A data processing method for identifying severe weather based on lidar

Abstract

The present invention relates to a data processing method for identifying severe weather based on laser radar, and belongs to the field of intelligent traffic radar data processing technology. The present invention combines laser radar to obtain trajectory data of sunny, rainy, and snowy weather, and proposes a method for identifying severe weather types using laser radar point cloud features, which expands the application range of laser radar, provides good weather reference conditions for laser radar to monitor vehicle roadside data, enriches the spatial characteristics of target trajectory information, and has strong practicality. The present invention fully extracts the extracted laser radar point cloud data, realizes the process from image input to feature extraction and comparison to output weather type and weather intensity level, directly outputs weather results end-to-end and predicts the duration of severe weather, and can fine-grainedly determine the weather conditions and duration of specific roads, making it more practical.

Description

Technical Field

[0001] The present invention relates to a data processing method for identifying severe weather based on a laser radar, and belongs to the technical field of intelligent traffic radar data processing. Background Art

[0002] Weather recognition and classification is a key classification problem in computer data processing. Weather conditions are a significant factor affecting vehicle safety, with traffic accidents caused by inclement weather such as rain and snow accounting for over 30% of all accidents. Severe rain and snow severely impact the normal operation of road traffic, particularly during rush hour, causing congestion to skyrocket. Highway pavement weather recognition specifically identifies inclement weather conditions on highways. By promptly assessing weather conditions through road imagery, it facilitates better traffic management. In recent years, numerous researchers have been conducting research in the fields of smart transportation and autonomous driving, but safe driving in inclement weather remains a challenge. Real-time monitoring and identification of inclement weather conditions, along with timely feedback to drivers, can further improve driving safety and lay the foundation for research related to vehicle safety in inclement weather.

[0003] Currently, there are two main methods for identifying severe weather: the first is to set up static meteorological monitoring stations on the road to collect data such as road water and snow depth, temperature, visibility, etc., to monitor and identify weather conditions on the road section. This method starts from the overall effect of rainy and snowy weather, ignores the dynamic changes of rain and snow during rainfall, and is difficult to accurately describe the dynamic relationship between traffic characteristics and rainfall and snowfall during rainfall and snowfall. This method is not suitable for related research in severe weather environments during driving; the second is through image processing and deep learning methods, especially in the aspect of human-computer interaction. The use of digital images for information processing and classification technology is specific to highway road weather, using surveillance videos or detecting the current weather category, but it is difficult to quantify the severity of the weather, and still faces refinement challenges in the task of road weather image classification. Summary of the Invention

[0004] In view of the shortcomings of existing severe weather recognition algorithms, the purpose of the present invention is to solve the shortcomings of the above recognition algorithms and propose a severe weather recognition method based on lidar.

[0005] As an important sensor, lidar is widely used in autonomous driving, roadside perception, high-precision mapping, etc. By emitting and receiving laser beams, lidar can quickly obtain the three-dimensional position point information of the target in space (i.e. point cloud data), and further determine the target's position, size, external contour and other key feature data.

[0006] The technical solutions of the present invention are as follows:

[0007] A data processing method for identifying severe weather based on laser radar comprises the following steps:

[0008] S1: Obtain lidar point cloud signal intensity data of different weather types and preprocess the data. Weather types include sunny, rainy, and snowy. Rainfall is divided into heavy rain, moderate rain, light rain, and drizzle. Snowfall is divided into blizzard, heavy snow, moderate snow, and light snow.

[0009] S2: Construct training and test sets based on the LiDAR point cloud signal intensity data obtained in S1, and construct correlation features between different weather types, i.e., sunny days, rain and snow intensity, and the corresponding state data of the point cloud image;

[0010] S3: Clustering the state features under different weather types to obtain point cloud clusters, and determining the data association features of the point cloud clusters corresponding to different weather types;

[0011] S4: Determine the consistency condition of the data association features of a certain point cloud sample and the trained sunny point cloud cluster, determine the consistency condition of the data association features of a certain point cloud sample and the trained rain or snow point cloud cluster, obtain the level of rain or snowfall and predict the duration of rain or snowfall.

[0012] Preferably, in step S1, the process of preprocessing the data includes:

[0013] S1-1: interpolate missing data and eliminate unreasonable data outside the set response threshold;

[0014] S1-2: Process the point cloud data using a combined noise reduction method to achieve point cloud data processing and target aircraft feature restoration in severe weather conditions;

[0015] The combined denoising method is implemented using a combination of filters, consisting of spatial filtering, ground filtering, voxel filtering, and statistical outlier filtering. Given that filtering algorithms typically require scanning and processing the entire point cloud spatial data, a spatial filtering algorithm is first used to remove irrelevant data from the space. Based on the target point cloud location, unnecessary walls, trees, and other objects are removed from the point cloud space, retaining the target point cloud data for sunny, rainy, and snowy weather conditions.

[0016] This invention improves image quality, removes high-frequency noise and interference, filters out noise points returned by the ground surface and ground objects, maintains point cloud shape characteristics while reducing data processing time complexity, and uses statistics to filter out points outside a threshold. Combined filters can effectively remove points missed by a single filter.

[0017] Preferably, step S2 includes:

[0018] S2-1: At the lidar test point at the same location, select the weather conditions of sunny days, heavy rain, moderate rain, light rain, and drizzle, and the snowfall conditions of blizzard, heavy snow, moderate snow, and light snow. Collect lidar 3D point cloud images respectively. Construct multiple sets of sample data sets for training and testing convolutional neural networks according to the continuous time periods of the same weather at the same location and the different time periods of different weather at the same location.

[0019] S2-2: In each 3D point cloud image collected by S2-1, the center of the image is used as the coordinate origin, and five key orientation points are defined at the upper left, lower left, upper right, lower right and center. A cube with a side length of n units is used as the region of interest. The position of each cube is represented by the coordinates of two diagonal points. Each point is composed of a three-dimensional vector, namely n1(x1, y1, z1), n2(x2, y2, z2)..., that is, the position of the first cube is represented by n12(x1-x2, y1-y2, z1 -z2)…, multiple groups of sample information are input into the convolutional neural network model (CNN). The lidar point cloud images of continuous time periods of the same weather at the same location and different time periods of different weather at the same location are extracted to construct data association feature relationships corresponding to sunny days, heavy rain, moderate rain, light rain, and drizzle in rainy weather, and blizzards, heavy snow, moderate snow, and light snow in snowy weather. The data association features are determined according to the weather conditions, including the threshold range of the total number of point clouds, the reflection intensity value of the point clouds, and the sparse characteristics of the point clouds.

[0020] Convolutional neural networks (CNNs) are suitable for computer vision tasks such as image processing and classification, and are particularly useful for solving image classification problems. Convolutional neural networks typically involve convolution and pooling of a dataset, taking in multi-channel inputs of color images, and then outputting the final result after multiple convolution and pooling cycles.

[0021] In this step, the total number of point clouds, the reflection intensity value of the point clouds, the sparse characteristics of the point clouds (sparseness, spatial distribution), etc. are determined by the output of the neural network. The same weather type will have a dynamic fluctuation range, that is, the threshold range.

[0022] Preferably, step S3 is specifically as follows:

[0023] S3-1: Obtaining a point cloud data set of extracted association information of different weather types, wherein the point cloud data set includes classified weather type point cloud data, determining a target point cloud quantity data and a processing weather type corresponding to the target point cloud data from the classified point cloud data;

[0024] S3-2: The weather types processed include sunny days, heavy rain, moderate rain, light rain, and drizzle in rainy days, and blizzards, heavy snow, moderate snow, and light snow in snowy days. Data is labeled according to weather types, such as 1, 2, 3, and so on. If the data threshold of the associated features is in the same range, that is, the same weather type, it will be labeled with the same number. The same type is grouped together as a cluster, and the target point cloud data is clustered to obtain the clustering results;

[0025] S3-3: After obtaining the clustering results, perform point cloud data visualization, displaying the weather types of the five regions of interest marked on the same point cloud map, and statistically analyzing the dynamic change data association characteristics (i.e., time-frequency status information) of the same region of interest at a continuous time interval, including the total number of point clouds in different regions of interest, the reflection intensity characteristics of point clouds in different regions of interest, and the distribution pattern of point cloud sparse characteristics;

[0026] S3-4: Determine the consistency of the above data association features of the five regions of interest in the same image, and then define the weather type of the point cloud image. The weather types include sunny days, heavy rain, moderate rain, light rain, and drizzle in rainy weather, and blizzards, heavy snow, moderate snow, and light snow in snowy weather. The point cloud feature range changes over time and the defined rainfall / snowfall level indicators are also analyzed.

[0027] Preferably, step S4 is specifically as follows:

[0028] S4-1: Preprocess a certain point cloud sample acquired by the current lidar and perform the steps of S1;

[0029] S4-2: Delineate five key points and regions of interest from the pre-processed current lidar point cloud image, and compare them with the threshold range of data association features under clear weather conditions. This includes the dynamic change data association features of the continuous time interval of the same region of interest, the total number of point clouds in different regions of interest, the reflection intensity of point clouds in different regions of interest, and the distribution pattern of point cloud sparse features to determine whether they are within the consistent feature range division under clear weather conditions.

[0030] S4-3: If the current lidar point cloud image is within the threshold range of the data association feature under sunny weather conditions, it is sunny weather; otherwise, execute step S4-4;

[0031] S4-4: Delineate five key azimuth points and regions of interest on the pre-processed current lidar point cloud image, and compare the current lidar point cloud image with the threshold range of the data association features for rainy weather and snowy weather obtained in step S2, including the dynamic change data association features of the continuous time interval of the same region of interest, that is, the total number of point clouds in different regions of interest, the reflection intensity of the point clouds in different regions of interest, and the distribution pattern of the sparse features of the point clouds. Based on the consistency of the above data association features of the five regions of interest in the same image, the type of rain or snowfall in the image is judged, and the level of rain or snowfall is determined, that is, heavy rain, moderate rain, light rain, and drizzle in rainy weather, and blizzard, heavy snow, moderate snow, and light snow in snowy weather.

[0032] Step S4-5: After determining the rainfall and snowfall levels, repeat step S3-3 to visualize the point cloud data, train the time-frequency status information of the current weather type at the location, form the current rainfall trend based on the time interval and rainfall intensity, and predict the duration of this rainfall type.

[0033] The results of many training samples are visualized to form a trend, and based on this trend (for example, it is decreasing), the duration of rainfall can be determined.

[0034] The rules and data obtained in step S2 of the present invention are in large quantities, and then step S3 is performed: first, the weather type is classified to determine whether it is rain, snow or sunny, and then refined, such as comparing the rainfall data of the same type to determine whether it is heavy rain or light rain. In addition, when consistency results appear at the five key orientation points in the same picture, it is ensured that the picture shows the same weather type, and then step S4 is entered as an example to input data to obtain the final determination of whether it is heavy rain, moderate rain, light rain or drizzle.

[0035] Where the present invention is not exhaustive, please refer to the prior art.

[0036] The beneficial effects of the present invention are:

[0037] 1. This paper combines LiDAR trajectory data from road scans in sunny, rainy, and snowy weather conditions to propose a method for identifying severe weather types using LiDAR point cloud features. This method fully utilizes the LiDAR signal attenuation during severe weather (the power comparison between the transmitted and received laser beams indicates attenuation changes) to identify the weather type and severity level. This method expands the application scope of LiDAR, provides good weather reference conditions for LiDAR monitoring of vehicle roadside data, enriches the spatial characteristics of target trajectory information, and is highly practical.

[0038] 2. The present invention fully extracts the extracted lidar point cloud data, realizing the process from image input to feature extraction and comparison to output of weather type and weather intensity level, directly outputting weather results end-to-end and predicting the duration of severe weather. Compared with using weather forecasts to judge the road conditions of highways, this method can fine-grainedly determine the weather conditions and duration of specific roads, which is more practical. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 Schematic diagram of the flow of the data processing method for identifying severe weather based on laser radar of the present invention;

[0040] Figure 2 Schematic diagram of selecting 5 regions of interest for the 3D point cloud image;

[0041] Figure 3 This is a schematic diagram of point cloud data for a rainfall event;

[0042] Figure 4 This is a schematic diagram of point cloud data for a snowfall event;

[0043] Figure 5 Schematic diagram of point cloud data before preprocessing for snowfall weather;

[0044] Figure 6 Schematic diagram of point cloud data after preprocessing in snowy weather. DETAILED DESCRIPTION

[0045] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, they will be described in detail below with reference to the accompanying drawings and specific embodiments, but are not limited thereto. Matters not fully described in the present invention shall be based on conventional techniques in the art.

[0046] Example 1

[0047] A data processing method based on lidar to identify severe weather, such as Figures 1-6 As shown, the following steps are included:

[0048] S1: Obtain lidar point cloud signal intensity data of different weather types and preprocess the data. Weather types include sunny, rainy, and snowy. Rainfall is divided into heavy rain, moderate rain, light rain, and drizzle. Snowfall is divided into blizzard, heavy snow, moderate snow, and light snow.

[0049] S2: Construct training and test sets based on the LiDAR point cloud signal intensity data obtained in S1, and construct correlation features between different weather types, i.e., sunny days, rain and snow intensity, and the corresponding state data of the point cloud image;

[0050] S3: Clustering the state features under different weather types to obtain point cloud clusters, and determining the data association features of the point cloud clusters corresponding to different weather types;

[0051] S4: Determine the consistency condition of the data association features of a certain point cloud sample and the trained sunny point cloud cluster, determine the consistency condition of the data association features of a certain point cloud sample and the trained rain or snow point cloud cluster, obtain the level of rain or snowfall and predict the duration of rain or snowfall.

[0052] Example 2

[0053] A data processing method for identifying severe weather based on laser radar is as described in Example 1, except that in step S1, the process of preprocessing the data includes:

[0054] S1-1: interpolate missing data and eliminate unreasonable data outside the set response threshold;

[0055] S1-2: Process the point cloud data using a combined noise reduction method to achieve point cloud data processing and target aircraft feature restoration in severe weather conditions;

[0056] The combined denoising method is implemented using a combination of filters, consisting of spatial filtering, ground filtering, voxel filtering, and statistical outlier filtering. Given that filtering algorithms typically require scanning and processing the entire point cloud spatial data, a spatial filtering algorithm is first used to remove irrelevant data from the space. Based on the target point cloud location, unnecessary walls, trees, and other objects are removed from the point cloud space, retaining the target point cloud data for sunny, rainy, and snowy weather conditions.

[0057] This invention improves image quality, removes high-frequency noise and interference, filters out noise points returned by the ground surface and ground objects, maintains point cloud shape characteristics while reducing data processing time complexity, and uses statistics to filter out points outside a threshold. Combined filters can effectively remove points missed by a single filter.

[0058] Example 3

[0059] A data processing method for identifying severe weather based on laser radar is as described in Example 2, except that step S2 includes:

[0060] S2-1: At the lidar test point at the same location, select the weather conditions of sunny days, heavy rain, moderate rain, light rain, and drizzle, and the snowfall conditions of blizzard, heavy snow, moderate snow, and light snow. Collect lidar 3D point cloud images respectively. Construct multiple sets of sample data sets for training and testing convolutional neural networks according to the continuous time periods of the same weather at the same location and the different time periods of different weather at the same location.

[0061] S2-2: In each 3D point cloud image collected by S2-1, the center of the image is used as the coordinate origin, and five key orientation points are defined at the upper left, lower left, upper right, lower right and center. A cube with a side length of n units is used as the region of interest. The position of each cube is represented by the coordinates of two diagonal points. Each point is composed of a three-dimensional vector, namely n1(x1, y1, z1), n2(x2, y2, z2)..., that is, the position of the first cube is represented by n12(x1-x2, y1-y2, z1 -z2)…, multiple groups of sample information are input into the convolutional neural network model (CNN). The lidar point cloud images of continuous time periods of the same weather at the same location and different time periods of different weather at the same location are extracted to construct data association feature relationships corresponding to sunny days, heavy rain, moderate rain, light rain, and drizzle in rainy weather, and blizzards, heavy snow, moderate snow, and light snow in snowy weather. The data association features are determined according to the weather conditions, including the threshold range of the total number of point clouds, the reflection intensity value of the point clouds, and the sparse characteristics of the point clouds.

[0062] Convolutional neural networks (CNNs) are suitable for computer vision tasks such as image processing and classification, and are particularly useful for solving image classification problems. Convolutional neural networks typically involve convolution and pooling of a dataset, taking in multi-channel inputs of color images, and then outputting the final result after multiple convolution and pooling cycles.

[0063] In this step, the total number of point clouds, the reflection intensity value of the point clouds, the sparse characteristics of the point clouds (sparseness, spatial distribution), etc. are determined by the output of the neural network. The same weather type will have a dynamic fluctuation range, that is, the threshold range.

[0064] Example 4

[0065] A data processing method for identifying severe weather based on laser radar is as described in Example 3, except that step S3 is specifically as follows:

[0066] S3-1: Obtaining a point cloud data set of extracted association information of different weather types, wherein the point cloud data set includes classified weather type point cloud data, determining a target point cloud quantity data and a processing weather type corresponding to the target point cloud data from the classified point cloud data;

[0067] S3-2: The weather types processed include sunny days, heavy rain, moderate rain, light rain, and drizzle in rainy days, and blizzards, heavy snow, moderate snow, and light snow in snowy days. Data are labeled 1, 2, 3, etc. according to the weather type. If the data threshold of the associated features is in the same range, that is, the same weather type, they are labeled with the same number. The same type is grouped together as a cluster, and the target point cloud data is clustered to obtain the clustering results;

[0068] S3-3: After obtaining the clustering results, perform point cloud data visualization, displaying the weather types of the five regions of interest marked on the same point cloud map, and statistically analyzing the dynamic change data association characteristics (i.e., time-frequency status information) of the same region of interest at a continuous time interval, including the total number of point clouds in different regions of interest, the reflection intensity characteristics of point clouds in different regions of interest, and the distribution pattern of point cloud sparse characteristics;

[0069] S3-4: Determine the consistency of the above data association features of the five regions of interest in the same image, and then define the weather type of the point cloud image. The weather types include sunny days, heavy rain, moderate rain, light rain, and drizzle in rainy weather, and blizzards, heavy snow, moderate snow, and light snow in snowy weather. The point cloud feature range changes over time and the defined rainfall / snowfall level indicators are also analyzed.

[0070] Example 5

[0071] A data processing method for identifying severe weather based on laser radar is as described in Example 4, except that step S4 is specifically as follows:

[0072] S4-1: Preprocess a certain point cloud sample acquired by the current lidar and perform the steps of S1;

[0073] S4-2: Delineate five key points and regions of interest from the pre-processed current lidar point cloud image, and compare them with the threshold range of data association features under clear weather conditions. This includes the dynamic change data association features of the continuous time interval of the same region of interest, the total number of point clouds in different regions of interest, the reflection intensity of point clouds in different regions of interest, and the distribution pattern of point cloud sparse features to determine whether they are within the consistent feature range division under clear weather conditions.

[0074] S4-3: If the current lidar point cloud image is within the threshold range of the data association feature under sunny weather conditions, it is sunny weather; otherwise, execute step S4-4;

[0075] S4-4: Delineate five key azimuth points and regions of interest on the pre-processed current lidar point cloud image, and compare the current lidar point cloud image with the threshold range of the data association features for rainy weather and snowy weather obtained in step S2, including the dynamic change data association features of the continuous time interval of the same region of interest, that is, the total number of point clouds in different regions of interest, the reflection intensity of the point clouds in different regions of interest, and the distribution pattern of the sparse features of the point clouds. Based on the consistency of the above data association features of the five regions of interest in the same image, the type of rain or snowfall in the image is judged, and the level of rain or snowfall is determined, that is, heavy rain, moderate rain, light rain, and drizzle in rainy weather, and blizzard, heavy snow, moderate snow, and light snow in snowy weather.

[0076] Step S4-5: After determining the rainfall and snowfall levels, repeat step S3-3 to visualize the point cloud data, train the time-frequency status information of the current weather type at the location, form the current rainfall trend based on the time interval and rainfall intensity, and predict the duration of this rainfall type.

[0077] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A data processing method for identifying severe weather based on laser radar, characterized in that: The following steps are involved: S1: Obtain lidar point cloud signal intensity data of different weather types and preprocess the data. Weather types include sunny, rainy, and snowy. Rainfall is divided into heavy rain, moderate rain, light rain, and drizzle. Snowfall is divided into blizzard, heavy snow, moderate snow, and light snow. S2: Construct training and test sets based on the LiDAR point cloud signal intensity data obtained in S1, and construct correlation features between different weather types, i.e., sunny days, rain and snow intensity, and the corresponding state data of the point cloud image; S3: Clustering the state features under different weather types to obtain point cloud clusters, and determining the data association features of the point cloud clusters corresponding to different weather types; S4: Determine the consistency condition of the data association features of a certain point cloud sample and the trained sunny point cloud cluster, determine the consistency condition of the data association features of a certain point cloud sample and the trained rain or snow point cloud cluster, obtain the rain or snow level and predict the duration of rain or snow; Step S2 includes: S2-1: At the lidar test point at the same location, select the weather conditions of sunny days, heavy rain, moderate rain, light rain, and drizzle, and the snowfall conditions of blizzard, heavy snow, moderate snow, and light snow. Collect lidar 3D point cloud images respectively. Construct multiple sets of sample data sets for training and testing convolutional neural networks according to the continuous time periods of the same weather at the same location and the different time periods of different weather at the same location. S2-2: In each 3D point cloud image collected by S2-1, the center of the image is used as the coordinate origin, and five key orientation points are delineated at the upper left, lower left, upper right, lower right and center respectively. A cube with a side length of n units is used as the region of interest. The position of each cube is represented by the coordinates of two diagonal points of the three-dimensional point. Each point is composed of a three-dimensional vector. Multiple groups of sample information are input into the convolutional neural network model. The lidar point cloud images of the same weather at the same location in continuous time periods and the same weather at the same location in different time periods are extracted and constructed, and the data association feature relationship corresponding to sunny days, heavy rain, moderate rain, light rain and drizzle in rainy weather, blizzard, heavy snow, moderate snow and light snow in snowy weather is constructed. The data association features are determined according to the weather conditions, including the total number of point clouds, the reflection intensity value of the point clouds, and the threshold range of the sparse features of the point clouds; Step S3 is specifically as follows: S3-1: Obtaining a point cloud data set of extracted correlation information of different weather types, determining the target point cloud quantity data and the processing weather type corresponding to the target point cloud data from the classified point cloud data; S3-2: The weather types processed include sunny days, heavy rain, moderate rain, light rain, and drizzle in rainy days, and blizzards, heavy snow, moderate snow, and light snow in snowy days. Data is labeled according to the weather type, and the target point cloud data is clustered to obtain the clustering results; S3-3: After obtaining the clustering results, perform point cloud data visualization, displaying the weather types of the five regions of interest marked on the same point cloud map, and statistically analyzing the dynamic change data association characteristics of the same region of interest at a continuous time interval, that is, the time-frequency status information, including the total number of point clouds in different regions of interest, the reflection intensity characteristics of point clouds in different regions of interest, and the distribution pattern of point cloud sparse characteristics; S3-4: Determine the consistency of the above data association features in the five regions of interest in the same image, and then delineate the weather type of the point cloud image. The weather types include sunny days, heavy rain, moderate rain, light rain, and drizzle in rainy days, and blizzard, heavy snow, moderate snow, and light snow in snowy days. The point cloud feature range changes over time and the delineated rainfall and snowfall level indicators are also analyzed. Step S4 is specifically as follows: S4-1: Preprocess a certain point cloud sample acquired by the current lidar and perform the steps of S1; S4-2: Delineate five key points and regions of interest from the pre-processed current lidar point cloud image, and compare them with the threshold range of data association features under clear weather conditions. This includes the dynamic change data association features of the continuous time interval of the same region of interest, the total number of point clouds in different regions of interest, the reflection intensity of point clouds in different regions of interest, and the distribution pattern of point cloud sparse features to determine whether they are within the consistent feature range division under clear weather conditions. S4-3: If the current lidar point cloud image is within the threshold range of the data association feature under sunny weather conditions, it is sunny weather; otherwise, execute step S4-4; S4-4: Delineate five key azimuth points and regions of interest (ROIs) on the pre-processed current LiDAR point cloud image, and compare the current LiDAR point cloud image with the threshold range of data association features for rainy and snowy weather. This includes dynamically changing data association features for the same region of interest at varying intervals, i.e., the total number of point clouds in different regions of interest, the reflection intensity of point clouds in different regions of interest, and the distribution pattern of sparse features of point clouds. Based on the consistency of these data association features across the five regions of interest in the same image, determine the type of rain or snowfall in the image, and also determine the level of rain or snowfall, i.e., heavy rain, moderate rain, light rain, or drizzle for rainy weather, and blizzard, heavy snow, moderate snow, or light snow for snowy weather. Step S4-5: After determining the rainfall and snowfall levels, repeat step S3-3 to visualize the point cloud data, train the time-frequency status information of the current weather type at the location, form the current rainfall trend based on the time interval and rainfall intensity, and predict the duration of this rainfall type.

2. The data processing method for identifying severe weather based on laser radar according to claim 1, characterized in that: In step (1), in step S1, the process of preprocessing the data includes: S1-1: interpolate missing data and eliminate data outside the set response threshold; S1-2: Process the point cloud data using a combined noise reduction method to achieve point cloud data processing and target aircraft feature restoration in severe weather conditions; The combined denoising method is implemented through a combined filter, which is composed of spatial filtering, ground filtering, voxel filtering and statistical outlier filtering.

Images