CN101876993B - Method for extracting and retrieving textural features from ground digital nephograms - Google Patents

Abstract

The invention discloses a texture feature extraction and retrieval method of a ground-based digital cloud image, including an extraction method and a retrieval method. The extraction method includes: converting a ground-based digital cloud image with three channels of color RGB into a single-channel pixel category map; analyzing the pixel category Figure and establish the co-occurrence matrix to obtain the histogram vector of the co-occurrence matrix; combine the histogram vectors of the co-occurrence matrix of multiple pixel categories to construct the texture feature vector of the ground-based digital cloud image; save the texture feature vector to the cloud image database. The retrieval includes: extracting the texture features of the sample cloud image according to the above-mentioned feature extraction method; sequentially calculating the similarity between the texture features of the sample cloud image and the texture features of each cloud image in the cloud image database; The results show that. The invention can automatically analyze and extract the effective texture features of the ground-based digital cloud image, and automatically retrieve the result cloud image similar to the sample cloud image from the cloud image database.

Description

A Texture Feature Extraction and Retrieval Method of Ground-Based Digital Cloud Image

technical field

The invention relates to the fields of meteorological detection technology, digital image processing, image retrieval and pattern recognition, in particular to a texture feature analysis and retrieval method of ground-based digital cloud images.

Background technique

In the Earth's atmospheric energy balance, clouds play a very significant role in regulating and are an important factor affecting climate change. On the other hand, the generation, development and evolution of clouds not only reflect the stability, movement and water vapor conditions of the atmosphere at that time, but also one of the important signs foretelling future weather changes. Therefore, cloud observation plays an extremely important role. At present, cloud observations mainly include space-based observations (that is, satellite remote sensing) and ground-based observations. Significant progress has been made in the feature analysis and automatic processing of satellite cloud images. However, ground-based cloud observation and cloud image analysis have long relied on the visual judgment of meteorological observers, which has become a bottleneck in the development of meteorological business automation.

At present, some ground-based all-sky cloud observation devices have been developed at home and abroad, such as the TSI (Total Sky Imager) developed by Yankee Environmental System Inc. of the United States, the WSI (Whole Sky Imager) developed by the University of California in the United States, and the Atmospheric Physics of the Chinese Academy of Sciences. The ASI (All Sky Imager) developed by the Institute and the ground-based all-sky cloud observation system TCI (Ground-based Total-skyCloud Imager) developed by the Chinese Academy of Meteorological Sciences, etc. These devices can automatically take images of the entire sky and generate color digital images—ground-based digital cloud images. Although the ground-based digital cloud image can be obtained automatically, and the cloud amount can basically be automatically calculated, however, the cloud shape analysis of the ground-based digital cloud image still mainly relies on manual analysis by experienced observers. Obviously, there are many defects in manual analysis: firstly, observers must be very familiar with complex observation rules and be able to use them proficiently; Analysis of ground-based cloud images at different times may produce different feature descriptions; in addition, the flow of forecasters and insufficient continuity of observations will also affect the accuracy of observations, and the same ground-based cloud image is often analyzed by different observers. produce different feature descriptions. Obviously, objectively and automatically analyzing and extracting effective features of ground-based digital cloud images is of great significance for the automation and intelligence of ground-based cloud observations. Texture features are an important feature of ground-based digital cloud images, which can objectively describe some characteristics of clouds and their sky background. Using digital image processing and artificial intelligence technology, ground-based digital cloud images can be represented by digital and effective texture feature vectors.

In addition, with the development of digital camera technology and its application in ground-based cloud observation, we will obtain more and more ground-based digital cloud images, which will form a very large-scale (often tens of thousands) image library . In practical applications, meteorological observers or meteorological researchers often need to retrieve specific cloud images in ground-based cloud image databases. In the traditional method, the searcher has two ways to obtain the cloud image: the most primitive method is to manually browse the cloud image database to obtain a specific cloud image; the other method is to search the cloud image according to keywords. In this method, the cloud image database administrator must Manually describe each cloud image in the cloud image database in words (called keywords) in advance, and associate the keyword with the cloud image and save it in the database; when searching, the searcher enters keywords, and the cloud image database searches The system searches by matching keywords. Obviously, when the scale of the cloud image database reaches a certain level, the above two methods are incompetent. Manually browsing a large-scale cloud image database is time-consuming, labor-intensive, and inefficient; although keyword-based retrieval is relatively efficient, it cannot be ignored that the premise of this method is that each cloud image in the cloud image database has the correct key Words are associated with it, and as far as the current technology is concerned, the machine cannot automatically add the correct keywords to the cloud map, but a skilled observer with professional knowledge needs to manually add the keyword description of the cloud map. Similarly, manually adding cloud image keywords is time-consuming and labor-intensive. Moreover, the added keywords are highly subjective and inconsistencies are not uncommon. Therefore, digital and effective texture feature extraction and cloud image retrieval method based on the texture feature can solve the problems in cloud image manual analysis and retrieval to a certain extent.

Contents of the invention

(1) Purpose of the invention

The purpose of the present invention is to provide a method for extracting and retrieving texture features of ground-based digital cloud images, objectively and automatically analyzing and extracting effective texture features of ground-based digital cloud images, and solving the problems existing in the manual analysis and retrieval of above-mentioned cloud images.

(2) Contents of the invention

A method for extracting texture features of ground-based digital cloud images, comprising the following steps:

S101: Convert the color RGB three-channel ground-based digital cloud image into a single-channel pixel category image using the following formula:

$\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; if</span>}{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; R</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}\\ \mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; if</span>}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ></span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; R</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\\ \mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; if</span>}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ></span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; R</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; and</span>}{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}\\ \mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; if</span>}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ></span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; R</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; and</span>}{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}\end{array}\right.<span\; class="notranslate">\; ,</span>$

Among them, I _B (x, y), I _R (x, y) represent the blue (Blue) component value and the red (Red) component value of the pixel at coordinates (x, y) in the input color ground-based digital cloud image respectively, and The value is directly read through the image file of the color ground-based digital cloud image, I _V (x, y) represents the brightness value of the pixel at the coordinate (x, y) in the input color ground-based digital cloud image, and C(x, y) represents the brightness value of the pixel in the color ground-based digital cloud image The category label of the pixel at coordinates (x, y), α ₁ , α ₂ are the threshold parameters of the blue-red band ratio, and β is the brightness threshold parameter.

S102: Obtain the histogram vector of the co-occurrence matrix by analyzing the pixel category map and establishing the co-occurrence matrix;

S103: Merge the histogram vectors of multiple pixel category co-occurrence matrices to construct a texture feature vector of the ground-based digital cloud image;

S104: Save the texture feature vector constructed in S103 to a cloud image database.

Wherein, the _IV (x, y) value in the step S101 is obtained by inputting the blue (Blue) component value red (Red) component value and the green (Green) component value of the pixel at coordinates (x, y) in the colored ground-based digital nephogram. The component values are calculated, and the calculation formula is: I _V (x, y) = 100*Max(I _R (x, y), I _G (x, y), I _B (x, y))/255, where I _G (x, y) is the green (Green) component value of the pixel at the coordinate (x, y).

Wherein, the _α1 value is 1.5, the _α2 value is 1.3, and the β value is 80.

Wherein, the step S102 includes the steps of:

S1021: Analyze the co-occurrence relationship between any two pixel categories in the pixel category map, and construct a co-occurrence matrix CCM:

$\mathrm{<span\; class="notranslate">\; CCM</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; w</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; <figure-callout\; id="1"\; label="h"\; filenames="H2009102385224E00031.png"\; state="\{\{state\}\}">h</figure-callout></span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; ifC</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; iandC</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&Delta;x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&Delta;y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; <figure-callout\; id="0"\; label="j"\; filenames="H2009102385224E00031.png"\; state="\{\{state\}\}">j</figure-callout></span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; otherwise</span>}\end{array}\right.<span\; class="notranslate">\; ,</span>$

Among them, i and j respectively represent the pixel category, and the values are {0, 1, 2, 3}, (Δx, Δy) represent the offset, w, h represent the length and width of the ground-based digital cloud image respectively, CCM(i, j ) indicates the frequency of occurrence of the position pair where the pixel category at (x, y) is i, and the pixel category at (x+Δx, y+Δy) is j in the pixel category map, and each CCM(i, j) is calculated separately, Get a 4×4 co-occurrence matrix;

S1022: Normalize the co-occurrence matrix CCM(i, j) according to the following normalization formula to obtain a normalized co-occurrence matrix CCM ^N ;

CCM ^N (i, j) = CCM (i, j)/(wh).

S1023: According to the following formula, each element of the normalized co-occurrence matrix CCM ^N is spliced row by row to obtain a 16-dimensional histogram vector F _S ;

F _S = (CCM ^N (0, 0), CCM ^N (0, 1), . . . , CCM ^N (3, 3)).

Wherein, the step S103 includes the steps of:

S1031: There are L position offsets of the co-occurrence relationship between two pixel categories, namely (Δx ₁ , Δy ₁ ), (Δx ₂ , Δy ₂ ), ..., (Δx _L-1 Δy _L-1 ), (Δx _L , Δy _L ), get L different co-occurrence matrices and L corresponding histogram vectors according to step S102

S1032: Process the L histogram vectors linearly superimposed and averaged according to the following formula

$\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; L</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; S</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; .</span>$

The texture feature vector of the 16-dimensional ground-based digital cloud image is obtained.

A texture feature retrieval method for ground-based digital cloud images, comprising the following steps:

S201: Calculate the texture feature vector of the sample digital cloud image according to the steps of steps S101, S102 and S103 of claim 1;

S202: Calculate the similarity between the texture feature vector of the sample digital cloud image and the texture feature vector of the cloud image in the cloud image database according to the following formula,

$\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; <figure-callout\; id="1"\; label="f"\; filenames="H2009102385224E00031.png"\; state="\{\{state\}\}">f</figure-callout></span>}}^{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 16</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\frac{\mathrm{<span\; class="notranslate">\; Min</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; <figure-callout\; id="1"\; label="f\; k"\; filenames="H2009102385224E00031.png"\; state="\{\{state\}\}">f</figure-callout></span><figure-callout\; id="1"\; label="f\; k"\; filenames="H2009102385224E00031.png"\; state="\{\{state\}\}">\; </figure-callout>}}_{\mathrm{<span\; class="notranslate">\; </span>}}^{\mathrm{<span\; class="notranslate">k</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; <figure-callout\; id="1"\; label="f\; k"\; filenames="H2009102385224E00031.png"\; state="\{\{state\}\}">f</figure-callout></span><figure-callout\; id="1"\; label="f\; k"\; filenames="H2009102385224E00031.png"\; state="\{\{state\}\}">\; </figure-callout>}}_{\mathrm{<span\; class="notranslate">\; </span>}}^{\mathrm{<span\; class="notranslate">k</span>}}}<span\; class="notranslate">\; ,</span>$

Among them, F ¹ and F ² represent the eigenvectors of the two ground-based digital cloud images, the larger the value of D, the more similar the two ground-based cloud images, on the contrary, the smaller the value of D, the greater the difference between the two ground-based cloud images;

S203: sort by the D value from large to small, and select the top M sheets with the greatest similarity as the retrieval result, where M is any integer greater than 0 and less than the total number of cloud images in the cloud image database;

S204: Display a cloud map of the retrieval result.

Wherein, before the step S201, there is a step of selecting a sample digital cloud image.

(3) Beneficial effects

The texture feature extraction and retrieval method of the ground-based digital cloud image proposed by the present invention has the following beneficial effects:

(1) In the inventive method, the feature of ground-based digital cloud image is described with an objective 16-dimensional numerical vector, rather than subjective literal description, and the extraction process of feature vector can be carried out fully automatically by computer, can greatly improve ground-based cloud image analysis and management efficiency. This texture feature vector implies the color, texture, and structural features of cloud images, and provides a mathematical basis for automatic analysis, identification, and retrieval of cloud images.

(2) In the method of the present invention, the management process of cloud image library in the background of cloud image retrieval is fully automated, and zero manual effort can be realized. That is to say, the administrator of the cloud library only needs to specify the collection of cloud images or the directory where the cloud images are stored, and the invention automatically analyzes the texture features of the cloud images and saves them in the database for retrieval and calling by the front desk. In addition, the present invention also provides a new retrieval method based on sample cloud image retrieval. The user only needs to specify a pair of sample cloud images, and the invention can automatically retrieve the result cloud images similar to the sample cloud images from the cloud library. Finally, the texture feature retrieval method based on the present invention has excellent performance and high retrieval accuracy.

(3) The pixel classification diagram of the method of the present invention classifies the pixels according to the blue-red band ratio and the pixel brightness, the classification method is simple and efficient, and the number of classes is simplified.

(4) The co-occurrence matrix in the method of the present invention is analyzed based on the pixel category map. Compared with the traditional grayscale-based method, the co-occurrence relationship of the pixel pair in this method has profound connotation and straightforward meaning. It depicts cloud pixels and cloud pixels , the symbiotic relationship between sky pixels and sky pixels, cloud pixels and sky pixels and other pixel pairs under different positional relationships. Moreover, the dimensionality of the co-occurrence matrix of the present invention is very low (4×4), which greatly reduces storage overhead.

(5) the co-occurrence matrix feature in the method of the present invention adopts the expression of histogram, has avoided the process of seeking high-order statistic in the traditional method, and this method conversion is simple, and, in the histogram vector, the meaning of each element is clear, and it represents The frequency of specific pixel category pairs in ground-based cloud images, this frequency feature has a better degree of discrimination in image retrieval.

(6) The average merging strategy is adopted when merging multiple histogram vectors in the method of the present invention. This strategy has high execution efficiency, avoids the loss of any single histogram characteristics, and maintains the simplification of the dimension of the final texture feature vector.

Description of drawings

Fig. 1 is a flow chart of texture feature extraction and retrieval method of ground-based digital cloud image;

Fig. 2 is a schematic diagram of converting the color ground-based digital cloud image into a pixel category map;

Figure 3 is a schematic diagram of two ground-based digital cloud images and their corresponding texture feature vectors;

Figure 4 is the interface of the cloud image retrieval demonstration system based on texture features.

Detailed ways

The texture feature extraction and retrieval method of the ground-based digital cloud image proposed by the present invention is described as follows in conjunction with the accompanying drawings and embodiments.

As shown in Figure 1, it includes the storage process (that is, the texture feature extraction process) and the retrieval process.

The steps of texture extraction are as follows:

Firstly, a digital cloud image is obtained. Generally, the state of the sky is automatically captured by an all-sky imager, and a digital cloud image is generated. The digital cloud image can also be collected by digital cameras/digital video cameras and other electronic devices.

The digital cloud image collected at this time is an RGB three-channel digital cloud image. A digital cloud image usually contains two parts, cloud and sky, and under different weather conditions, the brightness of the cloud also has a relatively large difference. The task of step S101 is to The pixels in the digital cloud image are classified, and the RGB value of the pixel in the original image is replaced by the category label value, and finally the pixel category map is obtained, that is, the color three-channel ground-based digital cloud image is converted into a single-channel pixel category map. The pixel category map is simplified, discarding some secondary information in the original RGB three-channel digital cloud map, and retaining its essential content.

For the pixel at position (x, y) in the original cloud image I, first read the color ground-based digital cloud image file to obtain I _B (x, y), I _R (x, y), and calculate the pixel blue (B) channel The ratio between the value and the red (R) channel value, that is, I _B (x, y)/I _R (x, y), if the ratio value is greater than 1.5, the pixel is a sky pixel, and the category label value C(x, y )=0; if the ratio value is between 1.3 and 1.5, the pixel is a transition pixel between cloud and sky, and the class label value C(x, y)=1; if the ratio value is less than 1.3, the pixel is a cloud pixel , and then according to the brightness information of the original pixels, the cloud pixels are divided into dark clouds and bright clouds, and the brightness value I _V (x, y) is obtained. Color (Blue) component value, red (Red) component value and green (Green) component value I _G (x, y) are calculated, and its calculation formula is: I _V (x, y)=100*Max(I _R ( x, y), I _G (x, y), I _B (x, y))/255, I _V (x, y)<80, it is a dark cloud, and the category label value C(x, y)=2 , otherwise it is a bright cloud, and the category label value C(x, y)=3. That is, α ₁ = 1.5, α ₂ = 1.3, β = 80. The selection of these three thresholds is related to the shooting parameter settings when the color ground-based digital cloud image is acquired. For cloud images with moderate exposure and correct color reproduction, the above threshold selection can be used , and for cloud images with long/short exposure time parameters (brighter/dark image brightness) and abnormal colors, it is first necessary to preprocess the cloud image and correct its brightness and color before using the above threshold. Correction, these three thresholds should be adjusted accordingly in this step to ensure accurate and reasonable extraction of image features. The specific calculation formula is as follows:

$\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; if</span>}{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; R</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>\mathrm{<span\; class="notranslate">\; 1.5</span>}\\ \mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; if</span>}\mathrm{<span\; class="notranslate">\; 1.5</span>}<span\; class="notranslate">\; ></span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; R</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>\mathrm{<span\; class="notranslate">\; 1.3</span>}\\ \mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; if</span>}\mathrm{<span\; class="notranslate">\; 1.3</span>}<span\; class="notranslate">\; ></span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; R</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; and</span>}{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; 80</span>}\\ \mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; if</span>}\mathrm{<span\; class="notranslate">\; 1.3</span>}<span\; class="notranslate">\; ></span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; R</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; and</span>}{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>\mathrm{<span\; class="notranslate">\; 80</span>}\end{array}\right.<span\; class="notranslate">\; ,</span>$

According to the above formula, from left to right, from top to bottom, cyclically calculate the category of each pixel in the cloud image I, and save the category label value in the pixel category map C. Figure 2(a) shows an original ground-based digital cloud image, and Figure 2(b) shows the pixel category map generated by the above steps. For the convenience of display, in Fig. 2(b), blue represents class 0, red represents class 1, white represents class 2, and gray represents class 3.

Step S102 is to obtain the histogram vector of the co-occurrence matrix based on analyzing the pixel category map and establishing the co-occurrence matrix. Obviously, pixels of different categories are arranged in different spatial positions to obtain different cloud images. For example, if a large number of sky pixels are arranged adjacently, a sky region will be formed; if a large number of cloud pixels are arranged adjacently, a cloud may be formed. The category co-occurrence matrix is the result of statistics on the condition that two pixels (called position pairs) satisfying a specific spatial position relationship belong to a specific pixel category respectively. Class co-occurrence matrix analysis consists of three sub-steps, namely class co-occurrence matrix generation, normalized co-occurrence matrix and class co-occurrence matrix feature representation.

Analyze the co-occurrence relationship between any two pixel categories in the pixel category map, and construct the co-occurrence matrix CCM:

$\mathrm{<span\; class="notranslate">\; CCM</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; w</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; <figure-callout\; id="1"\; label="h"\; filenames="H2009102385224E00031.png"\; state="\{\{state\}\}">h</figure-callout></span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; ifC</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; iandC</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&Delta;x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&Delta;y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; <figure-callout\; id="0"\; label="j"\; filenames="H2009102385224E00031.png"\; state="\{\{state\}\}">j</figure-callout></span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; otherwise</span>}\end{array}\right.<span\; class="notranslate">\; ,</span>$

Among them, i and j respectively represent the pixel category, and the values are {0, 1, 2, 3}, (Δx, Δy) represent the offset, w, h represent the length and width of the ground-based digital cloud image respectively, CCM(i, j ) indicates the frequency of occurrence of the position pair where the pixel category at (x, y) is i, and the pixel category at (x+Δx, y+Δy) is j in the pixel category map, and each CCM(i, j) is calculated separately, A 4×4 co-occurrence matrix is obtained.

Normalize the co-occurrence matrix CCM (i, j) according to the following normalization formula to obtain a normalized co-occurrence matrix CCM ^N ;

CCM ^N (i, j) = CCM (i, j)/(wh).

According to the following formula, each element of the normalized co-occurrence matrix CCM ^N is stitched row by row to obtain a 16-dimensional histogram vector F _S ;

F _S = (CCM ^N (0, 0), CCM ^N (0, 1), ..., CCM ^N (3, 3)).

The histogram vector F _S is represented as a feature of the category co-occurrence matrix satisfying the positional relationship (Δx, Δy).

进行线性叠加，并且取平均值，最终合并成一个16维的向量，合并公式如下：Step S103 is to combine the histogram vectors of multiple pixel category co-occurrence matrices to construct the texture feature vector of the ground-based digital cloud image. One location relationship is often not enough to describe the texture features of the ground-based digital cloud image. Therefore, multiple location relationships are constructed, and multiple category co-occurrence matrices and their feature representations are generated accordingly. Four positional relationships are included here, they are {(1, 0), (0, 1), (-1, 0), (0, -1)}, for each positional relationship, step S102 can be looped to get 4 different co-occurrence matrices, and 4 corresponding histogram vectors Then for these four histogram vectors

Carry out linear superposition, and take the average value, and finally merge into a 16-dimensional vector. The merge formula is as follows:

$\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 4</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; S</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; .</span>$

The combined 16-dimensional histogram vector F constitutes the texture feature vector of the ground-based digital cloud image. Figure 3(a) shows a digital cloud map of cumulus fractus, and Figure 3(b) shows the eigenvectors (expressed in histogram) corresponding to Figure 3(a); Figure 3(c) shows a digital cloud map of cumulus cumulus Cloud diagram, Fig. 3(d) shows the corresponding feature vectors (represented by histogram) in Fig. 3(c).

Step S104 saves the texture feature vector constructed in step S103 to the cloud image database for retrieval. Through the ADO interface of SQL Server 2000, write a program to write the feature vector into the database, and it can be realized by using mainstream database management software (such as SQL Server, Oracle) and programming languages (C++, JAVA).

If multiple ground-based digital cloud images are put into storage, the process of steps S101-S104 is repeated until all digital cloud images are processed.

The steps to retrieve a cloud image from the cloud image database are as follows:

First select the sample digital cloud image to be retrieved, and calculate the texture feature vector of the sample after selecting the sample digital cloud image to be retrieved, that is, step S201, this step can be carried out according to the steps S101, S102 and S103 in the texture feature extraction method .

Step S202 is to measure the similarity between the sample digital cloud image to be retrieved and the cloud image in the cloud image database, that is, calculate the similarity between the texture feature vector of the sample digital cloud image and the cloud image texture feature vector in the cloud image database. Calculate the similarity distance between the feature vector F ^e of the sample cloud image and the feature vector F ⁱ of the cloud image in the cloud image database in turn, i=1,...,N, where N represents the number of cloud images in the cloud image database. The similarity distance between eigenvectors is represented by the cross distance, and the calculation formula of the cross distance is as follows:

$\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; e</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 16</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\frac{\mathrm{<span\; class="notranslate">\; Min</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; e</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; e</span>}}}<span\; class="notranslate">\; .</span>$

Step S203 sorts the cloud images in the cloud image database in descending order of the similarity distance D(F ^e , F ⁱ ). Then, select the previous M images with the greatest similarity as the retrieval result, where M can be arbitrarily designated by the user to be any positive integer greater than 0 and less than the total number N of cloud image databases.

Step S204 presents the retrieved results to the user through a specific graphical user interface, and presents M cloud images of the retrieval results to the user. FIG. 4 shows an example of a display interface of the cloud image retrieval system.

Application of the present invention in stand-alone cloud image management and retrieval:

The invention is especially suitable for managing large-scale cloud images, and provides a convenient retrieval interface. For example, Professor Zhang, a meteorological scientist, has accumulated tens of thousands of ground-based digital cloud images in his long-term meteorological observation and research, and he stores them in one or several directories on the hard disk of his personal computer. Whenever he gets a new cloud image, he needs to find similar cloud images from the collection of typical cloud images and conduct comparative research. However, facing tens of thousands of digital cloud images on the computer, he cannot find similar cloud images quickly and efficiently. With the application of the present invention, Professor Zhang can quickly extract texture features from tens of thousands of digital cloud images and import them into the database, and this import process is "one-time import and permanent use"; when he gets a new cloud image, he can use the present invention to The cloud image retrieval method retrieves cloud images with similar content in the cloud image database within a few seconds.

Application of the present invention in network cloud image retrieval:

Meteorological management departments (such as Meteorological Bureau) or research departments (such as Academy of Meteorological Sciences) have built a large-scale ground-based cloud image database, and apply the present invention to extract the texture features of cloud images, import them into the database, and provide Internet-based Internet cloud image retrieval service, and announced the URL to access the cloud image retrieval service. When an observer takes a cloud image, he wants to retrieve cloud images similar to the cloud image from the cloud image database for comparative research. Therefore, he can log in to the website, upload and search the cloud images he has taken, and the system will return similar cloud images in the cloud image database within a few seconds.

Application of the present invention in automatic weather station ground-based digital cloud image comprehensive management system:

All-sky cloud imaging from ground-based automatic weather stations captures the state of the sky and generates digital cloud images. Then, send the digital cloud image to the remote host through the transmission channel. The host runs the texture feature extraction algorithm of the present invention to extract the texture features of the cloud image, and import the cloud image and feature vectors into the database. The host computer can also issue a cloud image retrieval service interface, and users can retrieve cloud images from the weather station at any time. In this way, the collection, analysis, management and retrieval functions of ground-based digital cloud images can be realized fully automatically without human intervention, forming an intelligent and automated ground-based digital cloud image comprehensive management system.

The above embodiments are only used to illustrate the present invention, but not to limit the present invention. Those of ordinary skill in the relevant technical field can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, all Equivalent technical solutions also belong to the category of the present invention, and the scope of patent protection of the present invention should be defined by the claims.

Claims (7)

1. a texture feature extraction method of ground-based digital cloud image, is characterized in that, comprises the following steps: S101: Convert the color RGB three-channel ground-based digital cloud image into a single-channel pixel category image using the following formula: $\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; if</span>}{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; R</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}\\ \mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; if</span>}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ></span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; R</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\\ \mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; if</span>}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ></span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; R</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; and</span>}{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}\\ \mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; if</span>}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ></span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; R</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; and</span>}{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}\end{array}\right.<span\; class="notranslate">\; ,</span>$ Among them, I _B (x, y), I _R (x, y) represent the blue component value and the red component value of the pixel at the coordinate (x, y) in the input color ground-based digital cloud image respectively, and their values are passed through the color ground-based digital cloud image The image file is directly read, I _V (x, y) represents the brightness value of the pixel at the coordinate (x, y) in the input color ground-based digital cloud image, and C(x, y) represents the coordinate (x, y) in the color ground-based digital cloud image The category label of the pixel at , α ₁ , α ₂ are the threshold parameters of the blue-red band ratio, and β is the brightness threshold parameter; S102: Obtain the histogram vector of the co-occurrence matrix by analyzing the pixel category map and establishing the co-occurrence matrix; S103: Merge the histogram vectors of multiple pixel category co-occurrence matrices to construct a texture feature vector of the ground-based digital cloud image; S104: Save the texture feature vector constructed in S103 to a cloud image database. 2. the texture feature extraction method of ground-based digital cloud image as claimed in claim 1, is characterized in that, the _IV (x, y) value in the described step S101 is by coordinate (x, y) place in input color ground-based digital cloud image The blue component value, red component value and green component value of the pixel are calculated, and the calculation formula is: _IV (x, y) = 100*Max(I _R (x, y), I _G (x, y), I _B (x, y))/255, where I _G (x, y) is the green component value of the pixel at coordinates (x, y). 3. the texture feature extraction method of ground-based digital cloud image as claimed in claim 1 or 2, is characterized in that, described α ₁ value is 1.5, α ₂ value is 1.3, and β value is 80. 4. the texture feature extraction method of ground-based digital cloud image as claimed in claim 1, is characterized in that, described step S102 comprises the step: S1021: Analyze the co-occurrence relationship between any two pixel categories in the pixel category map, and construct a co-occurrence matrix CCM: $\mathrm{<span\; class="notranslate">\; CCM</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; w</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; ifC</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; iandC</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&Delta;x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&Delta;y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; j</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; otherwise</span>}\end{array}\right.<span\; class="notranslate">\; ,</span>$ Among them, i and j respectively represent the pixel category, and the values are {0, 1, 2, 3}, (Δx, Δy) represent the offset, w, h represent the length and width of the ground-based digital cloud image respectively, CCM(i, j ) represents the frequency of occurrence of the pixel category at (x, y) in the pixel category map is i, and at the same time (x+Δx, y+Δy) is the frequency of the location of the pixel category j, and calculate each CCM (i, j), respectively, Get a 4×4 co-occurrence matrix; S1022: Normalize the co-occurrence matrix CCM(i, j) according to the following normalization formula to obtain a normalized co-occurrence matrix CCM ^N ; CCM ^N (i, j) = CCM (i, j)/(wh). S1023: According to the following formula, each element of the normalized co-occurrence matrix CCM ^N is spliced row by row to obtain a 16-dimensional histogram vector F _S ; F _S = (CCM ^N (0, 0), CCM ^N (0, 1), . . . , CCM ^N (3, 3)). 5. the texture feature extraction method of ground-based digital cloud image as claimed in claim 1, is characterized in that, described step S103 comprises the step: S1031: There are L position offsets of the co-occurrence relationship between two pixel categories, namely (Δx ₁ , Δy ₁ ), (Δx ₂ , Δy ₂ ), ..., (Δx _L-1 , Δy _L-1 ), ( Δx _L , Δy _L ), obtain L different co-occurrence matrices and L corresponding histogram vectors according to step S102

S1032: Process the L histogram vectors linearly superimposed and averaged according to the following formula

$\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; L</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; S</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; .</span>$ The texture feature vector of the 16-dimensional ground-based digital cloud image is obtained. 6. A texture feature retrieval method of ground-based digital cloud image, is characterized in that, comprises the following steps: S201: Calculate the texture feature vector of the sample digital cloud image according to the steps of steps S101, S102 and S103 of claim 1; S202: Calculate the similarity between the texture feature vector of the sample digital cloud image and the texture feature vector of the cloud image in the cloud image database according to the following formula, $\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 16</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\frac{\mathrm{<span\; class="notranslate">\; Min</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; ,</span>$ Among them, F ¹ and F ² represent the eigenvectors of the two ground-based digital cloud images, the larger the value of D, the more similar the two ground-based cloud images, on the contrary, the smaller the value of D, the greater the difference between the two ground-based cloud images; S203: sort by the D value from large to small, and select the top M sheets with the greatest similarity as the retrieval result, where M is any integer greater than 0 and less than the total number of cloud images in the cloud image database; S204: Display a cloud map of the retrieval result. 7. The texture feature retrieval method of ground-based digital cloud image according to claim 6, characterized in that, before the step S201, a step of selecting a sample digital cloud image is included.

Images