A sea fog concentration detection and early warning method based on fog perception density estimation

Abstract

The present invention discloses a sea fog concentration detection and early warning method based on fog-perceived density estimation. The method uses a sea fog image acquisition device to acquire a target image against a single background, such as the sea surface. After extracting fog-related statistical features from the image, the mean and covariance matrix of the eigenvectors are estimated using a multivariate normal distribution (MVG) to establish a fog-perceived density estimation model. The Mahalanobis distance between the image and both the foggy and fog-free models is then calculated to obtain the relative value of the fog concentration. After data smoothing, the fog concentration and visibility are fitted to obtain a fog concentration threshold, and an early warning is issued based on the threshold. The present invention addresses the drawback of being unable to obtain sea fog concentration from images against a single background, such as the sea surface. By quantifying the sea fog concentration and matching it with visibility, a timely early warning is issued based on the quantified concentration threshold.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer vision and digital image information processing, and in particular relates to a sea surface fog concentration detection and early warning method based on fog perception density estimation. Background Art

[0002] Sea fog is a major marine weather disaster. Heavy fog forces ships of all types to anchor or slow down at sea, resulting in losses of manpower, material resources, and time. Furthermore, accidents involving entanglement of kelp, laver, and other aquaculture areas are common, posing a significant threat to normal vessel navigation. In addition to maritime hazards, sea fog can also trigger serious land or air traffic accidents, disrupt power transmission, and even produce fog flashovers, causing power transmission lines to discharge into the fog, directly threatening power grid security. Therefore, understanding the basic laws and characteristics of coastal sea fog and predicting its concentration not only ensures the safety of maritime activities and transportation, but also facilitates the coordination of industrial, agricultural, fishery, and transportation production, thus possessing significant academic and practical value.

[0003] Currently, the Meteorological Bureau uses specialized monitoring instruments to monitor visibility. While this specialized equipment offers the advantage of precise monitoring, it also has the disadvantage of being limited by the instrumentation and unable to quickly assess current visibility conditions. Therefore, a simple and fast image processing method is necessary to promptly understand current visibility conditions and issue traffic warnings during foggy weather.

[0004] In recent years, visibility or fog density detection algorithms for foggy images have focused on images of urban roads or densely populated buildings. These algorithms can only provide early warnings for urban road traffic, but cannot accurately estimate visibility in environments with a monotonous background and a large amount of blank space, such as the ocean surface. Furthermore, these algorithms often extract a large number of features (greater than 5) and can only quantify the fog density level. For example, if the current fog density is light, moderate, or heavy, they cannot quantify the fog density value, making timely early warnings impossible. Summary of the Invention

[0005] The purpose of the present invention is to provide a sea surface fog concentration detection and early warning method based on fog perception density estimation to overcome the defect that sea fog concentration cannot be obtained through images under a single background such as the sea surface. By quantifying the sea fog concentration and matching it with visibility, an early warning can be given in time according to the quantified concentration threshold.

[0006] The present invention provides a sea fog concentration detection and early warning method based on fog perception density. The method achieves sea fog concentration detection and early warning by fixing a target image on a floating body on the sea surface and collecting and analyzing the target image data. The specific steps are as follows:

[0007] S1: Image acquisition: Real-time acquisition of target images with distinct features fixed on a floating object on the sea surface in different environments as test images;

[0008] S2: Extract fog-related statistical features of the foggy image dataset, the fog-free image dataset, and the collected test images; fog-related features include: saturation, Weber brightness contrast, and chromaticity variance;

[0009] S3: Based on the foggy image dataset and the fog-free image dataset, the mean and covariance matrix of the eigenvector are estimated using the multivariate normal distribution (MVG), and fog-aware density estimation models based on the fog-free MVG model and the foggy MVG model are established respectively;

[0010] S4: Extract the three-dimensional feature vector X from the test image and predict the fog-free level d1 and fog-affected level d2 of the test image by using the Mahalanobis distance metric with the fog-aware density estimation model based on the fog-free MVG model and the fog-affected MVG model respectively;

[0011] The Mahalanobis distance metric formula is:

[0012]

[0013] in is the eigenvector, σ n is the chromaticity variance, w n is the Weber luminance contrast, is the average saturation, μ i and ∑ i are the mean and covariance matrix of the feature sets of all blocks in the foggy and fog-free datasets, respectively, and ∑′ is the covariance matrix of the feature sets of all blocks in a given test image;

[0014] Using the averaging strategy, d1 and d2 are averaged and substituted into the following formula to calculate the relative fog concentration D:

[0015]

[0016] S5: After smoothing the obtained relative fog concentration D data, a curve of the relative fog concentration D versus real-time visibility is plotted, and the fog concentration threshold is determined based on the fitting of the relative fog concentration D value and the visibility;

[0017] S6: Early warning: When the real-time relative fog concentration value D is greater than the fog concentration threshold, it is judged that there is fog and an early warning is given.

[0018] In the present invention, in step S1, a sea fog image acquisition device is used to acquire images, which includes a floating body, an anchor system, a counterweight, a mast, a digital camera, and a target picture with obvious features; the floating body is connected to the anchor system and the counterweight, the target picture is fixed on the floating body, and a mast is set on the floating body, and a digital camera is mounted on the mast. The digital camera is aimed at the target picture to realize real-time acquisition of the target picture.

[0019] In the present invention, in step S1, target images are collected every 1 minute to obtain target images in different fog environments in real time.

[0020] In the present invention, in step S3, the multivariate normal distribution MVG estimation uses the standard maximum likelihood estimation method to describe the statistical characteristics of the data set.

[0021] In the present invention, in step S4, the relative fog concentration value D and the visibility value are smoothed in groups of 10 minutes, and samples with a statistical z value greater than 3 are removed as singular values; in the change curve diagram, the horizontal axis is the mean value of each visibility group, and the vertical axis is the mean value of each image D value group.

[0022] In the present invention, in step S5, the relative fog concentration value corresponding to the visibility of 1000 meters is determined as a threshold value and stored in a database to facilitate timely giving of early warning prompts.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] 1. The present invention is based on a sea surface image acquisition device. Under the single background of the sea surface, it takes pictures of designated targets and performs image processing to achieve sea fog and dense fog detection and early warning.

[0025] 2. This fog-sensing density-based fog concentration detection method reduces the number of eigenvectors and the algorithmic complexity. The mean and covariance matrix of the eigenvectors are estimated using the multivariate normal distribution (MVG). The relative fog concentration is calculated using the Mahalanobis distance and then fitted with the visibility to determine the fog threshold. This method provides timely warnings and facilitates maritime activities. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a side view of the image acquisition device.

[0027] Figure 2 This is a flowchart of the steps of image preprocessing.

[0028] Figure 3 This is a schematic diagram of the working steps of the sea fog concentration detection and early warning system based on fog perception density estimation.

[0029] Figure 4 This is a curve showing the change of fog concentration with visibility. DETAILED DESCRIPTION

[0030] The present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments to provide a comprehensive understanding of the technical solutions and advantages of the present invention. These descriptions and embodiments are only intended to explain the present invention and are not intended to limit the present invention.

[0031] The present invention proposes a sea fog image acquisition device, such as Figure 1 The image acquisition device consists of a digital camera 1, a mast 2, a buoy 4, an anchor 5, and a counterweight 6. A target image 3 with distinct features, rich colors, high contrast, and high clarity is affixed to the buoy 4. The camera 1 and target image 3 are positioned so that the camera 1 captures the entire target image 3 without leaving any blank spaces. This device enables real-time acquisition of target images in various environments, resolving the issue of inability to accurately estimate fog concentration from images due to a single sea surface background.

[0032] Furthermore, in order to realize sea fog concentration detection and early warning, the present invention processes the target image collected by the sea fog image acquisition device, and pre-processes the foggy image dataset and the fog-free image dataset, as follows ( Figure 2 ):

[0033] (1) Prepare a foggy image dataset, a fog-free image dataset, and the image to be tested, and extract image fog-related statistical features, including saturation, Weber brightness contrast, and chromaticity variance;

[0034] The formula for calculating saturation is:

[0035] s=I HSV (i,j,2)

[0036] Among them I HSV It is the image after the original image is converted to HSV space.

[0037] The formula for calculating Weber brightness contrast is:

[0038]

[0039] Where I and Ib represent the brightness of the foreground and background of the image respectively.

[0040] The formula for calculating chromaticity variance is:

[0041]

[0042] where μ rg and μ yb , σ rg and σ yb are the mean and variance of rg and yb respectively, where rg=rg,

[0043] yb=0.5(r+g)-b.

[0044] (2) Multivariate Normal Distribution (MVG) modeling estimates the mean and covariance matrix of the feature vector. After selecting three fog-related features, a fog-aware density estimation model is established by learning two MVG models (foggy MVG and fog-free MVG). Standard maximum likelihood estimation is used to describe the statistical characteristics of the dataset.

[0045] The formula of the MVG model is:

[0046]

[0047] Where X is the eigenvector, μ and Σ are the mean vector and covariance matrix of X. Here, the fog-free model and foggy model are described by a pair (μ1, Σ1) and (μ2, Σ2), respectively.

[0048] (3) Calculate the relative value of fog concentration by Mahalanobis distance. After learning two MVG models, a three-dimensional feature vector X is extracted for each image, and the Mahalanobis distance between the two fog-free and foggy models is used to predict the fog-free level d1 and foggy level d2 of an image.

[0049] The Mahalanobis distance metric formula is:

[0050]

[0051] where ∑′ is the covariance matrix of the feature set of all patches in a given test image.

[0052] Using the averaging strategy, d1 and d2 are averaged and substituted into the following formula to calculate the relative fog concentration D:

[0053]

[0054] In a specific embodiment, the sea fog concentration detection and early warning method based on fog perception density of the present invention is implemented as follows: Figure 3 ):

[0055] Step S1, image acquisition: Use a camera to capture a target image 3 every minute to obtain target images in different fog environments in real time.

[0056] Step S2, extracting image fog-related statistical features, including features including saturation, Weber brightness contrast and chromaticity variance;

[0057] Step S3, establishing a fog perception density estimation model by estimating the mean and covariance matrix of the eigenvector using the multivariate normal distribution (MVG);

[0058] Step S4: Calculate the relative value of fog concentration using Mahalanobis distance. After learning two MVG models, extract a three-dimensional feature vector X from each image and use the Mahalanobis distance metric between the two fog-free and foggy models to predict the fog-free level d1 and fog-rich level d2 of an image.

[0059] The Mahalanobis distance metric formula is:

[0060]

[0061] where ∑′ is the covariance matrix of the feature set of all patches in a given test image.

[0062] The foggy model and the fog-free model include a foggy dataset and a fog-free dataset;

[0063] Using the averaging strategy, d1 and d2 are averaged and substituted into the following formula to calculate the relative fog concentration D:

[0064]

[0065] Step S5: Data smoothing. Smooth the fog concentration value D and visibility value in groups of 10 minutes to remove singular values:

[0066] The definition of z-value in statistics is used to help remove outliers. The z-value represents how many standard deviations the sample value of a random variable that conforms to a normal distribution is from the mean. The calculation formula is:

[0067]

[0068] Where X is a random variable, μ is the mean, σ is the standard deviation, and the sample is a set of 10 pictures.

[0069] For each sample, the corresponding z value is calculated. Samples with z values greater than 3 are removed as outliers and replaced with the mean of the two preceding and following samples. After removing the outliers and replacing them with the new values, the mean D value for each group of 10 images is calculated. Similarly, visibility is grouped into 10 values (10 minutes), and the mean of each group is calculated after removing the singular value.

[0070] like Figure 4 A graph showing the change in fog concentration versus visibility in a specific embodiment is shown. The horizontal axis represents the mean of each visibility group, and the vertical axis represents the mean of each image D value group. The horizontal axis is plotted in ascending order. The graph shows a clear downward trend, demonstrating a pattern of high visibility and low fog concentration, or low visibility and high fog concentration. The fog concentration threshold is determined based on the fit between the fog concentration value and the visibility.

[0071] Step S6: Early warning. The relative fog concentration value of 1.9 corresponding to the visibility of 1000 meters is stored in the database to facilitate timely early warning.

[0072] In the embodiment, the warning prompt is:

[0073] When the fog concentration value is greater than 1.9, it is judged as foggy and an early warning is given.

Claims

1. A sea fog concentration detection and early warning method based on fog perception density, characterized in that: It collects and analyzes target image data by taking fixed target images of floating objects on the sea surface to achieve sea fog concentration detection and early warning. The specific steps are as follows: S1: Image acquisition: Real-time acquisition of images of fixed floating objects on the sea surface with obvious features in different environments as test images; S2: Extract fog-related statistical features of the foggy image dataset, the fog-free image dataset, and the collected test images; fog-related features include: saturation, Weber brightness contrast, and chromaticity variance; S3: Based on the foggy image dataset and the fog-free image dataset, the mean and covariance matrix of the eigenvector are estimated using the multivariate normal distribution (MVG), and fog-aware density estimation models based on the fog-free MVG model and the foggy MVG model are established respectively; S4: Extract the three-dimensional feature vector X from the test image and predict the fog-free level d1 and fog-affected level d2 of the test image by using the Mahalanobis distance metric with the fog-aware density estimation model based on the fog-free MVG model and the fog-affected MVG model respectively; The Mahalanobis distance metric formula is: ， Where X = [ , , ] is the eigenvector, is the chromaticity variance, is the Weber luminance contrast, is the average saturation, and are the mean and covariance matrices of the feature sets of all blocks in the foggy and fog-free datasets, respectively, is the covariance matrix of the feature set of all blocks in a given test image; Using the averaging strategy, d1 and d2 are averaged and substituted into the following formula to calculate the relative fog concentration D: ， S5: After smoothing the obtained relative fog concentration D data, a curve of the relative fog concentration D versus real-time visibility is plotted, and the fog concentration threshold is determined based on the fitting of the relative fog concentration D value and the visibility; S6: When the real-time relative fog concentration value D is greater than the fog concentration threshold, it is determined that there is fog and an early warning prompt is given.

2. The sea fog concentration detection and early warning method based on fog perception density according to claim 1 is characterized in that: In step S1, an image is collected using a sea fog image acquisition device, which includes a floating body, an anchor, a counterweight, a mast, a digital camera, and a target picture with obvious features; the floating body is connected by the anchor and the counterweight, the target picture is fixed on the floating body on the sea surface, and a mast is set on the floating body, and a digital camera is mounted on the mast. The digital camera is aimed at the target picture to realize real-time acquisition of the target picture.

3. The sea fog concentration detection and early warning method based on fog perception density according to claim 1 is characterized in that: In step S1, target images are collected every 1 minute to obtain target images in different fog environments in real time.

4. The sea fog concentration detection and early warning method based on fog perception density according to claim 1 is characterized in that: In step S3, the multivariate normal distribution MVG estimation uses the standard maximum likelihood estimation method to describe the statistical characteristics of the data set.

5. The sea fog concentration detection and early warning method based on fog perception density according to claim 1 is characterized in that: In step S4, the relative fog concentration value D and visibility value are smoothed in groups of 10 minutes, and samples with a statistical z value greater than 3 are removed as singular values; in the change curve graph, the horizontal axis is the mean of each visibility group, and the vertical axis is the mean of each image D value group.

6. The sea fog concentration detection and early warning method based on fog perception density according to claim 1 is characterized in that: In step S5, the relative fog concentration value corresponding to the visibility of 1000 meters is determined as a threshold and stored in the database to facilitate timely issuance of early warning prompts.

Images