Recognition system based on laser wide-area scanning low cloud and mist

Abstract

The invention discloses a low cloud and mist recognition system based on laser wide-area scanning. The system comprises the following steps: S1, performing inversion according to range resolution data of laser radar rapid wide-area scanning to obtain visibility information; s2, training a network model according to historical data of the forward scattering visibility meter and the laser radar; s3, drawing a horizontal visibility distribution map according to the visibility information; s4, performing image processing on the visibility distribution map, and identifying a low cloud and mist development path; drawing a low-visibility region boundary in the original image; s5, obtaining and outputting the visibility condition of each channel or road according to the identified low cloud development boundary; all visibility information in a scanning range is obtained by scanning the change process of fog in a port range in a large range, and the method has high temporal-spatial resolution, can observe the whole sea fog change process more finely, and is beneficial to guiding port navigation more finely.

Description

Technical Field

[0001] The present invention relates to the technical field of laser scanning, and particularly to an identification system for low cloud and fog based on laser wide-area scanning. Background Art

[0002] Fast and efficient transportation plays a significant role in promoting the rapid economic development. In the modern comprehensive transportation system, highways, ports, and airports, as the core hubs, bear the heavy responsibility of a huge amount of personnel flow and material transportation. Safety is an important guarantee for efficient land, sea, and air transportation. However, the comprehensive transportation system is severely affected by weather. For example, high-speed transportation is vulnerable to weather such as fog, rain, snow, and icing. Maritime transportation, especially ports, is vulnerable to fog, strong winds, and high waves. Air transportation is vulnerable to wind shear, fog, lightning, etc. Among them, foggy weather has the most extensive impact on land, sea, and air transportation. When fog appears, the visibility will be significantly reduced, seriously interfering with traffic operation. Due to the complexity and variability of the generation and dissipation of low cloud and fog, and the limited monitoring means and accuracy, it is often impossible to timely predict the generation and dissipation of foggy weather. At present, the way for each transportation system to deal with foggy weather is to close the high-speed channels or port channels, etc. It can only be released after the fog in the entire control area naturally dissipates to the safe visibility range, which not only brings great inconvenience to people's travel, but also has a negative impact on the efficient operation of the social economy, resulting in a series of reactions such as increased logistics costs and blocked trade. In fact, the dissipation of fog is not a group generation or group dissipation. For example, the generation and dissipation of sea fog are often local generation and local dissipation. The area of a port is generally large. If waiting for the fog in the entire port to dissipate before opening the port, it will seriously affect the transportation capacity of the port. Therefore, how to achieve accurate identification and accurate monitoring and prediction of the development trend of low cloud and fog is the key problem to solve the rapid operation of transportation.

[0003] The monitoring means of low cloud and fog mainly include traditional ground monitoring, satellite remote sensing monitoring, mobile monitoring, etc. The traditional ground monitoring data is accurate but the range is limited and it is lagging; satellite remote sensing monitoring has a wide range and high time resolution, but it is vulnerable to atmospheric conditions and the accuracy of a small area is insufficient; mobile monitoring is flexible and efficient, but the sensors need to be calibrated regularly and it is not suitable for long-term fixed-point monitoring. In contrast, lidar for measuring visibility has the advantages of strong real-time performance, high accuracy, and wide coverage, and can effectively make up for the deficiencies of the existing technology, providing more reliable technical support for the dynamic monitoring of low cloud and fog. Summary of the Invention

[0004] Object of the Invention: In order to overcome the inaccurate prediction of sea fog, which affects the transportation capacity of ports, the present invention provides an identification system for low cloud and fog based on laser wide-area scanning, which has the ability of high spatio-temporal resolution information, can accurately identify sea fog and provide accurate position information.

[0005] To achieve the above object, the technical solution adopted by the present invention is to provide a low-cloud and fog recognition system based on laser wide-area scanning, including the following steps:

[0006] S1. Invert the visibility information from the range resolution data obtained by the rapid wide-area scanning of the lidar;

[0007] S2. Train a U-shaped network model according to the historical data of the forward-scattering visibility meter and the lidar;

[0008] S3. Draw a horizontal visibility distribution map according to the visibility information obtained in S1;

[0009] S4. Perform threshold segmentation, edge detection, and contour extraction on the obtained horizontal visibility distribution map through a low-cloud and fog recognition algorithm to identify the development path of low-cloud and fog; obtain real-time data of the forward-scattering visibility meter and the lidar; match the data with the trained U-shaped network model and the low-cloud and fog recognition algorithm to obtain a matched low-cloud and fog recognition system; use the system to identify the development process of low-cloud and fog;

[0010] S5. Obtain and output the visibility conditions of each waterway or road according to the identified development boundary of low-cloud and fog.

[0011] The specific method for inverting the visibility information from the range resolution data obtained by the rapid wide-area scanning of the lidar in S1 is as follows: Invert the radar signal information into visibility information through a horizontal visibility inversion algorithm, and the horizontal visibility inversion algorithm is:

[0012] ;

[0013] ;

[0014] wherein, is the atmospheric extinction coefficient, is the visibility, is a coefficient determined by an empirical formula, is the wavelength, and R is the detection distance.

[0015] As a preferred embodiment of the present invention: S2 is specifically as follows:

[0016] Divide the visibility data of the forward-scattering visibility meter and the visibility data of the visibility lidar into a training set, a validation set, and a test set according to a preset ratio respectively. The visibility data of the forward-scattering visibility meter is used as a label, and the visibility data of the visibility lidar is used as input data, and use the visibility data to train a U-shaped network model, and the trained model is a denoising network.

[0017] As a preferred embodiment of the present invention: the training of the U-shaped network model specifically includes the following steps:

[0018] S2-1, obtaining the original signal data through wide-area rapid scanning of the lidar, and inversely calculating the visibility information with distance resolution according to the horizontal visibility inversion algorithm;

[0019] S2-2, inputting the inversely calculated visibility information and the visibility information of the forward-scattering visibility meter into the U-shaped network model in pairs, where the data of the forward-scattering visibility meter is used as the label, and the model realizes improving the measurement accuracy of low signal-to-noise ratio visibility by learning the mapping relationship between the input information and the corresponding label;

[0020] S2-3, defining the loss function according to the mean absolute error between the lidar visibility denoising result predicted by the model and the visibility of the forward-scattering visibility meter, specifically:

[0021] ;

[0022] where N is the number of samples; is the model prediction value, is the label value;

[0023] S2-4, taking the mean absolute error between the uncorrected data and the label as the baseline value, and setting the threshold of the loss function. If the loss value is higher than the preset threshold of the loss function, then update the U-shaped network model.

[0024] As a preferred embodiment of the present invention: the specific content of S3 is: drawing a horizontal visibility distribution map according to the visibility information with distance resolution and the real port waterway position or road surface position.

[0025] As a preferred embodiment of the present invention: the specific content of S4 is: using the OpenCV image processing library to perform threshold segmentation on the obtained horizontal visibility distribution map, extracting the area where the visibility is lower than the threshold, then using the edge detection algorithm, i.e., the Canny algorithm, to extract the boundary of the low visibility area, then connecting the discrete edges detected by the Canny algorithm into a continuous closed contour, and extracting the contour geometric information to form a complete fog boundary.

[0026] As a preferred embodiment of the present invention: the specific steps of the OpenCV image processing include the following:

[0027] S4-1, mapping the visibility value and the pixel value, the pixel value corresponds to the visibility value, and the visibility range is dynamically adjusted according to the real-time visibility; where the highest visibility at a certain moment the pixel value is 255, and the lowest visibility at a certain moment the pixel value is 0;

[0028] S4-2. Set the visibility threshold, extract the area with visibility lower than the threshold, and generate a binary image, namely the visibility binary map, where the low visibility area is white and other areas are black;

[0029] Determine the visibility threshold according to traffic regulations, and obtain the pixel value threshold corresponding to the visibility threshold through the mapping relationship between the visibility value and the pixel value ,

[0030] ;

[0031] Among them, is the visibility threshold;

[0032] S4-3. Use the Canny algorithm to detect the boundary of the low visibility area and generate a binary edge map; perform Canny edge detection on the binary image to extract the boundary of the low visibility area;

[0033] S4-4. Extract the contours in the edge map and draw the boundary of the low visibility area on the original image;

[0034] Use the contour extraction function in the OpenCV image processing library to extract the outer contour in the edge map and draw a red boundary of the low visibility area on the original image;

[0035] S4-5. Display the horizontal visibility distribution map after drawing the boundary.

[0036] As a preferred embodiment of the present invention: in the S5, the method for obtaining the visibility conditions of each waterway or road through the identified low cloud and fog development boundary specifically includes the following steps:

[0037] S5-1. Represent the position of the waterway or road with coordinates, load and read the visibility binary map; use the coordinate order of the image in OpenCV, and ensure a single-channel grayscale image when loading the binary image;

[0038] S5-2. Traverse all positions of the waterway or road;

[0039] S5-3. Judge whether the pixel value at the coordinate of the waterway or road exceeds the pixel value threshold ; For each waterway position, check its pixel value in the binary map. If it exceeds the pixel value threshold , record this position;

[0040] S5-4. Output the coordinate of this waterway or position.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] Scanning the change process of fog in a large range dynamically within the port can obtain visibility information of any point of sea fog within the scanning range, observe the entire sea fog change process more precisely, and is conducive to providing more refined guidance for navigation command. Moreover, this application has the ability of high spatio-temporal resolution information, can accurately identify sea fog, and correct the accuracy through the U-shaped network model and OpenCV image processing, making the measurement data more accurate. Using a wide-area laser system to quickly scan each channel area of the port, through data processing and manual correction, an identification map of the low-cloud and fog development boundary and high-precision visibility information with distance resolution are obtained. Determine whether a ship can navigate according to the navigation visibility threshold, and flexibly and quickly command the navigation of port ships according to the information of the real-time changing low-cloud and fog development boundary identification map, improving the port capacity. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a flowchart of the method according to an embodiment of the present invention;

[0044] Figure 2 It is a training flowchart of the U-shaped network model provided by an embodiment of the present invention;

[0045] Figure 3 It is a flowchart of OpenCV image processing provided by an embodiment of the present invention;

[0046] Figure 4 It is a flowchart of reasoning for the passing position rule provided by an embodiment of the present invention;

[0047] Figure 5 It is an identification map of the low-cloud and fog development boundary provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] The present invention will be further clarified below with reference to the drawings and specific embodiments. It should be understood that these examples are only used to illustrate the present invention and not to limit the scope of the present invention. After reading the present invention, various equivalent forms of modification of the present invention by those skilled in the art all fall within the scope defined by the appended claims of this application.

[0049] As Figures 1 - 5 shown, it is an identification system for low-cloud and fog based on wide-area laser scanning, specifically including the following steps:

[0050] S1. Invert visibility information according to the distance resolution data obtained by rapid wide-area scanning of the lidar;

[0051] S2. Train the network model according to the historical data of the forward-scattering visibility meter and the lidar to improve the measurement accuracy of low-signal-to-noise ratio visibility;

[0052] S3. Draw a horizontal visibility distribution map according to the visibility information obtained in S1;

[0053] S4. Use the low-cloud and fog recognition algorithm to perform threshold segmentation, edge detection, and contour extraction on the obtained horizontal visibility distribution map to identify the development path of low clouds and fog; obtain real-time forward-scattering visibility meter and lidar data; match this data with the trained U-shaped network model and the low-cloud and fog recognition algorithm to obtain the matched low-cloud and fog recognition system; use the system to identify the development process of low clouds and fog;

[0054] S5. Obtain and output the visibility conditions of each waterway or road according to the identified development boundary of low clouds and fog.

[0055] Among them, the specific method for inversely calculating the visibility information from the range-resolution data of the lidar's fast wide-area scan in S1 is: invert the radar signal information into visibility information through the horizontal visibility inversion algorithm, and the horizontal visibility inversion algorithm is:

[0056] ;

[0057] ;

[0058] Among them, is the atmospheric extinction coefficient, is the visibility, is a coefficient obtained from an empirical formula, is the wavelength, and R is the detection distance.

[0059] In S2, the specific method for training the type network model using the forward-scattering visibility meter and lidar data is:

[0060] Divide the visibility data of the forward-scattering visibility meter and the visibility data of the visibility lidar into a training set, a validation set, and a test set according to a preset ratio respectively. The visibility data of the forward-scattering visibility meter is used as a label, and the visibility data of the visibility lidar is used as input data. Use the visibility data to train the type network model. The trained model is a denoising network, and the denoising network realizes denoising by learning the difference features between the forward-scattering visibility data and the lidar visibility data, thereby improving the visibility measurement accuracy with low signal-to-noise ratio.

[0061] Among them, the training of the type network model specifically includes the following steps:

[0062] S2-1. The lidar performs a fast wide-area scan to obtain the original signal data, and inversely calculates the range-resolved visibility information according to the horizontal visibility inversion algorithm;

[0063] S2-2, pair the retrieved visibility information and the visibility information of the forward-scattering visibility meter and input them into the U-shaped network model. The data of the forward-scattering visibility meter is used as the label. By learning the mapping relationship between the input information and the corresponding label, the model improves the measurement accuracy of visibility with low signal-to-noise ratio.

[0064] S2-3, define the loss function according to the mean absolute error between the denoising result of the lidar visibility predicted by the model and the visibility of the forward-scattering visibility meter. Specifically:

[0065] ;

[0066] where N is the number of samples; is the predicted value of the model, is the label value;

[0067] S2-4, take the mean absolute error between the uncorrected data and the label as the baseline value, set the loss function threshold to 70% of the baseline value. If the loss value is higher than the preset threshold of the loss function, update the U-shaped network model.

[0068] The specific process of drawing the horizontal visibility distribution map in S3 is as follows: draw the horizontal visibility distribution map according to the distance-resolved visibility information and the real port waterway position or road surface position.

[0069] In S4, through the network model performs threshold segmentation, edge detection, and contour extraction on the obtained horizontal visibility distribution map to identify the low-cloud and fog development path. This step is specifically as follows: use the OpenCV image processing library to perform threshold segmentation on the obtained horizontal visibility distribution map, extract the area with visibility lower than the threshold, then use the edge detection algorithm, i.e., the Canny algorithm, to extract the boundary of the low-visibility area, and then connect the discrete edges detected by the Canny algorithm into a continuous closed contour and extract the contour geometric information to form a complete fog boundary. Among them, the OpenCV image processing specifically includes the following steps:

[0070] S4-1, map the visibility value and the pixel value, and the pixel value corresponds to the visibility value. The visibility range is dynamically adjusted according to the real-time visibility; where the highest visibility at a certain moment the pixel value is 255, and the lowest visibility at a certain moment the pixel value is 0;

[0071] S4-2, set the visibility threshold, extract the area with visibility lower than the threshold, and generate a binary image, i.e., the visibility binary image. The low-visibility area is white, and other areas are black;

[0072] Determine the visibility threshold according to traffic regulations, and obtain the pixel value threshold corresponding to the visibility threshold through the mapping relationship between the visibility value and the pixel value ,

[0073] ;

[0074] Among them, is the visibility threshold;

[0075] S4-3. Use the Canny algorithm to detect the boundary of the low visibility area and generate a binary edge map; perform Canny edge detection on the binary image to extract the boundary of the low visibility area;

[0076] S4-4. Extract the contours in the edge map and draw the boundary of the low visibility area on the original image;

[0077] Use the contour extraction function in the OpenCV image processing library to extract the outer contour in the edge map and draw a red boundary of the low visibility area on the original image;

[0078] S4-5. Display the horizontal visibility distribution map after drawing the boundary;

[0079] In S5, the method for obtaining the visibility conditions of each waterway or road through the identified low cloud and fog development boundary specifically includes the following steps:

[0080] S5-1. Represent the waterway or road position with coordinates, load and read the visibility binary map;

[0081] Use the coordinate order of the image in OpenCV to ensure a single-channel grayscale image when loading the binary image.

[0082] S5-2. Traverse all waterway or road positions;

[0083] S5-3. Judge whether the pixel value at the waterway or road coordinate exceeds the pixel value threshold; for each waterway position, check its pixel value in the binary map. If it exceeds the pixel value threshold, record this position;

[0084] S5-4. Output the coordinate of this waterway or position.

[0085] The large-scale dynamic scanning of the change process of fog within the port range has the function of measuring the thick fog in the entire sea area compared with the shore-based point-type in-situ measurement. It can obtain the visibility information of any point of the sea fog within the scanning range, and observe the entire sea fog change process more refinedly, which is conducive to providing more refined guidance for navigation command. Satellite observation is easily affected by clouds and convective weather, and the temporal and spatial resolution and accuracy of fog recognition are not high. In comparison, this invention patent has high spatio-temporal resolution information and is more accurate in identifying sea fog. The traditional lidar visibility algorithm needs to be calibrated and generally has low accuracy. This invention patent uses the U-shaped network model and OpenCV image processing for accuracy correction, making the measurement data more accurate.

[0086] This application uses a wide-area laser system to quickly scan each channel area of the port. After the above data processing and manual correction, a low-cloud and fog development boundary recognition map and high-precision visibility information with distance resolution are obtained. It is used to judge whether a ship can navigate according to the navigable visibility threshold, and flexibly and quickly direct the navigation of port ships according to the information of the real-time changing low-cloud and fog development boundary recognition map, so as to improve the port capacity.

[0087] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. An identification system for low cloud and fog based on laser wide-area scanning, characterized in that, Including the following steps: S1. Invert visibility information from the range resolution data of lidar's fast wide-area scanning; S2. Train a type of network model according to the historical data of the forward-scattering visibility meter and lidar; S3. Draw a horizontal visibility distribution map based on the visibility information obtained in S1; S4. Perform threshold segmentation, edge detection, and contour extraction on the obtained horizontal visibility distribution map through a low-cloud and fog recognition algorithm to identify the development path of low-cloud and fog; obtain real-time forward-scattering visibility meter and lidar data; match this data with the trained U-shaped network model and the low-cloud and fog recognition algorithm to obtain a matched low-cloud and fog recognition system; use the system to identify the development process of low-cloud and fog; S5. Obtain and output the visibility conditions of each waterway or road according to the identified development boundary of low-cloud and fog.

2. The recognition system for low cloud and fog based on laser wide-area scanning according to claim 1, wherein, The specific method of inverting visibility information from the range resolution data of lidar's fast wide-area scanning in S1 is as follows: Invert radar signal information into visibility information through a horizontal visibility inversion algorithm, and the horizontal visibility inversion algorithm is: ； ； Among them, is the atmospheric extinction coefficient, is the visibility, is a coefficient determined by an empirical formula, is the wavelength, and R is the detection distance.

3. The recognition system for low cloud and fog based on laser wide-area scanning according to claim 1, characterized in that, The specific content of S2 is: Divide the visibility data of the forward-scattering visibility meter and the visibility data of the visibility lidar into a training set, a validation set, and a test set according to a preset ratio respectively. The visibility data of the forward-scattering visibility meter is used as the label, and the visibility data of the visibility lidar is used as the input data, and use the visibility data to train the network model of type, and the trained model is a denoising network.

4. The recognition system for low cloud and fog based on laser wide-area scanning according to claim 3, characterized in that, The described training of the type network model specifically includes the following steps: S2-1. Lidar performs wide-area fast scanning to obtain original signal data, and inverts range-resolved visibility information according to the horizontal visibility inversion algorithm; S2-2. Input the inverted visibility information and the visibility information of the forward-scattering visibility meter into the U-shaped network model in pairs, where the forward-scattering visibility meter data is used as a label, and the model realizes the improvement of the visibility measurement accuracy with low signal-to-noise ratio by learning the mapping relationship between the input information and the corresponding label; S2-3. Define a loss function according to the mean absolute error between the lidar visibility denoising result predicted by the model and the visibility of the forward-scattering visibility meter, specifically: ； Among them, N is the number of samples; is the model prediction value, is the label value; S2-4. Use the mean absolute error between the uncorrected data and the label as the baseline value, and set the threshold of the loss function. If the loss value is higher than the preset threshold of the loss function, update the U-shaped network model.

5. The recognition system for low cloud and fog based on laser wide-area scanning according to claim 1, characterized in that, The specific content of S3 is: Draw a horizontal visibility distribution map according to the range-resolved visibility information and the real port waterway position or road surface position.

6. The recognition system based on laser wide-area scanning for low cloud and fog according to claim 1, characterized in that, The specific content of S4 is: Use the OpenCV image processing library to perform threshold segmentation on the obtained horizontal visibility distribution map, extract the area with visibility lower than the threshold, then use the edge detection algorithm, i.e., the Canny algorithm, to extract the boundary of the low-visibility area, and then connect the discrete edges detected by the Canny algorithm into a continuous closed contour, and extract the contour geometric information to form a complete fog boundary.

7. The recognition system based on laser wide-area scanning for low cloud and fog according to claim 6, characterized in that, The specific steps of the OpenCV image processing include the following: S4-1. Map the visibility value to the pixel value, where the pixel value corresponds to the visibility value, and the visibility range is dynamically adjusted according to the real-time visibility; among them, the highest visibility at a certain moment The pixel value is 255, and the lowest visibility at a certain moment The pixel value is 0; S4-2. Set the visibility threshold, extract the area with visibility lower than the threshold, and generate a binary image, i.e., a visibility binary map, where the low-visibility area is white and other areas are black; Determine the visibility threshold according to traffic regulations, and obtain the pixel value threshold corresponding to the visibility threshold through the mapping relationship between the visibility value and the pixel value , ； Among them, is the visibility threshold; S4-3. Use the Canny algorithm to detect the boundary of the low-visibility area and generate a binary edge map; perform Canny edge detection on the binary image to extract the boundary of the low-visibility area; S4-4. Extract the contours in the edge map and draw the boundary of the low-visibility area on the original image; Use the contour extraction function in the OpenCV image processing library to extract the outer contour in the edge map and draw a red boundary of the low-visibility area on the original image; S4-5. Display the horizontal visibility distribution map after drawing the boundary.

8. The recognition system for low cloud and fog based on laser wide-area scanning according to claim 7, wherein, In S5, the method for obtaining the visibility conditions of each waterway or road through the identified low cloud and fog development boundary specifically includes the following steps: S5-1. Represent the waterway or road position in coordinates, load and read the visibility binary image; use the coordinate order of the image in OpenCV, and ensure a single-channel grayscale image when loading the binary image. S5-2. Traverse all waterway or road positions. S5-3. Determine whether the pixel value at the waterway or road coordinate exceeds the pixel value threshold ; For each waterway position, check its pixel value in the binary image. If it exceeds the pixel value threshold , record this position; S5-4. Output the coordinates of the waterway or position.

Images