Laser radar cloud layer detection algorithm based on deep learning, storage medium and electronic device
Through the deep learning-based lidar cloud detection algorithm, the U2Net model is optimized and the FocalLoss and DiceLoss loss functions are combined, and the problems of low cloud detection accuracy, poor real-time performance and poor adaptability in the existing technology are solved, and efficient and accurate cloud detection is achieved.
Patent Information
- Application Number
- CN202411702454.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-26
- Publication Date
- 2025-03-18
AI Technical Summary
In the prior art, cloud detection accuracy is low, real-time performance is poor, and adaptability is poor, resulting in frequent missed and missed detection.
The deep learning-based lidar cloud detection algorithm is adopted, including lidar backscattered signal preprocessing, establishing cloud detection data sets, building and optimizing U2Net models, training models, and combining loss functions with FocalLoss and DiceLoss, realizing cloud detection of real-time lidar signals.
It realizes efficient and accurate cloud detection, can adapt to different meteorological conditions, significantly reducing the occurrence of missed and missed inspections.
Smart Images
Figure CN119649110A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of laser radar signal processing, and in particular to a laser radar cloud detection algorithm, storage medium and electronic device based on deep learning. Background Art
[0002] With the development of LiDAR technology, LiDAR has been widely used in atmospheric detection. As an active detection device, LiDAR emits a beam of laser into the atmosphere. During the transmission of the laser in the atmosphere, various media in the atmosphere, such as aerosols and air molecules, will interact with it by scattering and absorption. After the optical telescope receives the scattered signal and processes the signal, it uses the interaction between the laser and the atmospheric medium to produce different effects and phenomena to obtain high-resolution atmospheric information such as cloud height and aerosol concentration. The current cloud detection methods mainly rely on fixed thresholds and traditional signal processing techniques. However, these traditional cloud detection methods, such as differential zero crossing method (DEM) and histogram equalization method (VDEM), cannot accurately detect cloud information, and often miss or over-detect clouds. In addition, these methods are limited by the fact that they rely on threshold methods, which usually require manual setting of reasonable threshold ranges. Different threshold parameters may need to be set for different LiDAR states and different weather conditions, which greatly increases the uncertainty of cloud detection. Summary of the invention
[0003] In order to solve the above technical problems, the present invention proposes a deep learning-based lidar cloud detection algorithm, storage medium and electronic device. The specific technical solutions are as follows: The present application discloses an inventive concept, wherein a deep learning-based lidar cloud detection algorithm comprises: LiDAR backscatter signal preprocessing; Establish a lidar cloud detection dataset; Build a lidar cloud detection model; Train the lidar cloud detection model and save the training results; Preprocess the laser radar signals collected in real time; The real-time preprocessed signal is input into the trained lidar cloud detection model for cloud detection.
[0004] Furthermore, the laser radar backscatter signal preprocessing includes: Get the laser radar background signal; convert the laser radar backscatter signal The far-end signal is used as the background signal of the laser radar backscatter ; Remove the laser radar background signal; convert the laser radar backscatter signal and the background signal Substitute the following formula to obtain the laser radar background-removed backscatter signal: :
[0005] in, It's distance. is the background signal average; Calculate the logarithmic backscatter signal :The laser radar background backscatter signal Substitute the following formula to obtain the logarithmic backscatter signal:
[0006] Normalize the logarithmic backscatter signal; Substitute the following formula to obtain the logarithmic backscatter normalized signal :
[0007] in, is the mean of the logarithmic backscattered signal, and STD is the standard deviation of the logarithmic backscattered signal.
[0008] Furthermore, in step S3, a lidar cloud detection model is constructed, including: The lidar cloud detection model obtained after optimization based on U2Net can adapt to the lidar signal. The optimization part is to replace the two-dimensional convolution layer in U2Net with a one-dimensional convolution layer, change the two-dimensional batch normalization to a one-dimensional batch normalization, change the two-dimensional pooling layer to a one-dimensional pooling layer, and change the bilinear interpolation to a linear interpolation.
[0009] Furthermore, the specific steps of establishing a lidar cloud detection dataset are as follows: Annotate the vertical information of clouds in the lidar backscatter signal to obtain cloud masks ; The logarithmic backscatter normalized signal and cloud mask Generate dataset samples; In the data set sample Samples with all values of 0, i.e., cloudless samples, are screened to account for the set ratio of the total samples; The data set samples are divided into training set and validation set according to the set ratio as the cloud detection data set.
[0010] Furthermore, the training of the laser radar cloud detection model and saving the training results include: Set training parameters; Define the loss function; Iteratively train the model; Evaluate model performance metrics.
[0011] Furthermore, the definition of the loss function includes: The loss function is a combination of FocalLoss and DiceLoss. The specific formula is as follows:
[0012] Among them, λ1 and λ2 are weight parameters used to control the influence of Focal Loss and Dice Loss in the combined loss. represents the predicted probability of the positive class, that is, the probability that the model predicts it to be a cloud, α is the weight factor of the positive class, is represented as the focusing parameter, Indicates The probability that a value is predicted to be a positive class, Indicates The true label of the value (0 or 1), is the total signal length.
[0013] Furthermore, the evaluation model performance indicators include: It is obtained by calculating the mean intersection over union (MIOU), mean pixel accuracy (MPA) and F1-Score; the normalized logarithmic backscattered signal in the validation set samples The cloud mask is input into the trained cloud detection model for detection, and then the predicted cloud mask is compared with the real cloud mask in the training set samples. The pixel binary classification confusion matrix is first calculated, and then the performance index is calculated through the confusion matrix.
[0014] Furthermore, the pixel binary classification confusion matrix contains four types of information: the number of samples correctly predicted as positive (TP), the number of samples actually positive but predicted as negative (FN), the number of samples actually negative but predicted as positive (FP), and the number of samples correctly predicted as negative (TN); The mean intersection over union (MIOU) measures the overlap between the region predicted by the model and the true label, and its formula is as follows:
[0015] The mean pixel accuracy (MPA) measures the accuracy of the model's predictions for each category and averages the accuracy of all categories. The formula is as follows:
[0016] The F1-Score is the harmonic mean of precision and recall. The calculation formula of the F1-Score is as follows:
[0017]
[0018] .
[0019] The present application also discloses another inventive concept, a storage medium, in which a computer program is stored, wherein the computer program is configured to implement the above-mentioned deep learning-based lidar cloud detection algorithm when running.
[0020] The present application also discloses another inventive concept, an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the above-mentioned deep learning-based lidar cloud detection algorithm.
[0021] The advantages of the present invention are: (1) The present disclosure aims to solve the problems of low cloud detection accuracy, poor real-time performance, and poor adaptability in the prior art. By using a deep learning model, the present application can achieve efficient and accurate cloud detection, and can adapt to environmental changes under different meteorological conditions, effectively reducing the occurrence of missed detections and false detections.
[0022] (2) The deep learning model architecture of U2Net is optimized by changing its two-dimensional convolution, batch normalization, pooling layer and interpolation layer into a one-dimensional structure. This optimization not only maintains the original multi-scale feature extraction capability of U2Net, but also makes it more adaptable to the one-dimensional signal of LiDAR, thus improving the boundary capture accuracy and applicability.
[0023] (3) To address the imbalance problem of positive and negative samples in the lidar cloud detection dataset, the sample imbalance problem is alleviated by making the cloud-free samples account for no more than a set value of the total samples when generating samples, and the set value is less than 50%. The Dice Loss function and the Focal Loss function are combined to achieve this goal. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 A flowchart of a laser radar cloud detection method provided in an embodiment of the present application; Figure 2 A flowchart of preprocessing a laser radar backscatter signal provided in an embodiment of the present application; Figure 3A process for establishing a method for a lidar cloud detection data set provided in an embodiment of the present application; Figure 4 A schematic diagram of the structure of a laser radar cloud detection model provided in an embodiment of the present application; Figure 5 A flowchart of a model training method provided in an embodiment of the present application; Figure 6 Part (a) verifies the detection performance of the embodiment of the present application at a low signal-to-noise ratio; Part (b) verifies the detection performance of the embodiment of the present application at a high signal-to-noise ratio; Figure 7 Part (a) is a distance-corrected signal diagram of the lidar backscatter signal for one day; part (b) is the result of annotating the vertical cloud layer in the lidar backscatter signal; part (c) is the result of detecting the lidar backscatter signal for that day using the differential zero-crossing method (DEM); part (d) is the result of detecting the lidar backscatter signal for that day using the value distribution equalization method (VDEM); part (e) is the result of detecting the lidar backscatter signal for that day using a lidar cloud detection model provided in an embodiment of the present application. DETAILED DESCRIPTION
[0025] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.
[0026] In the field of computer vision, deep learning technology is used to perform semantic segmentation on images, making it possible to accurately segment image targets in complex scenes. The application prospects of deep learning technology in the field of cloud detection are very broad. By applying it to the processing of LiDAR backscatter signals, it can effectively overcome the shortcomings of traditional methods and realize automatic and efficient extraction of cloud information.
[0027] Based on this, this application discloses a deep learning-based lidar cloud detection algorithm, including: S1, preprocessing of laser radar backscatter signal; S2, establish a lidar cloud detection dataset; S3, build a lidar cloud detection model; S4, training the lidar cloud detection model and saving the training results; S5, preprocessing the laser radar signal collected in real time; S6. Input the real-time preprocessed signal into the trained lidar cloud detection model for cloud detection.
[0028] As a further preferred embodiment of the present invention: in step S1, the collected laser radar data is preprocessed, and the specific steps are as follows: S11, obtain the laser radar background signal; convert the laser radar backscatter signal The far-end signal is used as the background signal of the laser radar backscatter ; By obtaining the lidar background noise, the long-distance noise baseline can be extracted, which facilitates the removal of irrelevant atmospheric background information and thus enhances the signal-to-noise ratio.
[0029] S12, remove the laser radar background signal; and the background signal Substitute the following formula to obtain the laser radar background-removed backscatter signal: :
[0030] in, It's distance. is the background signal average; S13. Calculate the logarithmic backscatter signal :The laser radar background backscatter signal Substitute the following formula to obtain the logarithmic backscatter signal:
[0031] In this implementation, the logarithmic transformation can reduce the dynamic range of the signal, enhance the weak features of the distance signal, and make the cloud boundary more prominent.
[0032] S14, normalizing the logarithmic backscatter signal; normalizing the logarithmic backscatter signal Substitute the following formula to obtain the logarithmic backscatter normalized signal :
[0033] in, is the average value of the logarithmic backscatter signal, and STD is the standard deviation of the logarithmic backscatter signal. Standardization eliminates systematic deviations caused by measurement equipment or environmental conditions, making the signal distribution more uniform, which helps improve the accuracy and consistency of detection.
[0034] After preprocessing the lidar backscatter signal, it can be used to remove noise interference in the signal and highlight the signal characteristics.
[0035] As a further preferred embodiment of the present invention: in step S2, a laser radar cloud detection data set is established, and the specific steps are as follows: S21. Annotate the cloud vertical information in the lidar backscatter signal to obtain the cloud mask ; S22, using the logarithmic backscattering normalized signal and cloud mask Generate dataset samples; S23. The samples with all zero values, i.e., cloud-free samples, are screened to a set ratio of the total samples, which is 20% in one embodiment, 15% in another embodiment, 25% in another embodiment, and 30% in another embodiment; S24, dividing the data set samples according to a set ratio to form a training set and a validation set as a cloud detection data set. In one embodiment, the training set accounts for 80% and the validation set accounts for 20%.
[0036] The cloud detection dataset formed by this step can be used to train the lidar cloud detection model.
[0037] As a further preferred embodiment of this invention: in step S3, a laser radar cloud layer detection model is constructed. In one embodiment, the constructed laser radar detection model is as follows: Figure 4 As shown, the LiDAR cloud detection model is a LiDAR cloud detection model obtained after optimization based on U2Net, so that it can adapt to LiDAR signals. The optimization part is to replace the two-dimensional convolution layer in U2Net with a one-dimensional convolution layer, change the two-dimensional batch normalization to a one-dimensional batch normalization, change the two-dimensional pooling layer to a one-dimensional pooling layer, and change the bilinear interpolation to a linear interpolation. U2net was originally applied to target detection in two-dimensional images, while LiDAR signals are one-dimensional signals. This application innovatively optimizes the U2net model structure so that it can process LiDAR signals and complete real-time detection of clouds.
[0038] like Figure 5 As shown, as a further preferred embodiment of this embodiment: in step S4, the laser radar cloud detection model is trained and the training results are saved, and the specific steps are as follows: S41, setting training parameters, wherein the training parameters include learning rate, learning rate scheduling method, number of iterations, iteration batch size and optimizer; In this embodiment, by setting a suitable learning rate and coordinating the learning rate scheduling method, the instability in the training process can be effectively avoided, so that the model gradually converges to the optimal solution. The appropriate number of iterations and batch size settings ensure that the model can extract rich features during the learning process and improve the accuracy of cloud layer detection. Using an optimizer can accelerate the convergence of the model, improve the detection performance in complex environments, and make the model more reliable in practical applications.
[0039] As an optional implementation, the present application embodiment sets the learning rate to 0.0001, and in this process, the learning rate is gradually changed by the learning rate scheduler. The learning rate scheduling method uses the cosine annealing method, the number of iterations is 100 times, the batch size is 8, and the optimizer is the Adam optimizer.
[0040] S42. Define a loss function; the loss function is used to guide model training. By minimizing the loss function, the model can continuously adjust its own parameters to better adapt to the training data, thereby improving its accuracy.
[0041] The loss function defined in the embodiment of the present application is a combination of FocalLoss and DiceLoss, and the specific formula is as follows:
[0042] Among them, λ1 and λ2 are weight parameters used to control the influence of Focal Loss and Dice Loss in the combined loss. represents the predicted probability of the positive class, that is, the probability that the model predicts it to be a cloud, α is the weight factor of the positive class, is represented as the focusing parameter, Indicates The probability that a value is predicted to be a positive class, Indicates The true label of the value (0 or 1), is the total signal length.
[0043] As an optional implementation, in the embodiment of the present application, λ1 is 0.5, λ2 is 0.5, and α is 0.25. is 2.
[0044] S43, iterative training model.
[0045] In this implementation, for each iteration, the training set loss is determined based on the cloud detection prediction results and loss function output by the current iteration batch, and the lidar cloud detection model is adjusted using the back propagation principle according to the optimization strategy corresponding to the optimizer. When the number of iterations is reached, the training is terminated and the trained lidar cloud detection model is saved. S44, evaluation model performance indicators; As an optional implementation, the performance indicators of the model evaluated in this application are obtained by calculating the mean intersection over union (MIOU), mean pixel accuracy (MPA) and F1-Score. The cloud mask is input into the trained cloud detection model for detection, and then the predicted cloud mask is compared with the real cloud mask in the training set samples. The pixel binary classification confusion matrix is first calculated, and then the performance index is calculated through the confusion matrix.
[0046] The pixel binary classification confusion matrix contains four types of information: the number of samples correctly predicted as positive (TP), the number of samples that are actually positive but predicted as negative (FN), the number of samples that are actually negative but predicted as positive (FP), and the number of samples correctly predicted as negative (TN).
[0047] Mean intersection over union (MIOU) is a metric commonly used to evaluate the performance of object detection models. It measures the degree of overlap between the region predicted by the model and the true label. The value of MIOU ranges from 0 to 1, and the closer to 1, the better the performance of the model. Its formula is as follows:
[0048] The average pixel accuracy (MPA) measures the accuracy of the model's predictions for each category and averages the accuracy of all categories. The formula is as follows:
[0049] The F1-Score is the harmonic mean of precision and recall, which can effectively balance these two indicators. The value range of the F1-Score is between 0 and 1, where 1 represents the best performance and 0 represents the worst performance. The F1-Score calculation formula is as follows:
[0050]
[0051]
[0052] S5, preprocessing the laser radar signal collected in real time; During implementation, the actually collected laser radar signal is preprocessed as described in step S1.
[0053] S6. Input the real-time preprocessed signal into the trained lidar cloud detection model for cloud detection.
[0054] To verify the applicability of this application method, the following verification is performed: Case verification: like Figure 6 As shown, part (a) and part (b) respectively represent the results detected by each detection algorithm under low and high signal-to-noise ratio conditions. It can be seen from part (a) that under low signal-to-noise ratio conditions, compared with traditional algorithms, the results detected by the algorithm proposed in the embodiment of the present application are closest to the actual annotation results; it can be seen from part (b) that under high signal-to-noise ratio conditions, the differential zero crossing method (DEM) over-detects the cloud layer, and the value distribution equalization method (VDEM) misses the cloud layer. The algorithm proposed in the embodiment of the present application can accurately detect the cloud layer.
[0055] Indicator verification: A total of 288 profiles of the laser radar on August 27, 2022 were selected to detect the vertical structure of the cloud layer using traditional algorithms (differential zero crossing method, histogram equalization method) and the algorithm of the embodiment of the present application. The evaluation indicators are shown in Table 1. The detection results are shown in Figure 7 shown.
[0056] method MIOU(%) MPA(%) F1-Score Differential Zero Crossing Method 0.72947 0.82569 0.82248 Value distribution balance method 0.88930 0.89965 0.93853 The method of this embodiment 0.91842 0.95044 0.95600 Table 1 Combined with Table 1 and Figure 7 It can be seen that the cloud detection method provided in the embodiment of the present application is superior to the traditional algorithm in terms of MIOU, MPA and F1-Score evaluation indicators. Figure 7 In part (c), it can be seen that the differential zero crossing method (DEM) has obvious under-detection of clouds in this area. Figure 7 In part (d), it can be seen at B and C that although the value distribution balancing method detected the clouds in this area, a single cloud layer was detected as multiple cloud layers and a small number of cloud layers were missed. The detection algorithm provided in the embodiment of the present application can avoid these problems and is closest to the actual labeling results, achieving the highest accuracy compared to traditional algorithms.
[0057] Based on the same inventive concept, the present invention also proposes a readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it can implement the deep learning-based lidar cloud detection algorithm described in any one of the above-mentioned feature descriptions.
[0058] The readable storage medium can be a tangible device that can keep and store the instructions used by the instruction execution device, such as but not limited to an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device or any suitable combination of the above. The more specific example (non-exhaustive list) of the readable storage medium includes: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a convex structure in a groove on which instructions are stored, and any suitable combination of the above. The computer program described here can be downloaded from the readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device through a network, such as the Internet, a local area network, a wide area network and / or a wireless network. The network can include copper transmission cables, optical fiber transmission, wireless transmission, routers, firewalls, switches, gateway computers and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer program from the network and forwards the computer program for storage in a readable storage medium in each computing / processing device. The computer program for performing the operation of the present invention may be an assembly instruction, an instruction set architecture (ISA) instruction, a machine instruction, a machine-dependent instruction, a microcode, a firmware instruction, a state setting data, or a source code or object code written in any combination of one or more programming languages, including object-oriented programming languages such as Smalltalk, C++, etc., and conventional procedural programming languages such as "C" language or similar programming languages. The computer program may be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect through the Internet). In some embodiments, by utilizing the state information of a computer program to personalize an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit can execute computer-readable program instructions to implement various aspects of the present invention.
[0059] Here, various aspects of the present invention are described with reference to the flowchart and / or block diagram of the method, system and computer program product according to the embodiment of the present invention. It should be understood that each box of the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by a computer program. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer or other programmable data processing device, so as to produce a machine so that when these programs are executed by the processor of the computer or other programmable data processing device, a device for implementing the function / action specified in one or more boxes in the flowchart and / or block diagram is generated. These computer programs can also be stored in a readable storage medium, which makes the computer, programmable data processing device and / or other equipment work in a specific manner, so that the readable storage medium storing the computer program includes a manufactured product, which includes instructions for implementing various aspects of the function / action specified in one or more boxes in the flowchart and / or block diagram.
[0060] The computer program may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operational steps are executed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the computer program executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.
[0061] An embodiment of the present invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the steps in any of the above-mentioned embodiments of the deep learning-based lidar cloud detection algorithm.
[0062] In the description of this specification, the description with reference to the terms "some embodiments" or "examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiments or examples are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.
[0063] The above is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any technician in the relevant technical field, without departing from the scope of the technical solution of the present invention, makes any form of equivalent replacement or modification to the technical solution and technical content disclosed in the present invention, which does not depart from the content of the technical solution of the present invention and still falls within the protection scope of the present invention.
Claims
1. A deep learning-based lidar cloud detection algorithm, characterized in that: include: LiDAR backscatter signal preprocessing; Establish a lidar cloud detection dataset; Build a lidar cloud detection model; Train the lidar cloud detection model and save the training results; Preprocess the laser radar signals collected in real time; The real-time preprocessed signal is input into the trained lidar cloud detection model for cloud detection.
2. The deep learning-based lidar cloud detection algorithm according to claim 1, characterized in that: The laser radar backscatter signal preprocessing includes: Get the laser radar background signal; convert the laser radar backscatter signal The far-end signal is used as the background signal of the laser radar backscatter ; Remove the laser radar background signal; convert the laser radar backscatter signal and the background signal Substitute the following formula to obtain the laser radar background-removed backscatter signal: : in, It's distance. is the background signal average; Calculate the logarithmic backscatter signal :The laser radar background backscatter signal Substitute the following formula to obtain the logarithmic backscatter signal: Normalize the logarithmic backscatter signal; Substitute the following formula to obtain the logarithmic backscatter normalized signal : in, is the mean of the logarithmic backscattered signal, and STD is the standard deviation of the logarithmic backscattered signal.
3. The deep learning-based lidar cloud detection algorithm according to claim 1 or 2, characterized in that: In step S3, a lidar cloud detection model is constructed, including: The lidar cloud detection model obtained after optimization based on U2Net can adapt to the lidar signal. The optimization part is to replace the two-dimensional convolution layer in U2Net with a one-dimensional convolution layer, change the two-dimensional batch normalization to a one-dimensional batch normalization, change the two-dimensional pooling layer to a one-dimensional pooling layer, and change the bilinear interpolation to a linear interpolation.
4. The deep learning-based lidar cloud detection algorithm according to claim 1, characterized in that: The specific steps of establishing a lidar cloud detection dataset are as follows: Annotate the cloud vertical information in the lidar backscatter signal to obtain the cloud mask ; The logarithmic backscatter normalized signal and cloud mask Generate dataset samples; In the data set sample Samples with all values of 0, i.e., cloudless samples, are screened to account for the set ratio of the total samples; The data set samples are divided into training set and validation set according to the set ratio as the cloud detection data set.
5. The deep learning-based lidar cloud detection algorithm according to claim 1, characterized in that: The step of training the laser radar cloud detection model and saving the training results includes: Set training parameters; Define the loss function; Iteratively train the model; Evaluate model performance metrics.
6. The deep learning-based lidar cloud detection algorithm according to claim 5, characterized in that: The definition of the loss function includes: The loss function is a combination of FocalLoss and DiceLoss. The specific formula is as follows: Among them, λ1 and λ2 are weight parameters used to control the influence of Focal Loss and Dice Loss in the combined loss. represents the predicted probability of the positive class, that is, the probability that the model predicts it to be a cloud, α is the weight factor of the positive class, is represented as the focusing parameter, Indicates The probability that a value is predicted to be a positive class, Indicates The true label of the value (0 or 1), is the total signal length.
7. The deep learning-based lidar cloud detection algorithm according to claim 1, characterized in that: The evaluation model performance indicators include: It is obtained by calculating the mean intersection over union (MIOU), mean pixel accuracy (MPA) and F1-Score; the normalized logarithmic backscattered signal in the validation set samples The cloud mask is input into the trained cloud detection model for detection, and then the predicted cloud mask is compared with the real cloud mask in the training set samples. The pixel binary classification confusion matrix is first calculated, and then the performance index is calculated through the confusion matrix.
8. The deep learning-based lidar cloud detection algorithm according to claim 7, characterized in that: The pixel binary classification confusion matrix contains four types of information: the number of samples correctly predicted as positive (TP), the number of samples actually positive but predicted as negative (FN), the number of samples actually negative but predicted as positive (FP), and the number of samples correctly predicted as negative (TN); The mean intersection over union (MIOU) measures the overlap between the region predicted by the model and the true label, and its formula is as follows: The average pixel accuracy (MPA) measures the accuracy of the model's predictions for each category and averages the accuracy of all categories. The formula is as follows: The F1-Score is the harmonic mean of precision and recall. The calculation formula of the F1-Score is as follows: 。 9. A storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program is configured to implement the deep learning-based lidar cloud detection algorithm described in any one of claims 1 to 8 when running.
10. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to implement the deep learning-based lidar cloud detection algorithm described in any one of claims 1 to 8.