Ecological environment detection method and system based on multi-spectral remote sensing fusion technology

Abstract

The present application relates to the field of remote sensing image processing technology, and specifically to an ecological environment detection method and system based on multispectral remote sensing fusion technology, the method comprising: collecting multispectral remote sensing images of a target area; determining the foreground information rate of each pixel level in the remote sensing images of each band; calculating the distortion coefficient of the remote sensing images of each band; obtaining the relative loss coefficient of each pixel level in the remote sensing images of each band; dividing the remote sensing images of each band into super-pixel blocks, determining the band independence of each super-pixel block in the remote sensing images of each band; screening the candidate pixel points in each super-pixel block, determining the atmospheric light value, and defogging the remote sensing images of each band in combination with the dark channel prior method; and performing ecological environment detection based on the remote sensing images of all bands after defogging. Thus, the accuracy of ecological environment detection is improved.

Description

Technical Field

[0001] The present application relates to the technical field of remote sensing image processing, and in particular to an ecological environment detection method and system based on multispectral remote sensing fusion technology. Background Art

[0002] Ecological environment detection refers to the monitoring and evaluation of the natural environment and its ecosystem, which helps to rationally utilize and manage natural resources, such as water resources, forests and land, and ensure the sustainable use of resources. Currently, with the rapid development of remote sensing technology, remote sensing technology can quickly obtain soil information in a large area and identify land types, which helps to evaluate the impact of land use on the environment and improve the accuracy of ecological environment detection.

[0003] However, remote sensing technology is greatly affected by seasons and climate, especially in autumn, winter, cloudy and foggy days. The scattering and absorption of remote sensing signals by the atmosphere will seriously affect the data quality. The traditional atmospheric correction process requires defogging of remote sensing images, while the defogging effects of multispectral remote sensing images in different bands are inconsistent. The accuracy of atmospheric light value determination is low, resulting in distortion of spectral characteristics and large errors in ecological environment detection. Summary of the invention

[0004] In order to solve the above technical problems, the purpose of this application is to provide an ecological environment detection method and system based on multispectral remote sensing fusion technology. The technical solutions adopted are as follows:

[0005] In a first aspect, the present application provides an ecological environment detection method based on multispectral remote sensing fusion technology, the method comprising the following steps:

[0006] Collect multispectral remote sensing images of the target area;

[0007] Obtain the statistical histogram of all pixels in the remote sensing image of each band; analyze the asymmetry of the distribution of all pixels at each pixel level in the statistical histogram in the image, as well as the distance relationship between each pixel and the background pixel at each pixel level, and determine the foreground information rate of each pixel level in the remote sensing image of each band;

[0008] Analyze the difference in the number of foreground pixels and background pixels in the remote sensing image of each band, as well as the difference in brightness, to determine the distortion coefficient of the remote sensing image of each band; sort all pixel levels in the remote sensing image of each band based on the foreground information rate, analyze the difference in pixel distribution probability between each pixel level in the remote sensing image of each band and its sorted adjacent pixel level, as well as the difference in the degree of pixel confusion between the adjacent pixel level, to determine the relative loss coefficient of each pixel level in the remote sensing image of each band;

[0009] The remote sensing image of each band is divided into super-pixel blocks, and the band independence of each super-pixel block in the remote sensing image of each band is determined by combining the relative loss coefficient and the foreground information rate of each pixel level in each super-pixel block, as well as the number of pixels at each pixel level; the candidate pixels in each super-pixel block are screened by combining the band independence and the relative loss coefficient, and the atmospheric light value is determined based on the brightness of the candidate pixels of all super-pixel blocks in the remote sensing image of each band, and the remote sensing image of each band is defogged by combining the dark channel prior method;

[0010] The remote sensing images of all bands after defogging are fused, and the vegetation areas in the fused images are identified using a deep learning model to obtain the vegetation coverage of the target area for ecological environment detection.

[0011] In one embodiment, the process of determining the foreground information rate is as follows:

[0012] For each pixel level in the remote sensing image of each band, the skewness of the x-coordinate and the skewness of the y-coordinate of all pixel points in the image are calculated, and the difference between the skewness of the x-coordinate and the skewness of the y-coordinate is used as the distribution range coefficient of each pixel level;

[0013] The pixel level with the largest distribution probability in the statistical histogram of the remote sensing image of each band is recorded as a class of pixel levels, and the shortest distance between each pixel point in each pixel level and the pixel point corresponding to the class of pixel levels is calculated;

[0014] The foreground information rate is a fusion result of the mean of the shortest distances of all pixel points at each pixel level and the distribution range coefficient.

[0015] In one embodiment, the determination of the distortion coefficient includes:

[0016] Threshold segmentation is performed on the foreground information rate of all pixel levels in each band remote sensing image, and pixel points with pixel levels greater than or equal to the threshold are used as foreground pixel points, and the remaining pixel points are used as background pixel points; the discrete degree of the horizontal spacing between all any two pixel levels greater than or equal to the threshold in the corresponding statistical histogram is calculated;

[0017] Calculate the ratio of the average pixel value of all background pixels in each band remote sensing image to the discrete degree, which is recorded as the first ratio; calculate the ratio of the number of all foreground pixels to the number of background pixels in each band remote sensing image, which is recorded as the second ratio;

[0018] The distortion coefficient is a fusion result of the first ratio and the second ratio in each band remote sensing image.

[0019] In one embodiment, the sorting is to arrange the pixel levels in descending order of foreground information rate.

[0020] In one embodiment, the calculation formula of the relative loss coefficient is:

[0021] ; In the formula, is the relative loss coefficient of the vth pixel level in the remote sensing image of the bth band, is the distribution probability of the vth pixel level in the remote sensing image of the bth band, is the distribution probability of the v+1th pixel level in the remote sensing image of the bth band, is the degree of confusion of the pixel values ​​of all pixels at the vth pixel level in the remote sensing image of the bth band corresponding to the pixel values ​​at the same position in the remote sensing image of the b+1th band, is the degree of confusion of the pixel values ​​of all pixels at the v+1th pixel level in the remote sensing image of the bth band corresponding to the pixel values ​​at the same position in the remote sensing image of the b+1th band, and norm is the normalization function.

[0022] In one embodiment, the calculation formula of the band independence is:

[0023] ; In the formula, is the band independence of the sth superpixel block in the remote sensing image of the bth band, is the number of pixel-level categories in the sth superpixel block in the remote sensing image of the bth band, is the number of pixels at the vth pixel level in the sth superpixel block in the remote sensing image of the bth band, is the number of all pixels in the sth superpixel block in the remote sensing image of the bth band, is the foreground information rate of the vth pixel in the sth superpixel block in the bth band of the remote sensing image, is the relative loss coefficient of the vth pixel level in the remote sensing image of the bth band.

[0024] In one embodiment, screening candidate pixel points in each superpixel block includes:

[0025] Obtain a dark channel image of each band remote sensing image, arrange all pixels in each superpixel block in each band remote sensing image in descending order according to the dark channel value, calculate a normalized value of the product of the distortion coefficient of each band remote sensing image and the band independence of each superpixel block therein, calculate the ratio of the number of pixels in each superpixel block in each band remote sensing image to the number of preset candidate pixels, and the product of the normalized value as the screening ratio of each superpixel block in each band remote sensing image;

[0026] After arranging the pixel points in each pixel block, the pixel points with the above screening ratio are used as candidate pixel points in each super pixel block.

[0027] In one embodiment, the atmospheric light value is the pixel value with the maximum brightness among all candidate pixels of the remote sensing image of each band.

[0028] In one embodiment, the ecological environment detection includes:

[0029] If the vegetation coverage rate of the target area exceeds a preset percentage, the ecological environment of the target area is judged to be good, otherwise, the ecological environment of the target area is judged to be poor.

[0030] In the second aspect, an embodiment of the present application also provides an ecological environment detection system based on multi-spectral remote sensing fusion technology, including a memory, a processor, and a computer program stored in the memory and running on the processor, and when the processor executes the computer program, the steps of any one of the above methods are implemented.

[0031] This application has at least the following beneficial effects:

[0032] The present application collects multispectral remote sensing images of the target area; obtains a statistical histogram of all pixels in the remote sensing images of each band; analyzes the difference in skewness between the x-coordinate and the y-coordinate of all pixels at each pixel level in the statistical histogram, as well as the distance relationship between each pixel and the background pixel at each pixel level, to determine the foreground information rate of each pixel level in the remote sensing images of each band; the foreground information rate reflects the possibility that the pixel corresponding to the pixel level in the multispectral remote sensing image belongs to the foreground pixel, thereby improving the accuracy of the division of foreground information and background information in the multispectral remote sensing image; analyzes the foreground pixel in the remote sensing image of each band The difference in the number of pixels and the brightness between the pixels of the multispectral remote sensing image and the background pixels is used to determine the distortion coefficient of the remote sensing image of each band; the distortion coefficient reflects the change in the details of the multispectral remote sensing image and the change in the degree of fog interference with the image as the band increases, which improves the accuracy of determining the degree of detail loss of the multispectral remote sensing image; based on the foreground information rate, all pixel levels in the remote sensing image of each band are sorted, and the difference in the distribution probability of the pixels of each pixel level and its adjacent pixel level in the remote sensing image of each band, as well as the difference in the degree of pixel confusion with its adjacent pixel level, are analyzed to determine the pixel level of each pixel level in the remote sensing image of each band. Relative loss coefficient; the relative loss coefficient reflects the degree of detail loss of pixel levels adjacent to the foreground information rate between adjacent bands, and embodies the correlation between each pixel level and the band; the remote sensing image of each band is divided into super-pixel blocks, and the band independence of each super-pixel block in the remote sensing image of each band is determined by combining the relative loss coefficient and foreground information rate of each pixel level in each super-pixel block, as well as the number of pixels at each pixel level; the determination of band independence improves the accuracy of screening pixel levels that are not affected by band characteristics; combining the band independence and the relative loss coefficient, the candidate pixel points in each super-pixel block are screened, based on The brightness of candidate pixels of all superpixel blocks in the remote sensing image of each band is used to determine the atmospheric light value, and the remote sensing image of each band is defogged in combination with the dark channel prior method; the accuracy of determining the atmospheric light value is improved, thereby improving the defogging effect of the image and ensuring the clarity of the multispectral remote sensing image; the remote sensing images of all bands after defogging are fused, and the vegetation area in the fused image is identified using a deep learning model to obtain the vegetation coverage rate of the target area for ecological environment detection; the problem of large errors in ecological environment detection caused by poor image defogging effect is avoided, and the accuracy of ecological environment detection is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0034] Figure 1 A flowchart of the steps of an ecological environment detection method based on multispectral remote sensing fusion technology provided in one embodiment of the present application;

[0035] Figure 2 Determine the flow chart for atmospheric light values. DETAILED DESCRIPTION

[0036] In order to further explain the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the following is a detailed description of the ecological environment detection method and system based on multi-spectral remote sensing fusion technology proposed in the present application, its specific implementation method, structure, features and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0037] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0038] The specific scheme of the ecological environment detection method and system based on multispectral remote sensing fusion technology provided by the present application is described in detail below with reference to the accompanying drawings.

[0039] See also Figure 1 , which shows a flowchart of the steps of an ecological environment detection method based on multispectral remote sensing fusion technology provided by an embodiment of the present application, the method includes the following steps:

[0040] S1, collect multispectral remote sensing images of the target area.

[0041] This embodiment uses remote sensing satellites to obtain multi-spectral remote sensing images of the target area; wherein, the selection of the target area can be set by the implementer according to actual conditions, and this embodiment does not limit this.

[0042] In special weather conditions, such as autumn and winter, when there are more clouds and fog, the collected multispectral remote sensing images will inevitably be affected by fog, resulting in blurry multispectral remote sensing images. Therefore, it is necessary to perform defogging on the collected multispectral remote sensing images.

[0043] The dark channel prior method is a commonly used image defogging algorithm. Its principle is based on an observed image statistical law, that is, in most natural scenes, pixels in non-sky areas have lower brightness values ​​on at least one color channel. Based on this observation, the dark channel prior method assumes that the darkest area in the image (usually the non-sky area) has a lower brightness value on at least one color channel, and this brightness value is related to the fog concentration, so as to obtain an estimated value of the fog concentration and perform defogging based on it.

[0044] However, in multispectral remote sensing images, the information expression of the same target area in different bands is different, which leads to different fog concentration estimates in remote sensing images of different bands of the same target area. The defogging effect varies greatly, affecting the accuracy of land type identification in the target area and causing errors in ecological environment detection.

[0045] Therefore, the multispectral remote sensing images analyzed in this embodiment mainly include the following bands: blue light band (450-495nm), green light band (495-570nm), red light band (620-750nm), near infrared band (750-1300nm), and short-wave infrared band (1300-2500nm), a total of 5 bands. Multispectral remote sensing images in different bands have different prominent information expressions. For example, the green light band is more sensitive to vegetation coverage information, and the short-wave infrared band is more sensitive to the mineral composition and moisture content of the soil.

[0046] S2, obtaining the statistical histogram of all pixels in the remote sensing image of each band; analyzing the difference in the skewness of the x-coordinate and the y-coordinate of all pixels at each pixel level in the statistical histogram, as well as the distance relationship between each pixel and the background pixel at each pixel level, to determine the foreground information rate of each pixel level in the remote sensing image of each band.

[0047] In the dark channel prior method, the atmospheric light A value is used to represent the intensity of light scattered by the atmosphere in the scene, which directly affects the clarity and contrast of the defogging image. The color and brightness of the foreground information in the multispectral remote sensing image will affect the determination of the atmospheric light A value. The foreground information highlighted by multispectral remote sensing images of different bands is not the same, and the defogging process needs to ensure that the foreground information in each remote sensing image is preserved and enhanced. Therefore, this embodiment analyzes the foreground information in the multispectral remote sensing image of each band.

[0048] Firstly, for the multispectral remote sensing image of each band, a pixel statistical histogram is obtained, in which each pixel value represents a pixel level, and analysis is performed based on the position information of all pixel points at each pixel level, specifically, the mean of the x-coordinates of all pixel points at each pixel level and the standard deviation of the x-coordinates are calculated, and the skewness of the x-coordinates of all pixel points at each pixel level is calculated based on the mean and the standard deviation, and the skewness of the y-coordinates of all pixel points at each pixel level is obtained by using the same calculation method as the skewness of the x-coordinates, and the difference between the skewness of the x-coordinates of all pixel points at each pixel level and the skewness of the y-coordinates is used as the distribution range coefficient of each pixel level. The larger the distribution range coefficient, the more the pixel points of this type of pixel level are biased towards a specific local distribution.

[0049] It should be noted that, in this embodiment, the absolute value of the difference between the skewness of the x-coordinate and the skewness of the y-coordinate of all pixel points at each pixel level is used as the distribution range coefficient of each pixel level; the implementer can choose other feasible difference calculation methods according to actual conditions, such as the square of the difference, the ratio, etc.

[0050] In the pixel statistical histogram of the multispectral remote sensing image of each band, the pixel level with the largest distribution probability is recorded as a type of pixel level, which represents the background information in the remote sensing image, and the background information is the main geological type of the area where the remote sensing image is collected; the distribution probability is calculated as the ratio of the number of pixel points at each pixel level to the number of all pixel points in the remote sensing image.

[0051] The shortest distance between each pixel point at each pixel level in the multispectral remote sensing image of each band and the pixel points of the first type of pixel level is calculated, and based on the shortest distance and the distribution range coefficient, the foreground information rate of each pixel level in the remote sensing image of each band is determined, wherein the foreground information rate is a fusion result of the mean of the shortest distance of all pixel points at each pixel level and the distribution range coefficient.

[0052] It should be noted that fusion means combining multiple variables, which can be calculated by addition, multiplication, or a combination of addition and multiplication. This embodiment uses multiplication as the calculation method for fusion.

[0053] In this embodiment, the expression of the foreground information rate of each pixel level in the remote sensing image of each band can be:

[0054] ; In the formula, represents the foreground information rate of the i-th pixel level in the remote sensing image of each band, Represents the distribution range coefficient of the i-th pixel level in the remote sensing image of each band, It represents the shortest distance between the pixel point of the i-th pixel level in the remote sensing image of each band and the pixel point of the corresponding class of pixel level. It represents the mean of the shortest distances between all pixels at the i-th pixel level in the remote sensing image of each band and the pixels at the corresponding pixel level.

[0055] It should be understood that the larger the mean of the shortest distance is, the farther the i-th pixel level is from the background information, and the more likely it is to be a specific pixel level in geology; the larger the foreground information rate is, the more likely it is that the i-th pixel level is foreground information.

[0056] S3, analyzing the difference in the number of foreground pixels and background pixels in the remote sensing image of each band, and the difference in brightness, to determine the distortion coefficient of the remote sensing image of each band; sorting all pixel levels in the remote sensing image of each band based on the foreground information rate, analyzing the difference in pixel distribution probability between each pixel level and its adjacent pixel levels in the remote sensing image of each band, as well as the difference in pixel confusion degree with its adjacent pixel levels, to determine the relative loss coefficient of each pixel level in the remote sensing image of each band.

[0057] As the band becomes larger, the scattering effect of the multispectral remote sensing image decreases, that is, the interference from fog will become smaller, but at the same time, the richness and details of the image information will increase as the band becomes larger. Therefore, the characteristics when the band is low are weak image details but thick fog, and the characteristics when the band is large are rich image details but thin fog. Therefore, the multispectral remote sensing images of each band are analyzed.

[0058] Specifically, the OTSU method is used to obtain the segmentation threshold of the foreground information rate of all pixel levels in each band of the multispectral remote sensing image, and the pixel points corresponding to the pixel level whose foreground information rate is greater than or equal to the segmentation threshold are taken as foreground pixel points, and the pixel points corresponding to the pixel level whose foreground information rate is less than the segmentation threshold are taken as background pixel points; the implementer can choose other existing feasible threshold segmentation algorithms according to the actual situation, and this embodiment does not limit it here.

[0059] For the pixel statistical histogram of the multispectral remote sensing image of each band, the lateral spacing between any two pixel levels that are greater than or equal to the segmentation threshold, that is, the absolute value of the difference between the two pixel levels, is calculated, and the discrete degree of all the lateral intervals in the pixel statistical histogram is calculated to indicate the strength of the texture distribution of the foreground information in the multispectral remote sensing image.

[0060] It should be noted that the degree of dispersion can be calculated by using variance, standard deviation, coefficient of variation, etc. This embodiment uses variance as the calculation method for the degree of dispersion.

[0061] Based on the above analysis, the distortion coefficient of the remote sensing image of each band is calculated, specifically: the ratio of the average pixel value of all background pixels in the remote sensing image of each band to the discrete degree is calculated, recorded as the first ratio, and the ratio of the number of all foreground pixels to the number of background pixels in the remote sensing image of each band is calculated, recorded as the second ratio;

[0062] The distortion coefficient is a fusion result of the first ratio and the second ratio in each band remote sensing image.

[0063] It should be understood that fusion means combining multiple variables, which can be specifically achieved by addition, multiplication, or a combination of addition and multiplication. This embodiment uses multiplication as the calculation method for fusion.

[0064] In this embodiment, the calculation expression of the distortion coefficient is:

[0065] ; Where F represents the distortion coefficient of the remote sensing image in each band, Represents the number of foreground pixels in the remote sensing image of each band, Represents the number of background pixels in the remote sensing image of each band, Represents the mean pixel value of all background pixels in the remote sensing image of each band, Indicates the discrete degree of the remote sensing image in each band.

[0066] Will Recorded as the second ratio, when it is in the low band, the second ratio is smaller, and when it is in the high band, the second ratio is larger; It is recorded as the first ratio. The larger the first ratio is, the larger the fog in the background of the remote sensing image is. Since the characteristic of a lower band is that the image details are weak but the fog is thick, the first ratio is larger when the fog is heavy for low-band remote sensing images, but the second ratio is smaller, and the image information distortion is limited. The characteristic of a larger band is that the image details are more but the fog is thin. Therefore, the first ratio is smaller when the fog is less, but the second ratio is larger. Although the fog is small, the loss of details is relatively large. Therefore, the greater the distortion coefficient of the remote sensing image.

[0067] In multispectral remote sensing images of different bands, the pixel levels of the prominent bias are different, and the foreground information distribution is also different; for the multispectral remote sensing images of each band, the pixel levels in its pixel statistical histogram are rearranged in descending order according to the foreground information rate to obtain the histogram after the pixel level is rearranged.

[0068] Since the spectral information between adjacent bands is highly correlated, it is helpful to extract the characteristics and information of the ground objects. Therefore, the fusion of multispectral remote sensing images mostly involves adjacent bands. Therefore, the relative loss coefficient of each pixel level in the remote sensing image of each band is calculated according to the pixel level changes of adjacent bands. The specific calculation method is:

[0069] ; In the formula, is the relative loss coefficient of the vth pixel level in the remote sensing image of the bth band, is the distribution probability of the vth pixel level in the remote sensing image of the bth band, is the distribution probability of the v+1th pixel level in the remote sensing image of the bth band, is the degree of confusion of the pixel values ​​of all pixels at the vth pixel level in the remote sensing image of the bth band corresponding to the pixel values ​​at the same position in the remote sensing image of the b+1th band, is the degree of confusion of the pixel values ​​of all pixels at the v+1th pixel level in the remote sensing image of the bth band corresponding to the pixel values ​​at the same position in the remote sensing image of the b+1th band, and norm is the normalization function.

[0070] It should be noted that in the process of calculating the relative loss coefficient, the degree of chaos is calculated using information entropy. Based on the degree of chaos in the distribution of extractable pixel values, the implementer can choose other existing feasible calculation methods. This embodiment does not limit this, and the vth pixel level and the v+1th pixel level are the pixel level orderings in the rearranged histogram.

[0071] The vth pixel level and its adjacent pixel level have similar foreground information rates, that is, they may both be relatively prominent in the same band, but there is a relative foreground and background relationship between them; for the vth pixel level in the bth band, the foreground information rate of its next adjacent v+1th pixel level is lower than that of the vth pixel level, so the pixel distribution probability ratio of the vth pixel level to the v+1th pixel level represents the foreground information ratio when the v+1th pixel level is used as background information and the vth pixel level is used as foreground information;

[0072] The information entropy value representing the pixel values ​​of all pixels at the vth pixel level in the remote sensing image of the b+1th band. The larger the value, the more all pixels at the vth pixel level in the bth band are transformed into other pixel levels in the next band, showing richer information, that is, the higher the foreground loss rate of the vth pixel level in the b+1th band; The ratio is the ratio of the foreground loss rate of all pixels at the vth pixel level and the v+1th pixel level in the b+1th band remote sensing image;

[0073] Represents the difference between the ratio of the foreground information of the vth pixel level and the v+1th pixel level to the foreground loss rate. The larger the value, the smaller the loss of the foreground information of the vth pixel level relative to the v+1th pixel level. Conversely, the smaller the value, the greater the relative loss of the foreground information of the vth pixel level. The difference can be positive or negative, so the norm normalization function is used to normalize it, and the constant 1 is subtracted for inverse proportional transformation, so The larger the value is, the greater the relative loss coefficient of the vth pixel level in the remote sensing image of the bth band is.

[0074] This embodiment calculates the information loss between each pixel level and the pixel level with relative foreground and background relationship according to the foreground information rate. It can be reflected that for two pixel levels with similar foreground information rates in the b-th band, when the band changes, the pixel level in the relative foreground information has a greater loss, which means that the band correlation of the v-th pixel level is stronger, and such pixel level is less likely to be used as an effective pixel level in the b-th band remote sensing image.

[0075] S4, divide the remote sensing image of each band into super-pixel blocks, and determine the band independence of each super-pixel block in the remote sensing image of each band by combining the relative loss coefficient and foreground information rate of each pixel level in each super-pixel block, as well as the number of pixels at each pixel level; combine the band independence and the relative loss coefficient to screen the candidate pixel points in each super-pixel block, determine the atmospheric light value based on the brightness of the candidate pixel points of all super-pixel blocks in the remote sensing image of each band, and defog the remote sensing image of each band by combining the dark channel prior method.

[0076] Due to the blurring effect of the atmosphere, the multispectral remote sensing image of each band is segmented into superpixels to obtain superpixel blocks. The present embodiment uses the SLIC (Simple Linear Iterative Clustering) algorithm for superpixel segmentation. The SLIC algorithm is a well-known technology. The implementer can select other feasible algorithms according to the actual situation, and the present embodiment does not limit it here.

[0077] According to the relative loss coefficients of all pixel levels in the superpixel block, the band independence of each superpixel block in each band of remote sensing images is calculated. The specific calculation method is:

[0078] ; In the formula, is the band independence of the sth superpixel block in the remote sensing image of the bth band, is the number of pixel-level categories in the sth superpixel block in the remote sensing image of the bth band, is the number of pixels at the vth pixel level in the sth superpixel block in the remote sensing image of the bth band, is the number of all pixels in the sth superpixel block in the remote sensing image of the bth band, is the foreground information rate of the vth pixel in the sth superpixel block in the bth band of the remote sensing image, is the relative loss coefficient of the vth pixel level in the remote sensing image of the bth band.

[0079] represents the proportion of pixels at the vth pixel level in the sth superpixel block as the weight, Represents the ratio of the foreground information rate of the vth pixel level to the relative loss coefficient. The larger the ratio, the more likely the vth pixel level is to be band-irrelevant foreground information. represents the weighting of all pixel levels of the s-th superpixel block to obtain the band independence of the s-th superpixel block, The larger the value, the higher the band independence of the sth superpixel block, and the less affected by the band characteristics the atmospheric light value A screened from the pixel level within the superpixel block.

[0080] In this embodiment, a 3*3 local window is set to obtain a dark channel image of each band of remote sensing images; wherein, the acquisition of the dark channel image is an existing well-known technology, and this embodiment will not be described in detail here.

[0081] In the dark channel image, the pixels in each superpixel block are arranged from large to small according to the dark channel intensity; then the band independence of each superpixel block is used as a weight to screen out the candidate pixels in each superpixel block in the dark channel image, and the screening ratio of each superpixel block in the remote sensing image of each band is calculated. The specific calculation method is:

[0082] ; In the formula, is the screening ratio of the sth superpixel block in the remote sensing image of the bth band, 0.1% is the default candidate pixel number ratio, norm is the normalization function, represents the distortion coefficient of the remote sensing image in the bth band, is the band independence of the sth superpixel block in the remote sensing image of the bth band, is the number of all pixels in the sth superpixel block in the remote sensing image of the bth band.

[0083] The larger the value, that is, the higher the degree of distortion and the higher the band independence, the more pixel candidate points should be in the superpixel block. The subsequently screened atmospheric light value A can not only adapt to the defogging effect of different bands, but also be conducive to the fusion of multi-band remote sensing images.

[0084] After arranging the dark channel intensity from large to small, the sth superpixel block is arranged The pixel point of is taken as the candidate pixel point, and the same method is used to obtain the candidate pixel points in each super pixel block.

[0085] According to all the candidate pixels obtained in the multispectral remote sensing image of each band, the pixel value with the maximum brightness is taken as the atmospheric light value A. The flow chart of atmospheric light value determination is as follows: Figure 2 shown.

[0086] According to the atmospheric light value A of each band of the multispectral remote sensing image, the atmospheric transmittance of each band can be calculated, and the multispectral remote sensing image of each band is defogged using the dark channel prior method to obtain a fog-free image of each band. The dark channel prior method is an existing well-known technology and will not be described in detail in this embodiment.

[0087] S5, performs image fusion on the remote sensing images of all bands after defogging, uses the deep learning model to identify the vegetation area in the fused image, and obtains the vegetation coverage rate of the target area for ecological environment detection.

[0088] Open source multispectral remote sensing images with land type annotations are used as training samples. This embodiment uses 5,000 training samples. The implementer can determine the number of training samples according to the actual situation. This embodiment does not limit this. All training samples are divided into training set and validation set in a ratio of 7:3, and input into the convolutional neural network for training. The loss function is the cross entropy function, and the gradient descent method is used for training until convergence. Among them, this embodiment uses a semantic segmentation neural network to identify the land type in the multispectral remote sensing image. The land type is divided into vegetation and others. The annotation method is that the pixels belonging to vegetation are manually labeled as 1, and other pixels are manually labeled as 0. The convolutional neural network and the training process of the convolutional neural network are both existing well-known technologies, and this embodiment will not be described in detail here.

[0089] The fog-free images obtained in all bands of this embodiment are fused using an image fusion algorithm, and then input into the trained convolutional neural network for land type recognition to obtain the vegetation coverage area. The ratio of the vegetation coverage area to the total area of ​​the target area is taken as the vegetation coverage rate of the target area. If the vegetation coverage rate of the target area exceeds the preset percentage, the ecological environment of the target area is judged to be good; otherwise, the ecological environment of the target area is judged to be poor. In this embodiment, the preset percentage is 30%, and the implementer can set it according to the actual situation. This embodiment does not limit it here.

[0090] Based on the same inventive concept as the above method, an embodiment of the present application also provides an ecological environment detection system based on multi-spectral remote sensing fusion technology, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the steps of any one of the above-mentioned ecological environment detection methods based on multi-spectral remote sensing fusion technology are implemented.

[0091] It should be noted that the above sequence of the embodiments of the present application is for description only and does not represent the advantages and disadvantages of the embodiments. The above is a description of a specific embodiment of this specification. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0092] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0093] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present application should be included in the protection scope of the present application.

Claims

1. An ecological environment detection method based on multispectral remote sensing fusion technology, characterized in that: The method comprises the following steps: Collect multispectral remote sensing images of the target area; Obtain the statistical histogram of all pixels in the remote sensing image of each band; analyze the asymmetry of the distribution of all pixels at each pixel level in the statistical histogram in the image, as well as the distance relationship between each pixel and the background pixel at each pixel level, and determine the foreground information rate of each pixel level in the remote sensing image of each band; Analyze the difference in the number of foreground pixels and background pixels in the remote sensing image of each band, as well as the difference in brightness, to determine the distortion coefficient of the remote sensing image of each band; sort all pixel levels in the remote sensing image of each band based on the foreground information rate, analyze the difference in pixel distribution probability between each pixel level in the remote sensing image of each band and its sorted adjacent pixel level, as well as the difference in the degree of pixel confusion between the adjacent pixel level, to determine the relative loss coefficient of each pixel level in the remote sensing image of each band; The remote sensing image of each band is divided into super-pixel blocks, and the band independence of each super-pixel block in the remote sensing image of each band is determined by combining the relative loss coefficient and the foreground information rate of each pixel level in each super-pixel block, as well as the number of pixels at each pixel level; the candidate pixels in each super-pixel block are screened by combining the band independence and the relative loss coefficient, and the atmospheric light value is determined based on the brightness of the candidate pixels of all super-pixel blocks in the remote sensing image of each band, and the remote sensing image of each band is defogged by combining the dark channel prior method; The remote sensing images of all bands after defogging are fused, and the vegetation areas in the fused images are identified using a deep learning model to obtain the vegetation coverage of the target area for ecological environment detection.

2. The ecological environment detection method based on multi-spectral remote sensing fusion technology as claimed in claim 1, characterized in that: The process of determining the foreground information rate is as follows: For each pixel level in the remote sensing image of each band, the skewness of the x-coordinate and the skewness of the y-coordinate of all pixel points in the image are calculated, and the difference between the skewness of the x-coordinate and the skewness of the y-coordinate is used as the distribution range coefficient of each pixel level; The pixel level with the largest distribution probability in the statistical histogram of the remote sensing image of each band is recorded as a class of pixel levels, and the shortest distance between each pixel point in each pixel level and the pixel point corresponding to the class of pixel levels is calculated; The foreground information rate is a fusion result of the mean of the shortest distances of all pixel points at each pixel level and the distribution range coefficient.

3. The ecological environment detection method based on multi-spectral remote sensing fusion technology as claimed in claim 1, characterized in that: The determination of the distortion coefficient includes: Threshold segmentation is performed on the foreground information rate of all pixel levels in each band remote sensing image, and pixel points with pixel levels greater than or equal to the threshold are used as foreground pixel points, and the remaining pixel points are used as background pixel points; the discrete degree of the horizontal spacing between all any two pixel levels greater than or equal to the threshold in the corresponding statistical histogram is calculated; Calculate the ratio of the average pixel value of all background pixels in each band remote sensing image to the discrete degree, which is recorded as the first ratio; calculate the ratio of the number of all foreground pixels to the number of background pixels in each band remote sensing image, which is recorded as the second ratio; The distortion coefficient is a fusion result of the first ratio and the second ratio in each band remote sensing image.

4. The ecological environment detection method based on multi-spectral remote sensing fusion technology as claimed in claim 1, characterized in that: The sorting is to arrange the pixel levels in descending order of foreground information rate.

5. The ecological environment detection method based on multi-spectral remote sensing fusion technology as claimed in claim 1, characterized in that: The calculation formula of the relative loss coefficient is: ; In the formula, is the relative loss coefficient of the vth pixel level in the remote sensing image of the bth band, is the distribution probability of the vth pixel level in the remote sensing image of the bth band, is the distribution probability of the v+1th pixel level in the remote sensing image of the bth band, is the degree of confusion of the pixel values ​​of all pixels at the vth pixel level in the remote sensing image of the bth band corresponding to the pixel values ​​at the same position in the remote sensing image of the b+1th band, is the degree of confusion of the pixel values ​​of all pixels at the v+1th pixel level in the remote sensing image of the bth band corresponding to the pixel values ​​at the same position in the remote sensing image of the b+1th band, and norm is the normalization function.

6. The ecological environment detection method based on multi-spectral remote sensing fusion technology as claimed in claim 1, characterized in that: The calculation formula of the band independence is: ; In the formula, is the band independence of the sth superpixel block in the remote sensing image of the bth band, is the number of pixel-level categories in the sth superpixel block in the remote sensing image of the bth band, is the number of pixels at the vth pixel level in the sth superpixel block in the remote sensing image of the bth band, is the number of all pixels in the sth superpixel block in the remote sensing image of the bth band, is the foreground information rate of the vth pixel in the sth superpixel block in the bth band of the remote sensing image, is the relative loss coefficient of the vth pixel level in the remote sensing image of the bth band.

7. The ecological environment detection method based on multi-spectral remote sensing fusion technology as claimed in claim 1, characterized in that: The screening of candidate pixel points in each superpixel block includes: Obtain a dark channel image of each band remote sensing image, arrange all pixels in each superpixel block in each band remote sensing image in descending order according to the dark channel value, calculate a normalized value of the product of the distortion coefficient of each band remote sensing image and the band independence of each superpixel block therein, calculate the ratio of the number of pixels in each superpixel block in each band remote sensing image to the number of preset candidate pixels, and the product of the normalized value as the screening ratio of each superpixel block in each band remote sensing image; After arranging the pixel points in each pixel block, the pixel points with the above screening ratio are used as candidate pixel points in each super pixel block.

8. The ecological environment detection method based on multi-spectral remote sensing fusion technology as claimed in claim 1, characterized in that: The atmospheric light value is the pixel value with the maximum brightness among all the candidate pixel points of the remote sensing image of each band.

9. The ecological environment detection method based on multi-spectral remote sensing fusion technology as claimed in claim 1, characterized in that: The ecological environment detection includes: If the vegetation coverage rate of the target area exceeds a preset percentage, the ecological environment of the target area is judged to be good, otherwise, the ecological environment of the target area is judged to be poor.

10. An ecological environment detection system based on multispectral remote sensing fusion technology, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 9 are implemented.

Images