. The formula for calculating the output feature map of the convolutional layer is:

[0051] Output multi-channel color point cloud data feature map = ∑[input multi-channel color point cloud data * filter] + bias;

[0052] Step B5, Pooling layer construction: The pooling layer downsamples the feature map of the multi-channel color point cloud data output by the convolutional layer, thereby reducing the dimension and number of parameters of the multi-channel color point cloud data;

[0053] Step B6, Fully Connected Layer Construction: The multi-channel color point cloud data feature map output from the pooling layer is unfolded into a vector, and calculated by weighted summation plus bias. The output formula of the fully connected layer is:

[0054] The p-th feature representation of the output multi-channel color point cloud data is calculated as ∑[the p-th feature vector of the input multi-channel color point cloud data * weight] + bias.

[0055] As a preferred embodiment of the present invention, in step S5, the model quality assessment, combined with steps S2 and S3, evaluates the quality of the generated 3D model. A 3D model quality assessment model is established using laser loss prediction, noise reduction degree, model geometric quality indicators, and the average area of ​​model facets. The formula for the 3D model quality assessment model is:

[0056] W = αAEL * AFL - Z LOSS -CZX;

[0057] Where: W is the quality assessment value of the 3D model, α is the quality control factor, AEL is the predicted laser loss value, AFL is the degree of noise reduction, and Z... LOSS CZX is the geometric quality index of the model, and CZX is the average area of ​​the model's facets.

[0058] As a preferred embodiment of the present invention, the formula for calculating the geometric quality index of the model is as follows:

[0059]

[0060] In the formula: Z LOSS The geometric quality index of the model, where E is the total number of edges on the surface of the 3D model, and Length(e) is the geometric quality index of the model. r ) represents the length of the surface edge of the 3D model.

[0061] As a preferred embodiment of the present invention, the formula for calculating the average area of ​​the model patches is as follows:

[0062]

[0063] In the formula: CZX is the average area of ​​the model facets, F is the total number of facets in the 3D model, and Area(f d ) represents the area of ​​a facet in the 3D model.

[0064] According to another aspect of the present invention, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory, characterized in that the processor executes the computer program to implement the steps of the above-described method.

[0065] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein computer program instructions are stored thereon, characterized in that the computer program instructions, when executed by a processor, implement the steps of the above-described method.

[0066] In summary, the technical solutions conceived by this invention have the following beneficial effects compared with the prior art:

[0067] This invention acquires multi-channel color point cloud data using a multispectral lidar and performs noise reduction processing; it analyzes laser loss caused by atmospheric absorption and scattering, and constructs a laser loss prediction model, which helps in subsequent correction and model evaluation; by calculating the average, minimum, and maximum point spacing of the multi-channel color point cloud data, it constructs a noise reduction evaluation model, which helps improve the quality of the multi-channel color point cloud data; it fuses the multi-channel color point cloud data, uses the fused color point cloud data to generate a 3D model, and establishes a 3D model quality evaluation model to evaluate the quality of the generated 3D model, avoiding the influence of noise and laser transmission loss, and enhancing the realism and accuracy of the generated 3D model. Attached Figure Description

[0068] Figure 1 This is a flowchart illustrating a method for unidirectional atmospheric laser transmission and multichannel fusion according to an embodiment of the present invention. Detailed Implementation

[0069] The technical solutions of this invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described technical solutions are only a part of this invention, and not all of it. All other technical solutions obtained by those skilled in the art based on the technical solutions of this invention without inventive effort are within the scope of protection of this invention.

[0070] like Figure 1 The present invention provides an atmospheric laser unidirectional transmission multi-channel fusion method according to an embodiment of the present invention, comprising the following steps:

[0071] Step 1, Multi-channel data acquisition: Acquire multi-channel color point cloud data using a multispectral lidar and perform noise reduction processing;

[0072] Step 2, Laser Loss Analysis: Analyze the laser loss caused by atmospheric absorption and scattering, and construct a laser loss prediction model to predict the loss of laser signals of different wavelengths during transmission. The formula for the laser loss prediction model is:

[0073] Z LOSS =(LOSS abp +LOSS sct )*d;

[0074] In the formula: Z LOSS LOSS is the predicted value for laser loss. abp LOSS is the amount of atmospheric absorption loss. sct Where d is the atmospheric scattering loss and d is the transmission distance;

[0075] Step 3, Noise Reduction Assessment: By calculating the average, minimum, and maximum point spacing of the multi-channel color point cloud data, a noise reduction assessment model is constructed to evaluate the degree of noise reduction in the channel color point cloud data.

[0076] Step 4, Multi-channel fusion and 3D model generation: Fuse the multi-channel color point cloud data and use the fused color point cloud data to generate a 3D model;

[0077] Step 5, Model Quality Assessment: Combining Step 2 and Step 3, assess the quality of the generated 3D model.

[0078] In this embodiment, acquiring multi-channel color point cloud data using a multispectral lidar helps to capture the features of the measured object more comprehensively. Noise reduction processing improves the quality of the channel color point cloud data and reduces the impact of measurement errors and other interferences. The laser loss prediction model can more accurately correct the laser data and improve the reliability of the data. By calculating the average, minimum, and maximum point spacing of the multi-channel color point cloud data, the degree of noise reduction is evaluated, helping to determine the clarity and reliability of the channel color point cloud data. The multi-channel color point cloud data is fused to generate a high-quality 3D model, providing a quantitative evaluation of the model's accuracy, clarity, and reliability.

[0079] In this embodiment, preferably, step two involves laser loss analysis, which analyzes the laser loss caused by atmospheric absorption and scattering. The formula for analyzing atmospheric absorption loss is as follows:

[0080]

[0081] In the formula: LOSS abp d is the atmospheric absorption loss, d is the transmission distance, α is the atmospheric absorption coefficient, and dz is the integral variable;

[0082] In this embodiment, preferably, the analytical formula for atmospheric scattering loss is:

[0083]

[0084] In the formula: LOSS sct I0 is the atmospheric scattering loss, I0 is the initial intensity of the laser, and λ is the laser wavelength. z The wavelength of the scattered light is denoted as .

[0085] In this embodiment, by integrating the atmospheric absorption coefficient α over the transmission distance d, a quantitative estimate of the loss of laser signal due to atmospheric absorption during transmission is provided, which can correct the signal strength reduction caused by atmospheric absorption, thereby improving the accuracy of the data; by integrating the initial laser intensity I0 and wavelength parameters, a quantitative estimate of the loss of laser signal due to atmospheric scattering during transmission is provided, which can correct the signal strength reduction caused by atmospheric scattering, thereby improving the accuracy of the data.

[0086] In this embodiment of the invention, step three, noise reduction assessment, involves constructing a noise reduction assessment model by calculating the average, minimum, and maximum point spacing of the multi-channel color point cloud data. The formula for the noise reduction assessment model is as follows:

[0087]

[0088] In the formula: CZX represents the noise reduction level, MZX represents the average point spacing of the multi-channel color point cloud data, MinZX represents the minimum point spacing of the multi-channel color point cloud data, and MaxZX represents the maximum point spacing of the multi-channel color point cloud data.

[0089] In this embodiment of the invention, the average point spacing, minimum point spacing, maximum point spacing, and standard deviation of the multi-channel color point cloud data are calculated. The formula for the average point spacing of the multi-channel color point cloud data is:

[0090]

[0091] In the formula: MZX is the average pixel spacing of the multi-channel color point cloud data, z is the total number of pixels in the multi-channel color point cloud data, and P is the total number of pixels in the multi-channel color point cloud data. i For the i-th multi-channel color point cloud data, P ih h is the nearest neighbor of the i-th multi-channel color point cloud data.

[0092] The average point spacing of multi-channel color point cloud data gives the average distribution density of points in the multi-channel color point cloud data. When the average point spacing is small, it indicates that the point cloud data is relatively dense; when the average point spacing is large, it indicates that the point cloud data is relatively sparse, and it can identify density changes in local areas.

[0093] In this embodiment, the formula for the minimum point spacing of multi-channel color point cloud data is:

[0094] MinZX=min i min h ||P i -P ih ||;

[0095] In the formula: MinZX is the minimum point spacing of the multi-channel color point cloud data, mini For the minimum value operation of the index channel color point cloud data, min h This is an operation to find the minimum nearest neighbor point in the index channel color point cloud data.

[0096] The minimum point spacing of multi-channel color point cloud data represents the distance between the two closest multi-channel color point cloud data. It can be used to locate potentially noisy areas in a local point cloud, which helps with subsequent noise processing and correction.

[0097] In this embodiment, the formula for the maximum pixel spacing of multi-channel color point cloud data is:

[0098] MaxZX = max i max h ||P i -P ih ||;

[0099] In the formula: MaxZX is the maximum pixel spacing of the multi-channel color point cloud data, max i The max operation is used to maximize the color point cloud data in the index channel. h This operation sets the maximum value of the nearest neighbor points in the index channel color point cloud data.

[0100] The maximum pixel pitch of multi-channel color point cloud data can be used to identify gaps or sparse regions in a scene, making it easier to understand the structure and characteristics of the entire point cloud data.

[0101] In this embodiment, step four, multi-channel fusion and 3D model generation, involves fusing multi-channel color point cloud data. During the fusion process, a duplicate data removal network is established to remove duplicate multi-channel color point cloud data. The steps for establishing the duplicate data removal network are as follows:

[0102] Step A1, Feature Representation: The feature representation of the multi-channel color point cloud data is learned through a convolutional neural network. The feature representation of the multi-channel color point cloud data is as follows:

[0103]

[0104] In the formula: G M This represents the total feature representation of multi-channel color point cloud data, where ReLU is the ReLU activation function. This is the first feature representation of multi-channel color point cloud data. This represents the p-th feature of the multi-channel color point cloud data;

[0105] Step A2 involves constructing a deduplication network to remove duplicate multi-channel color point cloud data. The deduplication network is represented as follows:

[0106]

[0107] Where: PLG MJ The results of the deduplication network are represented as follows: Q represents the total number of duplicate multi-channel color point cloud data, and w u Let G be the weight matrix. M represents the overall feature representation of multi-channel color point cloud data, and softmax is the softmax activation function;

[0108] Step A3, calculate the network loss function: calculate the loss function of the deduplication network using cross-entropy loss;

[0109] Step A4, Point Cloud Fusion: The multi-channel color point cloud data is fused. The fused multi-channel color point cloud data is represented as follows:

[0110]

[0111]

[0112] Where: RXH fused For the representation of the fused multi-channel color point cloud data, RXH n The data after removing duplicate multi-channel color point cloud data, Qp is the number of data points after removing duplicate multi-channel color point cloud data, ω n This represents the weight of the nth channel.

[0113] In this embodiment, by establishing a duplicate data removal network, duplicate information in multi-channel color point cloud data can be effectively removed, improving the quality and uniqueness of the data. This helps to generate more accurate and representative 3D models and enhances the expressive power and reliability of multi-channel color point cloud data during the fusion process.

[0114] In this embodiment of the invention, step A3 calculates the network loss function. The loss function of the deduplication network is calculated using cross-entropy loss. The formula for the loss function of the deduplication network is as follows:

[0115] LCE=-∑ u PLG MJ *lg(p u );

[0116] In the formula: LCE is the loss function result of the deduplication network, p u This represents the probability distribution of the laser filtering network results.

[0117] By minimizing the cross-entropy loss function, the deduplication network can learn to adjust its weights and parameters to make the network's output closer to the true label, which helps improve the network's accuracy and robustness, and helps to better understand and capture duplicate data in multi-channel color point cloud data, thus better performing the task of deduplication.

[0118] In this embodiment of the invention, step S1, laser signal feature extraction, involves learning the feature representation of multi-channel color point cloud data through a convolutional neural network. The specific steps are as follows:

[0119] Step B1, Data Processing: Divide the multi-channel color point cloud data into a training signal set and a test signal set;

[0120] Step B2, construct the convolutional neural network model: The convolutional neural network model includes several input layers, convolutional layers, pooling layers, and fully connected layers;

[0121] Step B3, Input Layer Construction: The size of the input layer depends on the data dimension of the multi-channel color point cloud data. When the multi-channel color point cloud data is single-channel with a width of W and a height of H, the shape of the input layer is W×H×1. The construction of the input layer is based on the dimension of the multi-channel color point cloud data to provide an appropriate input shape, so that the model can effectively process the point cloud data.

[0122] Step B4, Convolutional Layer Construction: The convolutional layer uses filters to perform convolution operations on the input multi-channel color point cloud data and adds bias. The formula for calculating the output feature map of the convolutional layer is:

[0123] The output multi-channel color point cloud data feature map = ∑[input multi-channel color point cloud data * filter] + bias. By learning the local features of the multi-channel color point cloud data through convolution operations and activation functions, it helps to extract the spatial structure and key features in the data.

[0124] Step B5, Pooling layer construction: The pooling layer downsamples the feature map of the multi-channel color point cloud data output by the convolutional layer, thereby reducing the dimension and number of parameters of the multi-channel color point cloud data, reducing the dimension of the feature map, accelerating the calculation and reducing the risk of overfitting. Downsampling the output of the convolutional layer can reduce the dimension and number of parameters of the multi-channel color point cloud data and improve the calculation efficiency.

[0125] Step B6, Fully Connected Layer Construction: The multi-channel color point cloud data feature map output from the pooling layer is unfolded into a vector, and calculated by weighted summation plus bias. The output formula of the fully connected layer is:

[0126] The p-th feature representation of the output multi-channel color point cloud data is calculated as ∑[the p-th feature vector of the input multi-channel color point cloud data * weight] + bias.

[0127] In this embodiment, step five, model quality assessment, combines steps two and three to evaluate the quality of the generated 3D model. A 3D model quality assessment model is established using laser loss prediction, noise reduction level, model geometric quality indicators, and average facet area. The formula for the 3D model quality assessment model is:

[0128] W = αAEL * AFL - Z LOSS -CZX;

[0129] Where: W is the quality assessment value of the 3D model, α is the quality control factor, AEL is the predicted laser loss value, AFL is the degree of noise reduction, and Z... LOSS CZX is the geometric quality index of the model, and CZX is the average area of ​​the model's facets.

[0130] This system comprehensively considers laser loss prediction, noise reduction, model geometry quality, and average patch area to provide a holistic assessment of the overall 3D model quality. Quality control factors adjust the weights of different factors in the assessment based on requirements, making the model quality evaluation more aligned with practical needs. The 3D model quality assessment values ​​can guide subsequent optimization efforts, thereby improving the overall 3D model quality. This avoids the limitations of a single indicator and more comprehensively reflects the quality status during the 3D model generation process, thus enhancing the accuracy and usability of the generated 3D model.

[0131] In one embodiment, the formula for calculating the model's geometric quality index is as follows:

[0132]

[0133] In the formula: Z LOSS The geometric quality index of the model, where E is the total number of edges on the surface of the 3D model, and Length(e) is the geometric quality index of the model. r ) represents the length of the surface edge of the 3D model.

[0134] In one embodiment, the formula for calculating the average patch area of ​​the model is:

[0135]

[0136] In the formula: CZX is the average area of ​​the model facets, F is the total number of facets in the 3D model, and Area(f d ) represents the area of ​​a facet in the 3D model.

[0137] Another embodiment of the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-described method.

[0138] Another embodiment of the present invention provides a computer-readable storage medium storing computer program instructions thereon, characterized in that the computer program instructions, when executed by a processor, implement the steps of the above-described method.

[0139] The technical solution of this invention acquires multi-channel color point cloud data using a multispectral lidar and performs noise reduction processing; analyzes laser loss caused by atmospheric absorption and scattering, and constructs a laser loss prediction model, which helps in subsequent correction and model evaluation; by calculating the average, minimum, and maximum point spacing of the multi-channel color point cloud data, a noise reduction evaluation model is constructed, which helps improve the quality of the multi-channel color point cloud data; the multi-channel color point cloud data is fused, and a three-dimensional model is generated using the fused color point cloud data. A three-dimensional model quality evaluation model is established to evaluate the quality of the generated three-dimensional model, avoiding the influence of noise and laser transmission loss, and enhancing the realism and accuracy of the generated three-dimensional model.

[0140] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0141] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for fusion of multiple channels via unidirectional atmospheric laser transmission, characterized in that, Includes the following steps: S1 acquires multi-channel color point cloud data through multispectral lidar and performs noise reduction processing; S2 analyzes the laser loss caused by atmospheric absorption and scattering, and constructs a laser loss prediction model to predict the loss of laser signals of different wavelengths during transmission. Specifically, the laser loss prediction model is as follows: Z LOSS =(LOSS abp +LOSS sct )*d; In the formula, Z LOSS LOSS is the predicted value for laser loss. abp LOSS is the amount of atmospheric absorption loss. sct Where d is the atmospheric scattering loss and d is the transmission distance; S3 constructs a noise reduction assessment model by calculating the average, minimum, and maximum point spacing of multi-channel color point cloud data, which is used to evaluate the noise reduction level of channel color point cloud data. S4 fuses multi-channel color point cloud data and uses the fused color point cloud data to generate a 3D model. S5 evaluates the quality of the generated 3D model based on the model constructed in steps S2 and S3.

2. The atmospheric laser unidirectional transmission multi-channel fusion method according to claim 1, wherein, The noise cancellation assessment model described in S3 is as follows: In the formula, CZX represents the noise reduction level, MZX represents the average point spacing of the multi-channel color point cloud data, MinZX represents the minimum point spacing of the multi-channel color point cloud data, and MaxZX represents the maximum point spacing of the multi-channel color point cloud data.

3. The atmospheric laser unidirectional transmission multi-channel fusion method according to claim 1, wherein, Step S4, fusing multi-channel color point cloud data, includes establishing a duplicate data removal network during the fusion process to remove duplicate multi-channel color point cloud data. The establishment of the duplicate data removal network includes: S41 learns the feature representation of multi-channel color point cloud data through a convolutional neural network. The feature representation of multi-channel color point cloud data is as follows: In the formula, G M This represents the total feature representation of multi-channel color point cloud data, where ReLU is the ReLU activation function. This is the first feature representation of multi-channel color point cloud data. This represents the p-th feature of the multi-channel color point cloud data; S42 removes duplicate multi-channel color point cloud data by constructing a deduplication network. The deduplication network is represented as follows: In the formula, PLG MJ The results of the deduplication network are represented as follows: Q represents the total number of duplicate multi-channel color point cloud data, and w u Let G be the weight matrix. M represents the feature set of multi-channel color point cloud data, and softmax is the softmax activation function. S43 calculates the loss function of the deduplication network using cross-entropy loss; S4 fuses multi-channel color point cloud data, and the fused multi-channel color point cloud data is represented as follows: In the formula, RXH fused For the representation of the fused multi-channel color point cloud data, RXH n The data after removing duplicate multi-channel color point cloud data, Qp is the number of data points after removing duplicate multi-channel color point cloud data, ω n This represents the weight of the nth channel.

4. The atmospheric laser unidirectional transmission multi-channel fusion method according to claim 3, wherein, In step S43, the network loss function is calculated. The loss function of the deduplication network is calculated using cross-entropy loss. The formula for the loss function of the deduplication network is as follows: LCE=-∑ u PLG MJ *lg(p u ); In the formula, LCE is the loss function result of the deduplication network, p u This represents the probability distribution of the laser filtering network results.

5. The atmospheric laser unidirectional transmission multi-channel fusion method according to claim 1, wherein, The quality evaluation of the generated 3D model in S5 includes: establishing a 3D model quality evaluation model based on laser loss prediction, noise reduction degree, model geometric quality index, and average patch area of ​​the model. The formula for the 3D model quality evaluation model is: W=αAEL*AFL-Z LOSS -CZX; In the formula, W is the quality assessment value of the three-dimensional model, α is the quality control factor, AEL is the predicted laser loss value, AFL is the degree of noise reduction, and Z... LOSS CZX is the geometric quality index of the model, and CZX is the average area of ​​the model's facets.

6. The atmospheric laser unidirectional transmission multi-channel fusion method according to claim 3, wherein, Step S41, which involves learning the feature representation of multi-channel color point cloud data using a convolutional neural network, specifically includes: S411 divides multi-channel color point cloud data into training signal set and test signal set; S412 Constructing a Convolutional Neural Network Model: A convolutional neural network model includes several input layers, convolutional layers, pooling layers, and fully connected layers; S413 constructs the input layer, the size of which depends on the data dimension of the multi-channel color point cloud data; S414 constructs a convolutional layer, wherein the convolutional layer uses filters to perform convolution operations on the input multi-channel color point cloud data and adds a bias, and the formula for calculating the output feature map of the convolutional layer is: Output multi-channel color point cloud data feature map = ∑[input multi-channel color point cloud data * filter] + bias; S415, Construct a pooling layer, which is used to downsample the feature map of the multi-channel color point cloud data output by the convolutional layer to reduce the dimension and number of parameters of the multi-channel color point cloud data; S416 constructs a fully connected layer, expands the multi-channel color point cloud data feature map output from the pooling layer into a vector, and calculates it by weighted summation plus bias. The output formula of the fully connected layer is: The p-th feature representation of the output multi-channel color point cloud data is calculated as ∑[the p-th feature vector of the input multi-channel color point cloud data * weight] + bias.

7. The atmospheric laser unidirectional transmission multi-channel fusion method according to claim 1, wherein, The average pixel spacing of the multi-channel color point cloud data is: In the formula: MZX is the average point spacing of the multi-channel color point cloud data, p is the total amount of multi-channel color point cloud data, Pi is the i-th multi-channel color point cloud data, and Pih is the nearest neighbor h of the i-th multi-channel color point cloud data.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to claims 1-7.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the steps of the method described in claims 1-7.

Images