Deep learning based integrated satellite remote sensing image analysis system and method

Abstract

The application discloses a comprehensive satellite remote sensing image analysis system and method based on deep learning, belongs to the technical field of remote sensing image analysis, and comprises an image acquisition and preprocessing module, a water storage capacity detection module, a surface floating object detection module and an overall state evaluation module. The application obtains a visible light remote sensing image of a current lake through a satellite and pre-processes the visible light remote sensing image to obtain a pre-processed visible light remote sensing image, acquires water storage capacity based on the pre-processed visible light remote sensing image, and then acquires three surface floating object state parameters, i.e., a local area ratio of floating objects, a local quantity of floating objects and a local dispersion degree of floating objects, so that the surface floating object states of each sub-region of the corresponding lake can be accurately represented, and subsequent overall state evaluation work can be performed in combination with the water storage capacity, so that the evaluation result is more accurate.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image analysis technology, and specifically to a comprehensive satellite remote sensing image analysis system and method based on deep learning. Background Technology

[0002] Satellite remote sensing imagery, primarily captured or recorded by satellites, shows different features of the Earth's surface and their reflection or radiation information in different spectral bands. With the continuous advancement of remote sensing technology and the widespread application of satellites, the acquisition of satellite remote sensing imagery will become more convenient and efficient.

[0003] Satellite remote sensing imagery can identify a wide variety of objects and phenomena on the Earth's surface. For example, through deep learning and machine learning algorithms, it can monitor natural objects, man-made objects, and special objects or phenomena. Natural objects include water bodies, vegetation, soil, and topography; man-made objects include buildings, roads and transportation facilities, vehicles, and agricultural facilities; and special objects or phenomena include disasters, mineral resources, and climate change. In summary, satellite remote sensing imagery technology has broad application prospects and significant value, capable of detecting and identifying many common objects and phenomena, providing crucial data support for scientific research, policy making, and business decision-making.

[0004] Existing methods for assessing the overall condition of lakes mainly rely on manual detection to obtain relevant parameters, and then use these parameters combined with human experience to obtain lake condition monitoring results. These methods suffer from problems such as high labor costs and low monitoring efficiency.

[0005] The above problems urgently need to be solved. To address this, a comprehensive satellite remote sensing image analysis system based on deep learning is proposed. Summary of the Invention

[0006] The technical problem to be solved by this invention is: how to solve the problems of high labor costs and low monitoring efficiency in existing methods of assessing the overall state of lakes, and provides a comprehensive satellite remote sensing image analysis system based on deep learning.

[0007] The present invention solves the above-mentioned technical problems through the following technical solutions: the present invention includes an image acquisition and preprocessing module, a water storage detection module, a surface floating object detection module, and an overall state assessment module;

[0008] The image acquisition and preprocessing module is used to acquire visible light remote sensing images of the current lake via satellite and preprocess them to obtain preprocessed visible light remote sensing images.

[0009] The water storage detection module is used to detect the current water storage of the lake based on the preprocessed visible light remote sensing image and obtain the water storage A.

[0010] The surface floating object detection module is used to detect the surface floating object state parameters of each sub-region of the current lake in the preprocessed visible light remote sensing image, and obtain the local area ratio S of the floating objects. i , number of floating objects in a local area C i With the local dispersion F of floating objects i , where the subscript i represents the i-th subregion;

[0011] The overall state assessment module is used to calculate the overall state score of the lake based on the obtained surface floating object state parameters and water storage of each sub-region of the current lake.

[0012] Furthermore, the image acquisition and preprocessing module includes a satellite imaging unit and a preprocessing unit. The satellite imaging unit is used to capture visible light remote sensing images of the current lake on a satellite. The preprocessing unit is used to perform noise reduction and image enhancement processing on the visible light remote sensing images to obtain preprocessed visible light remote sensing images.

[0013] Furthermore, the water storage detection module includes a target detection unit, a contour detection unit, and a water storage acquisition unit. The target detection unit is used to detect lake targets in the preprocessed visible light remote sensing image using a trained lake target detection model to obtain a lake target detection box. The contour detection unit is used to perform outer contour detection on the lake region within the lake target detection box using the contour detection function in OpenCV, obtain the coordinates of each point on the outer contour line of the lake region in the image, and calculate the area of ​​the lake region in the image based on the coordinates of each point in the image, denoted as Sp. The water storage acquisition unit is used to search and compare the area Sp of the lake region in the image in a preset area-water level height database to obtain the water level height Hp corresponding to the area Sp of the lake region in the image, and then search and compare the water level height Hp in a preset water level height-water storage capacity database to obtain the water storage capacity corresponding to the water level height Hp, i.e., obtain the current water storage capacity A of the lake.

[0014] Furthermore, the surface floating object detection module includes a sub-region division unit, an area ratio calculation unit, a quantity calculation unit, and a dispersion calculation unit; the sub-region division unit is used to divide the current lake area into multiple sub-regions; the area ratio calculation unit is used to calculate the local area ratio S of floating objects in each sub-region based on the divided sub-regions. i The quantity calculation unit is used to count and record the local quantity C of floating objects in each sub-region according to the divided sub-regions. i The dispersion calculation unit is used to calculate the local dispersion F of floating objects in each sub-region based on the divided sub-regions. i .

[0015] Furthermore, the specific processing procedure of the surface floating object detection module is as follows:

[0016] Step S31: Based on the coordinates of each point on the outer contour line of the lake region obtained by the contour detection unit, calculate the geometric center point Z of the lake region using OpenCV, and draw a straight line L through the geometric center point Z along the direction parallel to the long side of the lake target detection box. z Intersect the two short sides of the lake target detection box at points D1 and D2 respectively. Calculate the midpoint of line segment ZD1, denoted as point D3, and simultaneously calculate the midpoint of line segment ZD2, denoted as point D4. Points D1, D3, ZD2, D4, and D2 are arranged sequentially on line segment D1D2. Draw straight lines H through points D3, ZD2, and D4 along directions parallel to the short sides of the lake target detection box. D3 H Z H D4 And record the line H D3 H Z H D4 The coordinates of the intersection points with the outer contour line of the lake area are used to divide the current lake area into four sub-regions;

[0017] Step S32: Based on the divided sub-regions, calculate the ratio of the local floating area S in each sub-region. i The calculation formula is as follows:

[0018] S i =Pz i / P i ;

[0019] Among them, Pz i P represents the total area of ​​all floating objects in the i-th sub-region in the image. i This represents the area of ​​the i-th sub-region in the image;

[0020] Step S33: Based on the divided sub-regions, count and record the local number C of floating objects in each sub-region. i ;

[0021] Step S34: Calculate the local dispersion F of floating objects in each sub-region based on the divided sub-regions. i The calculation formula is as follows:

[0022] F i = (Q1 + Q2 + ... + Q) j-1 +Q j );

[0023] Among them, Q j This represents the distance in the image between the geometric center point of the j-th floating object within the i-th sub-region and the geometric center point of that sub-region.

[0024] Furthermore, in steps S32 to S34, the floating objects in the sub-region are detected using a trained floating object detection model.

[0025] Furthermore, in the overall state assessment module, the formula for calculating the overall state score T of the lake is as follows:

[0026] T = S' wavg +C′ wavg +F′ wavg +A′;

[0027] Among them, S′ wavg The ratio of the local area of ​​floating objects in each sub-region to S i The weighted average S wavg The result obtained after dimensionless processing; C′ wavg The local number C of floating objects in each sub-region i The weighted average C wavg The result obtained after dimensionless processing; F′ wavg The local dispersion F of floating objects in each sub-region i The weighted average F wavg The result obtained after dimensionless processing; A′ is the result obtained after dimensionless processing of water storage A.

[0028] Furthermore, in the overall lake condition score T:

[0029] Weighted average S wavg The calculation formula is as follows:

[0030] S wavg = w1*S1+w2*S2+w3*S3+w4*S4;

[0031] Weighted average C wavg The calculation formula is as follows:

[0032] C wavg = w1*C1+w2*C2+w3*C3+w4*C4;

[0033] Weighted average F wavg The calculation formula is as follows:

[0034] F wavg = w1*F1 + w2*F2 + w3*F3 + w4*F4;

[0035] Among them, w1, w2, w3, and w4 are the weights of the corresponding sub-regions, which are positively correlated with the distance between the geometric center point of the corresponding sub-region and the geometric center point of the lake region. That is, the farther the geometric center point of the sub-region is from the geometric center point of the lake region, the greater the corresponding weight.

[0036] This invention also provides a comprehensive satellite remote sensing image analysis method based on deep learning. This method utilizes the aforementioned image analysis system to analyze satellite remote sensing images and then assesses the overall state of the lake. The method includes the following steps:

[0037] Step S1: Acquire a visible light remote sensing image of the current lake via satellite and preprocess it to obtain a preprocessed visible light remote sensing image;

[0038] Step S2: Based on the preprocessed visible light remote sensing image, detect the current water storage of the lake and obtain the water storage volume;

[0039] Step S3: Divide the lake area into sub-regions, and then detect the surface floating object state parameters of each sub-region of the current lake;

[0040] Step S4: Calculate the overall state score of the lake based on the obtained surface floating object state parameters and water storage of each sub-region of the current lake.

[0041] Compared with the prior art, the present invention has the following advantages: This integrated satellite remote sensing image analysis system and method based on deep learning acquires visible light remote sensing images of the current lake by satellite and preprocesses them to obtain preprocessed visible light remote sensing images. Based on the preprocessed visible light remote sensing images, the water storage is obtained, and then three surface floating object state parameters, namely, the local area ratio of floating objects, the local number of floating objects, and the local dispersion of floating objects, are obtained. These parameters can accurately characterize the surface floating object state of each sub-region of the corresponding lake. Combined with the water storage, subsequent overall state assessment is carried out, making the assessment results more accurate. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of the structure of the integrated satellite remote sensing image analysis system based on deep learning in an embodiment of the present invention;

[0043] Figure 2 This is a schematic diagram of sub-region division in an embodiment of the present invention (taking a near-elliptical lake as an example);

[0044] Figure 3 This is a flowchart illustrating the integrated satellite remote sensing image analysis method based on deep learning in an embodiment of the present invention. Detailed Implementation

[0045] The embodiments of the present invention are described in detail below. These embodiments are implemented based on the technical solution of the present invention, and provide detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the following embodiments.

[0046] like Figure 1As shown, this embodiment provides a technical solution: a comprehensive satellite remote sensing image analysis system based on deep learning, including an image acquisition and preprocessing module, a water storage detection module, a surface floating object detection module, and an overall state assessment module;

[0047] The image acquisition and preprocessing module is used to acquire visible light remote sensing images of the current lake via satellite and preprocess them to obtain preprocessed visible light remote sensing images.

[0048] In this embodiment, the image acquisition and preprocessing module includes a satellite imaging unit and a preprocessing unit. The satellite imaging unit is used to capture visible light remote sensing images of the current lake on a satellite and send them to the preprocessing unit for preprocessing. The preprocessing unit is used to perform noise reduction and image enhancement processing on the visible light remote sensing images to obtain preprocessed visible light remote sensing images.

[0049] More specifically, the satellite imaging unit is an ultra-high-definition camera installed on the satellite, capable of acquiring ultra-high-definition visible light images. In this embodiment, noise reduction methods include, but are not limited to, Gaussian filtering, median filtering, and mean filtering, to significantly reduce image noise. Image enhancement methods include, but are not limited to, histogram equalization and grayscale transformation, to improve image quality.

[0050] The water storage detection module is used to detect the current water storage of the lake and obtain the water storage A.

[0051] In this embodiment, the water storage detection module includes a target detection unit, a contour detection unit, and a water storage acquisition unit. The target detection unit is used to detect lake targets in the preprocessed visible light remote sensing image using a trained lake target detection model to obtain a lake target detection box. The contour detection unit is used to perform outer contour detection on the lake region in the lake target detection box using the contour detection function in OpenCV, obtain the coordinates of each point on the outer contour line of the lake region in the image, and calculate the area of ​​the lake region in the image based on the coordinates of each point in the image, denoted as Sp. The water storage acquisition unit is used to search and compare the area Sp of the lake region in the image in a preset area-water level height database to obtain the water level height Hp corresponding to the area Sp of the lake region in the image, and then search and compare the water level height Hp in a preset water level height-water storage database to obtain the water storage corresponding to the water level height Hp, that is, to obtain the current water storage A of the lake. This invention associates the area of ​​a lake region in an image with its water level and establishes a corresponding database, thereby enabling convenient and rapid acquisition of the water level. Then, by utilizing the correlation between the water level and the water storage capacity, the current water storage capacity of the lake can be accurately and quickly obtained.

[0052] More specifically, the lake target detection model is trained based on the SSD network (deep learning network).

[0053] More specifically, in the water storage acquisition unit, the area-water level database has a pre-set mapping relationship between the area of ​​a lake region in an image and the corresponding water level. This mapping relationship is obtained by acquiring a large amount of area data of the lake region in the image in the early stage and manually measuring the water level under the corresponding area data, and finally obtaining the mapping relationship through mathematical fitting.

[0054] More specifically, in the water storage acquisition unit, the water level-storage database pre-establishes a mapping relationship between water level and corresponding water storage. This mapping relationship is obtained during the initial simulation calculation of water storage. During the initial water storage simulation calculation, elevation data of the lake area's topographic surface is obtained using a digital elevation model. By analyzing the elevation data, the lake's boundaries and bottom area can be determined. Combined with the water level, the water storage can then be calculated. The corresponding mapping relationship can then be established between the calculated water storage and the corresponding water level.

[0055] The surface floating object detection module is used to detect the surface floating object state parameters of each sub-region of the current lake and obtain the local area ratio S of the floating objects. i , number of floating objects in a local area C i With the local dispersion F of floating objects i , where the subscript i represents the i-th sub-region; in this invention, by using three surface floating object state parameters, namely the local area ratio of floating objects, the local number of floating objects, and the local dispersion of floating objects, the surface floating object state of each sub-region of the lake can be accurately characterized, which is beneficial for subsequent overall state assessment and makes the assessment results more accurate.

[0056] In this embodiment, the surface floating object detection module includes a sub-region division unit, an area ratio calculation unit, a quantity calculation unit, and a dispersion calculation unit; the sub-region division unit is used to divide the current lake area into multiple sub-regions; the area ratio calculation unit is used to calculate the local area ratio S of floating objects in each sub-region based on the divided sub-regions. i The quantity calculation unit is used to count and record the local quantity C of floating objects in each sub-region according to the divided sub-regions. i The dispersion calculation unit is used to calculate the local dispersion F of floating objects in each sub-region based on the divided sub-regions. i .

[0057] More specifically, when using OpenCV to calculate the geometric center point Z of a lake region, the concept of the centroid of a polygon is used to calculate the center point; the centroid can be obtained by calculating the weighted average position of all vertices, and then the geometric center point Z can be calculated.

[0058] More specifically, the specific processing procedure of the surface floating object detection module is as follows:

[0059] Step S31: Based on the coordinates of each point on the outer contour line of the lake region obtained by the contour detection unit, calculate the geometric center point Z of the lake region using OpenCV, and draw a straight line L through the geometric center point Z along the direction parallel to the long side of the lake target detection box. z Intersect the two short sides of the lake target detection box at points D1 and D2 respectively. Calculate the midpoint of line segment ZD1, denoted as point D3, and simultaneously calculate the midpoint of line segment ZD2, denoted as point D4. Points D1, D3, ZD2, D4, and D2 are arranged sequentially on line segment D1D2. Draw straight lines H through points D3, ZD2, and D4 along directions parallel to the short sides of the lake target detection box. D3 H Z H D4 And record the line H D3 H Z H D4 The coordinates of the intersection points with the outer contour line of the lake area are used to divide the current lake area into four sub-regions, such as... Figure 2 As shown;

[0060] Step S32: Based on the divided sub-regions, calculate the ratio of the local floating area S in each sub-region. i The calculation formula is as follows:

[0061] S i =Pz i / P i ;

[0062] Among them, Pz i P represents the total area of ​​all floating objects in the i-th sub-region in the image. The calculation method is the same as calculating the area Sp of a lake region in the image: first, a trained floating object detection model is used to detect floating objects within the sub-region; then, outer contour detection is performed; finally, the total area of ​​all floating objects in the i-th sub-region in the image is calculated. i This represents the area of ​​the i-th sub-region in the image;

[0063] It should be noted that the floating object detection model is trained based on the YOLO v3 network;

[0064] Step S33: Based on the divided sub-regions, count and record the local number C of floating objects in each sub-region.i The number of floating objects is obtained by using a floating object detection model to detect floating objects in a sub-region.

[0065] Step S34: Calculate the local dispersion F of floating objects in each sub-region based on the divided sub-regions. i The calculation formula is as follows:

[0066] F i = (Q1 + Q2 + ... + Q) j-1 +Q j );

[0067] Among them, Q j F represents the distance in the image between the geometric center point of the j-th floating object within the i-th sub-region and the geometric center point of that sub-region. i The larger the value, the more dispersed the floating objects in that sub-region. The calculation methods for the geometric center point of floating objects and the geometric center point of sub-regions are the same as those for the geometric center point of lake regions, and will not be repeated here.

[0068] The overall state assessment module is used to calculate the overall state score of the lake based on the obtained surface floating object state parameters and water storage of each sub-region of the current lake.

[0069] In this embodiment, the formula for calculating the overall lake condition score T is as follows:

[0070] T = S' wavg +C′ wavg +F′ wavg +A′;

[0071] Among them, S′ wavg The ratio of the local area of ​​floating objects in each sub-region to S i The weighted average S wavg The result obtained after dimensionless processing; C′ wavg The local number C of floating objects in each sub-region i The weighted average C wavg The result obtained after dimensionless processing; F′ wavg The local dispersion F of floating objects in each sub-region i The weighted average F wavg The result obtained after dimensionless processing; A′ is the result obtained after dimensionless processing of water storage A.

[0072] It should be noted that the dimensionless processing in this embodiment is a normalization processing.

[0073] In more specific terms:

[0074] Weighted average S wavg The calculation formula is as follows:

[0075] S wavg = w1*S1+w2*S2+w3*S3+w4*S4;

[0076] Weighted average C wavg The calculation formula is as follows:

[0077] C wavg = w1*C1+w2*C2+w3*C3+w4*C4;

[0078] Weighted average F wavg The calculation formula is as follows:

[0079] F wavg = w1*F1 + w2*F2 + w3*F3 + w4*F4;

[0080] Among them, w1, w2, w3, and w4 are the weights of the corresponding sub-regions, which are positively correlated with the distance between the geometric center point of the corresponding sub-region and the geometric center point of the lake region. That is, the farther the geometric center point of the sub-region is from the geometric center point of the lake region, the greater the corresponding weight, which is conducive to calculating a more accurate overall lake status score.

[0081] like Figure 3 As shown, this embodiment also provides a comprehensive satellite remote sensing image analysis method based on deep learning. This method utilizes the aforementioned image analysis system to analyze satellite remote sensing images and then assesses the overall state of the lake. The method includes the following steps:

[0082] Step S1: Acquire a visible light remote sensing image of the current lake via satellite and preprocess it to obtain a preprocessed visible light remote sensing image;

[0083] Step S2: Based on the preprocessed visible light remote sensing image, detect the current water storage of the lake and obtain the water storage volume;

[0084] Step S3: Divide the lake area into sub-regions, and then detect the surface floating object state parameters of each sub-region of the current lake;

[0085] Step S4: Calculate the overall state score of the lake based on the obtained surface floating object state parameters and water storage of each sub-region of the current lake.

[0086] It should be noted that the specific implementation process of steps S1 to S4 above can be found in the processing process of each module and unit in the image analysis system mentioned above, and will not be repeated here.

[0087] In summary, the deep learning-based integrated satellite remote sensing image analysis system and method described above acquires visible light remote sensing images of the current lake via satellite and preprocesses them to obtain preprocessed visible light remote sensing images. Based on the preprocessed visible light remote sensing images, the water storage capacity is obtained. Then, three surface floating object state parameters are obtained: local floating object area ratio, local floating object quantity, and local floating object dispersion. These parameters can accurately characterize the surface floating object state of each sub-region of the corresponding lake. Combined with the water storage capacity, subsequent overall state assessment is performed, making the assessment results more accurate.

[0088] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A deep learning-based integrated satellite remote sensing image analysis system, characterized in that, Comprising: An image acquisition and preprocessing module, a water storage capacity detection module, a surface floating object detection module, and a whole state evaluation module; The image acquisition and preprocessing module is configured to acquire a visible light remote sensing image of a current lake by a satellite and pre-process the visible light remote sensing image to obtain a pre-processed visible light remote sensing image. The water storage capacity detection module is configured to detect the water storage capacity of the current lake based on the pre-processed visible light remote sensing image, and obtain the water storage capacity A. The surface floating object detection module is used for detecting the surface floating object state parameters of each sub-region of the current lake from the preprocessed visible light remote sensing image, and obtaining the local area ratio S of the floating object i , the local number C of the floating object i , and the local dispersion F of the floating object i , wherein the subscript i represents the i th sub-region. The surface floating object detection module comprises a sub-region division unit, an area ratio calculation unit, a quantity calculation unit and a dispersion degree calculation unit; the sub-region division unit is used for dividing the current lake region into a plurality of sub-regions; the area ratio calculation unit is used for calculating the local area ratio S of the floating object of each sub-region according to the divided sub-regions i ; the quantity calculation unit is used for counting and recording the local quantity C of the floating object of each sub-region according to the divided sub-regions i ; and the dispersion degree calculation unit is used for calculating the local dispersion degree F of the floating object of each sub-region according to the divided sub-regions i . The specific processing procedure of the surface floating object detection module is as follows: Step S31: based on the coordinates of each point on the outer contour line of the lake region obtained by the contour detection unit in the image, the geometric center point Z of the lake region is calculated using OpenCV, and a straight line L is made along the direction parallel to the long side of the lake target detection frame through the geometric center point Z z , respectively intersecting the two short sides of the lake target detection frame at point D1 and point D2, calculating the midpoint of the line segment ZD1, denoted as point D3, and simultaneously calculating the midpoint of the line segment ZD2, denoted as point D4, and arranging the points D1, D3, Z, D4, D2 in order on the line segment D1D2, making straight lines H D3 , H Z , H D4 through points D3, Z, D4 along the direction parallel to the short side of the lake target detection frame, and recording the intersection coordinates of the straight lines H D3 , H Z , H D4 and the outer contour line of the lake region, and further dividing the current lake region into four sub-regions; Step S32: According to the divided sub-regions, the local area ratio S of floating objects in each sub-region is calculated i The calculation formula is as follows: S i = Pz i / P i ; wherein Pz i represents the total sum of the areas of all the floating objects in the i-th sub-region in the image, P i represents the area of the i-th sub-region in the image; Step S33: According to the divided sub-regions, the local quantity C of floating objects in each sub-region is counted and recorded i ; Step S34: According to the divided sub-regions, the local dispersivity F of each sub-region is calculated i The calculation formula is as follows: F i = (Q1+Q2+...+Q j-1 +Q j ); wherein Q j represents the distance between the geometric center of the jth floating object in the ith sub-region and the geometric center of the sub-region in the image; The whole state evaluation module is configured to calculate a whole state score of the lake according to the surface floating object state parameters and the water storage capacity of each sub-region of the current lake. In the whole state evaluation module, the calculation formula of the whole state score T of the lake is as follows: T = S' wavg + C' wavg + F' wavg + A' S' = S / S wavg S' = S / S i S' = S / S wavg C' = C / C wavg C' = C / C i F' = F / F wavg F' = F / F wavg F' = F / F i F' = F / F wavg A' = A / A In the whole state score T of the lake: Weighted average value S wavg The calculation formula is as follows: S wavg = w1 S1 + w2 S2 + w3 S3 + w4 S4; Weighted average value C wavg The calculation formula is as follows: C wavg = w1 C1 + w2 C2 + w3 C3 + w4 C4; Weighted average value F wavg The calculation formula is as follows: F wavg = w1 F1 + w2 F2 + w3 F3 + w4 F4; Wherein, w1, w2, w3, and w4 are weights of the corresponding sub-regions, and are positively correlated with the distance between the geometric center point of the corresponding sub-region and the geometric center point of the lake region, that is, the farther the geometric center point of the sub-region is from the geometric center point of the lake region, the greater the corresponding weight is.

2. The deep learning-based integrated satellite remote sensing image analysis system of claim 1, wherein, The image acquisition and preprocessing module comprises a satellite shooting unit and a preprocessing unit. The satellite shooting unit is configured to shoot and acquire the visible light remote sensing image of the current lake on the satellite. The preprocessing unit is configured to perform noise reduction and image enhancement processing on the visible light remote sensing image to obtain the pre-processed visible light remote sensing image.

3. The deep learning-based integrated satellite remote sensing image analysis system of claim 2, wherein, The water storage capacity detection module comprises a target detection unit, a contour detection unit, and a water storage capacity acquisition unit. The target detection unit is configured to detect the lake target in the pre-processed visible light remote sensing image by using a trained lake target detection model to obtain a lake target detection frame. The contour detection unit is configured to perform outer contour detection on the lake region in the lake target detection frame by using a contour detection function in OpenCV, to obtain the coordinates of each point on the outer contour line of the lake region in the image, and to calculate the area of the lake region in the image according to the coordinates of each point in the image, denoted as Sp. The water storage capacity acquisition unit is configured to find and compare the area Sp of the lake region in the image in a pre-set area-water level height database to obtain the water level height Hp corresponding to the area Sp of the lake region in the image, and then find and compare the water level height Hp in a pre-set water level height-water storage capacity database to obtain the water storage capacity corresponding to the water level height Hp, that is, to obtain the water storage capacity A of the current lake.

4. The deep learning-based integrated satellite remote sensing image analysis system of claim 1, wherein, In steps S32 to S34, the floating object in the sub-region is detected by using the trained floating object detection model.

5. A comprehensive satellite remote sensing image analysis method based on deep learning, characterized in that, The image analysis system according to any one of claims 1 to 4 is used to analyze the satellite remote sensing image, and further evaluate the whole state of the lake, comprising the following steps: Step S1: acquiring a visible light remote sensing image of a current lake by a satellite and pre-processing the visible light remote sensing image to obtain a pre-processed visible light remote sensing image; Step S2: detecting the water storage capacity of the current lake based on the pre-processed visible light remote sensing image, and obtaining the water storage capacity; Step S3: Sub-region division operation is performed on the lake region, and then the surface floating object state parameters of each sub-region of the current lake are detected; Step S4: According to the obtained surface floating object state parameters of each sub-region of the current lake and the water storage capacity, the overall state score of the lake is calculated.

Images