Coal surface moisture prediction method based on 3D reconstruction and illumination self-adaption

Abstract

The invention relates to the technical field of moisture prediction, in particular to a 3D reconstruction and illumination adaptive coal surface moisture prediction method, which comprises the following steps of: planning an unmanned aerial vehicle route to collect an image, reconstructing coal pile surface space structure information, forming a local triangular grid, dividing an illumination continuous area, establishing a space corresponding relation between the illumination area and a hyperspectral pixel, and predicting the moisture content of a coal pile. And geometric registration and reflection transition correction are executed, depression points and shoulder positions of a reflection curve are detected in the identification interval, moisture absorption characteristics are constructed, a moisture migration link is established in the slope direction, and a coal surface moisture prediction result is obtained. According to the method, a spatial structure and illumination distribution are coupled, a dynamic correction relation is established to unify reflection information, image registration and spectrum reconstruction are combined to strengthen geometric consistency, a moisture migration trend is extracted by utilizing depression point and slope correlation, link and accumulation distribution is formed, and a corrected reflection characteristic represents a spectrum attenuation rule. The spatial continuity and numerical stability of moisture identification are ensured, and the water content grade division is coherent and comparable.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of water prediction, in particular to a coal surface water prediction method based on 3D reconstruction and light adaptation. BACKGROUND

[0002] The technical field of water prediction includes related technologies for accurately predicting water content on the surface or inside an object by measuring and analyzing, which is widely used in agriculture, environmental monitoring, industrial production and other fields. Especially in the drying, storage and processing of materials, accurate detection and prediction of water content are of great significance to improve product quality and production efficiency. The core content of water prediction technology usually involves estimating and predicting water content based on different sensors, optical methods, thermal imaging technology, etc. combined with mathematical models or algorithms. With the continuous progress of artificial intelligence, machine learning and other technologies, water prediction methods are constantly innovating to achieve higher accuracy, wider applicability and real-time monitoring.

[0003] Among them, the coal surface water prediction method based on 3D reconstruction and light adaptation refers to a technical method for predicting the water content on the surface of coal using three-dimensional reconstruction technology and light adaptation technology. It mainly solves several technical problems in coal surface water prediction. Especially in traditional manual sampling or sensor measurement, there are problems such as limited coverage and low efficiency, which are difficult to meet the real-time monitoring needs of large-area stockyards. Therefore, 3D reconstruction technology is used to model the surface of coal, and the detection accuracy of water distribution is improved by accurately reconstructing the surface morphology. Combined with light adaptation technology, the light changes are compensated to ensure stable water prediction under different light conditions. The method involved solves the influence of surface complexity and environmental factors on water prediction accuracy that previous methods cannot handle by reasonably combining the morphological characteristics of the coal surface with the light conditions.

[0004] The existing technology relies on a single light assumption in coal surface water monitoring. Changes in light intensity and incident angle will cause reflectivity distortion, causing the response of hyperspectral data to shift. It is difficult to maintain geometric alignment during image stitching, resulting in surface structure information breaking and reflection differences superimposing, and water distribution showing discrete and uneven characteristics. Traditional prediction methods cannot track the indirect influence of light differences on water identification, and spatial local measurement is prone to reflection errors in areas with high and low fluctuations in the stack, lacking continuous expression of water migration paths, which further affects the accuracy and consistency of overall water distribution judgment. SUMMARY

[0005] In order to solve the technical problems existing in the prior art, the present application provides a coal surface water prediction method based on 3D reconstruction and light adaptation. The technical solution is as follows: A coal surface moisture prediction method with 3D reconstruction and light self-adaptation, comprising the following steps: S1: Obtain image frames of the coal pile area and number them to form an image sequence, detect matching feature points in the overlapping area, calculate parallax information to recover three-dimensional coordinates, complete spatial alignment and reconstruct the surface morphology of the coal pile, extract the elevation difference and connection order of adjacent points, and generate surface spatial structure information; S2: According to the elevation difference and connection order of adjacent points in the surface spatial structure information, a local triangular mesh is established to form a patch set, the patch normal is fitted to determine the tilt direction, the sun and the backlight are determined in combination with the sun direction, the shadow and the occlusion boundary are identified, and the light partition information is generated; S3: Based on the three-dimensional nodes composed of adjacent patch boundary points in the light partition information, the spatial position and boundary of the light area are extracted and the spatial correspondence with the hyperspectral pixels is established, the sun and the backlight samples are separated, the reflection difference of the same pixel is compared and the transition is corrected, and the light corrected reflection information is generated; S4: According to the corrected spectral curve in the light corrected reflection information, the reflectivity change trend is analyzed, the continuous descending interval is determined, the depth of the depression point and the shoulder width of the shoulder are detected, the reflection attenuation order is established and the slope direction and the sun label are associated, the same region is aggregated, and the moisture absorption characteristic information is generated.

[0006] As a further scheme of the present application, the surface spatial structure information includes a set of spatial point coordinates, a neighboring topological relationship, an elevation difference parameter and a boundary connectivity attribute, the light partition information includes a light area range, an incident angle parameter, an occlusion boundary feature and patch index information, the light corrected reflection information includes a pixel coordinate mapping table, a reflectivity correction curve, a multi-band reflectivity sequence and a light area correspondence relationship, and the moisture absorption characteristic information includes a depression point spectral position, a shoulder distance feature, a slope direction association parameter and a sun classification label.

[0007] As a further scheme of the present application, the surface spatial structure information acquisition step is: S101: Obtain the flight line parameters and geographic coordinates of the coal pile operation area, set the flight height, flight direction and image overlap rate, use a UAV to carry a visible light camera to continuously collect image frames along the planned flight line, number them in the order of shooting to form an image sequence, detect the brightness difference value of the image frames and adjust the gray scale distribution according to the brightness balance threshold value, and at the same time correct the image edge offset according to the lens distortion coefficient, to generate a brightness balanced image set; S102: Based on the brightness balanced image set, detect the overlapping area between adjacent images, select paired frames with an overlap rate reaching an overlap matching threshold value, extract texture gradient and edge corner information in the overlapping area, calculate the gray difference value of the feature points and perform matching comparison, calculate the parallax distance according to the pixel displacement of the matching points, and generate a parallax displacement matrix; S103: calculating three-dimensional coordinates of the matching points according to the parallax displacement matrix and the camera pose parameters, registering the spatial coordinates, establishing a point adjacency relationship according to the spatial distance, calculating the height difference of adjacent points, arranging the height difference in a connection order to form a height sequence, and generating surface spatial structure information.

[0008] As a further scheme of the present application, the light partition information acquisition step is: S201: detecting a spatial connectivity relationship between adjacent points according to the height difference and the connection order of the adjacent points in the surface spatial structure information, calculating a Euclidean distance of the adjacent points, combining points meeting a condition into a local triangular mesh, connecting and forming a continuous patch set according to the spatial position, and obtaining a triangular mesh node set; S202: calculating a normal vector and an inclination angle value of each patch node in the triangular mesh node set, obtaining a solar azimuth and elevation angle according to shooting time and geographic position data, judging an included angle between a solar incident direction and a patch normal, determining whether the patch is in a sunlight irradiation range in combination with the inclination angle, and generating a patch light parameter set; S203: calculating a shadow position and a self-occlusion boundary according to a change in the inclination direction and the height difference of adjacent patches in the patch light parameter set, marking a boundary line of a light irradiation area, dividing a light continuous area, recording an incident direction and a patch relationship of each area, and generating light partition information.

[0009] As a further scheme of the present application, the light correction reflection information acquisition step is: S301: extracting a spatial position and a boundary range of each light area according to three-dimensional nodes composed of adjacent patch boundary points in the light partition information, converting three-dimensional coordinates of the patch nodes into hyperspectral image pixel coordinates, establishing a spatial correspondence relationship between the light area and the hyperspectral pixel, and generating a light area coordinate correspondence set; S302: separating pixel samples of a sunlit area and a shaded area according to the light area coordinate correspondence set, extracting pixels of the sunlit area and the shaded area, selecting pixels having the same position and corresponding to the same ground point as homonymous pixels, comparing reflectivity differences and marking deviations, performing reflection transition correction along a sunlit area and a shaded area boundary line, and generating a reflection correction list; S303: merging correction results in the reflection correction list, generating a unified reflection curve, extracting a hyperspectral pixel multi-band reflectivity sequence after light correction, recording spectral curve changes, and generating light correction reflection information.

[0010] As a further scheme of the present application, the water absorption characteristic information acquisition step is: S401: Obtain the hyperspectral pixel reflectance sequence after the illumination correction, calculate the reflectance difference of adjacent wavelength points, determine the wavelength drop interval as the identification interval according to the continuously descending section, detect the lowest reflection point of the reflection curve in the identification interval and calibrate the rising points on both sides as shoulder points, determine the depth of the depression point and the shoulder width distance by using the reflectance difference, and generate a reflection change feature group; S402: According to the reflection change feature group, combine the facet slope information and the sun exposure label, pair the depth of the depression point and the shoulder width distance according to the slope relationship, calculate the difference of the depth of the depression point and the difference of the shoulder width ratio, screen the same direction facet set according to the slope difference threshold, and generate a slope correlation feature set; S403: Based on the slope correlation feature set, compare the attenuation distribution of the depression point, aggregate and classify the depth of the depression point and the shoulder width distance according to the distribution rule, establish the corresponding items of the slope, the sun exposure label and the position of the depression point in the same group, extract the corresponding relationship between the attenuation order of the depression point and the slope direction, and obtain the water absorption feature information.

[0011] As a further scheme of the present application, the method further comprises: S5: Call the corresponding items of the depression point position and the slope recorded in the water absorption feature information, obtain the relative elevation distribution of each depression point along the slope by combining the spatial coordinates of the facet, establish the water migration link, distribute the accumulation outflow attribute according to the reflection attenuation order, and form a continuous area water content level label by summarizing the intersection relationship of adjacent links, output the facet-level water content annotation list and rasterized summary, and generate the coal surface water prediction result; The coal surface water prediction result comprises a water content level label, a facet-level water content annotation table and a rasterized water distribution map.

[0012] As a further scheme of the present application, the acquisition step of the coal surface water prediction result is: S501: Call the corresponding items of the depression point position and the slope in the water absorption feature information, calculate the relative elevation of each depression point along the slope direction by combining the spatial coordinates of the facet, detect the continuously changing section according to the elevation distribution, judge the coincidence relationship between the extension direction of the slope line and the projection position of the depression point, determine the continuously changing interval of the elevation along the slope direction, and generate a slope elevation distribution group; S502: According to the slope elevation distribution group, detect the slope vector change rate at the intersection position of each depression point and the adjacent facet slope line, identify the high difference gradient mutation point as a convergence node, connect the depression point, the convergence node and the transition node on the slope line in turn according to the spatial order, establish the water migration link along the slope direction, compare and analyze the spectral reflection change trend of each node in the link according to the reflection attenuation order, determine the accumulation attribute or the outflow attribute of the node, and generate a water migration link set accordingly; S503: Based on the water migration link set, the intersection node relationship of adjacent links is summarized, the connection density and slope angle distribution of adjacent links are calculated, the continuous area is divided according to the connection density threshold, the proportion of accumulated node number in each area is counted and a water content level label is assigned, the water content label of the facet level is sorted and rasterized, and the coal surface water content prediction result is obtained.

[0013] As a further scheme of the present application, the process of calculating the relative height difference of each depression point along the slope direction is specifically: After obtaining the spatial coordinates of the facet, a spatial interpolation algorithm based on Euclidean distance is used to perform weighted calculation on the elevation difference between adjacent facet nodes, the stability of the slope line extension trend is determined through a preset elevation change rate threshold, and the elevation difference gradient sequence is extracted in the continuous change interval of the elevation difference to form the slope elevation difference distribution group; The process of detecting the slope vector change rate is specifically: The normal vectors of adjacent nodes in the slope elevation difference distribution group are analyzed for similarity, and when the similarity is lower than a preset change threshold, a slope mutation point is determined, and the spatial position of the mutation point is marked as the accumulation node.

[0014] As a further scheme of the present application, the process of determining the accumulation or outflow attribute of the link node according to the reflection attenuation order is specifically: According to the combination parameters of the depth difference and shoulder width difference of the depression point in the water absorption characteristic information, the accumulation or outflow attribute of the link node is judged according to a preset reflection attenuation threshold, the node with a reflection attenuation value exceeding the threshold is marked as an accumulation node, and the node with a reflection attenuation value not exceeding the threshold is marked as an outflow node; The process of dividing the continuous area according to the connection density threshold is specifically: The connection number and adjacent slope angle distribution of each accumulation node in the water migration link set are calculated, when the node connection number is greater than a preset connection density threshold and the adjacent slope angle distribution difference is less than a preset angle threshold, the corresponding area is defined as the same water content level area, and the coal surface water content prediction result is generated according to the proportion of accumulation nodes.

[0015] The technical scheme provided by the embodiment of the present application has at least the following beneficial effects: In the present application, by coupling the surface spatial structure of the coal pile with the light distribution, a dynamic correction relationship between the light and the reflectivity is established, the unified expression of the reflection information under different illumination conditions is realized, the spatial registration of the multi-source image and the spectral curve reconstruction are combined, the geometric consistency of the hyperspectral data under the complex surface environment is strengthened, the water migration trend is extracted by using the correlation of the depression point and the slope direction, the continuous water flow link and the accumulation distribution relationship are formed, the reflection characteristics after correction can accurately represent the spectral attenuation law of the water-containing area, the spatial continuity and the numerical stability in the water identification process are ensured, and the water-containing grade division is more coherent, comparable and quantifiable. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A method flowchart of the present application is shown in the figure. Figure 2 A surface spatial structure information acquisition flowchart of the present application is shown in the figure. Figure 3 An illumination partition information acquisition flowchart of the present application is shown in the figure. Figure 4 An illumination correction reflection information acquisition flowchart of the present application is shown in the figure. Figure 5 A water absorption characteristic information acquisition flowchart of the present application is shown in the figure. Figure 6 A coal surface water content prediction result acquisition flowchart of the present application is shown in the figure. DETAILED DESCRIPTION

[0017] The technical solutions in the present application will be described below with reference to the drawings.

[0018] In the embodiments of the present application, the words such as "example", "for example" are used to represent as an example, illustration or explanation. Any embodiment or design scheme described as "example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. In fact, the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present application, the meaning expressed by "and / or" can be both, or can be one of the two.

[0019] In the embodiments of the present application, "image" and "picture" can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent. "Of", "corresponding" and "corresponding" can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent.

[0020] In the embodiments of the present application, sometimes the subscript such as W1 can be written in the form of non-subscript such as W1. When the distinction is not emphasized, the meanings expressed are consistent.

[0021] In order to make the technical problems, technical solutions and advantages to be solved by the present application clearer, specific embodiments will be described in detail below with reference to the drawings.

[0022] Please refer to Figure 1 The present application provides a technical solution: a 3D reconstruction and light self-adaptive coal surface moisture prediction method, comprising the following steps: S1: planning the unmanned aerial vehicle route in the coal pile operation area, setting the flight height, heading and image overlap rate, executing continuous aerial photography by the unmanned aerial vehicle carrying a visible light camera, collecting image frames in sequence according to the route and numbering to form an image sequence, performing brightness equalization and distortion correction on the image sequence, detecting the image overlap area and screening the effective matching frames, extracting the texture and edge feature points of the overlap area to perform feature matching, obtaining the pixel displacement of the matching points in the adjacent images and converting it into disparity information, recovering the three-dimensional space coordinates of the matching points according to the disparity information, completing the space alignment combined with the camera pose, establishing the adjacency relationship according to the spatial distance, reconstructing the surface relief form and boundary connectivity of the coal pile, extracting the elevation difference and connection order of adjacent points, and generating surface spatial structure information; S2: According to the elevation difference and connection order of adjacent points in the surface spatial structure information, the adjacent points are combined into a local triangular mesh according to the spatial connectivity relationship, forming a continuous surface patch set, performing plane fitting on each patch to determine the patch normal and tilt direction, obtaining the illumination direction and height of the sun according to the shooting time and geographical position data, comparing the angle between the sun incident direction and the patch normal and combining the tilt direction to determine the effective projection range of the light, marking the sun and backlight state of each patch, identifying the shadow and self-occlusion boundary according to the change of the tilt direction of adjacent patches and the relative height relationship, dividing the light continuous area and recording the incident direction, occlusion boundary and patch corresponding relationship of each area, and generating light partition information; S3: Based on the three-dimensional nodes composed of adjacent patch boundary points in the light partition information, the spatial position and boundary range of each light area are extracted, the coordinates of the patch nodes are converted to the hyperspectral image pixel coordinate system, the spatial correspondence between the light area and the hyperspectral pixel is established and geometric registration is performed, the pixel samples of the sun and backlight areas are separated, and the sun and backlight segments are formed, respectively. Select the pixels with the same position and corresponding to the same ground point as homonymous pixels, compare the reflection differences and mark the deviations, perform reflection transition correction along the sun and backlight boundary, merge the correction results to form a unified reflection curve list, extract the multi-band reflectance sequence of the hyperspectral pixels after light correction to form a spectral curve record, and generate light correction reflection information; S4: According to the spectral curve of the hyperspectral pixel after illumination correction in the illumination correction reflection information, the reflectance curve formed by the change of reflectance with wavelength is read, the reflectance change trend between adjacent wavelength points is obtained, and the continuously decreasing wavelength interval is determined as the identification interval. The positions of the dip point and the shoulder on both sides of the reflectance curve in the identification interval are detected, the dip point depth and shoulder width characteristics are determined, and the reflection attenuation sequence is constructed. The facet slope direction is associated with the sun exposure label. The dip point attenuation distribution is compared, and the same type of area is aggregated according to the combination characteristics of the dip point depth and the shoulder spacing. The corresponding items of the dip point position, the shoulder spacing, the slope direction and the sun exposure label are recorded, and the water absorption characteristic information is generated; S5: The corresponding items of the dip point position and the slope direction recorded in the water absorption characteristic information are called, and the relative elevation distribution of each dip point along the slope direction is obtained combined with the spatial coordinates of the facet. The water migration link is established along the slope direction, and the convergence node is marked at the intersection position of the dip point and the adjacent facet slope line. The link nodes are assigned with accumulation and outflow attributes according to the reflection attenuation sequence. The intersection relationship of adjacent links is summarized to form the water content level label of the continuous area. The facet-level water content annotation list and rasterized summary are output, and the coal surface water content prediction result is generated.

[0023] The surface spatial structure information includes a set of spatial point coordinates, an adjacency topological relationship, an elevation difference parameter and a boundary connectivity attribute. The illumination partition information includes illumination area range, incident angle parameter, shielding boundary feature and facet index information. The illumination correction reflection information includes pixel coordinate mapping table, reflectance correction curve, multi-band reflectance sequence and illumination area corresponding relationship. The water absorption characteristic information includes dip point spectral position, shoulder spacing feature, slope direction associated parameter and sun exposure classification label. The coal surface water content prediction result includes water content level label, facet-level water content annotation table and rasterized water content distribution map.

[0024] Please refer to Figure 2 The acquisition steps of the surface spatial structure information are as follows: S101: Obtain the flight line parameters and geographical coordinates of the coal pile operation area, set the flight height, flight direction and image overlap rate, use the unmanned aerial vehicle to carry the visible light camera to continuously collect image frames along the planned flight line, number the image frames in shooting order to form an image sequence, detect the brightness difference value of the image frames and adjust the gray scale distribution according to the brightness balance threshold value, and correct the image edge offset according to the lens distortion coefficient to generate a set of brightness balanced images; The flight path parameters and geographic coordinates of the coal pile operation area are obtained. The GPS coordinates of the four corner points of the coal pile operation area are read to determine the rectangular boundary range. For example, the boundary coordinates are set to (116.40, 39.90) to (116.41, 39.91). Within this range, a parallel reciprocating flight path is planned. The drone's flight altitude is set to a vertical distance of 120 meters from the highest point of the coal pile, the heading angle is set to 15 degrees east of due north, the image lateral overlap rate is set to 80%, and the heading overlap rate is set to 75%. The drone, equipped with a visible light camera, flies along the set flight path, triggering the shutter every 2 seconds. The acquired images are digitally encoded from 0001 to 1000 according to the order of acquisition time. The pixel matrix of each frame is traversed, and the arithmetic mean of the gray values ​​of all pixels in a single frame is calculated as the average brightness value of that frame (90). The brightness standard value is set to 128, and the multiplication gain coefficient is calculated. This gain coefficient The red, green, and blue channels of all pixels in the frame are applied respectively (e.g. For the pixel positions at the four corners of the image, the camera radial distortion parameters obtained in advance are called. and Read the raw pixel coordinates And calculate its radius to the image center. Through calculation The horizontal and vertical coordinates of edge pixels are shifted and corrected to reconstruct the pixel arrangement and generate a brightness-balanced image set.

[0025] S102: Based on the brightness equalization image set, detect the overlapping area between adjacent images, filter the paired frames whose overlap rate reaches the overlap matching threshold, extract the texture gradient and edge corner information in the overlapping area, calculate the gray value difference of feature points and perform matching comparison, calculate the disparity distance based on the pixel displacement of the matching points, and generate the disparity displacement matrix. Based on a brightness-equalized image set, read two adjacent frames from the image set. and The system retrieves the coordinates of the camera center at the moment of shooting and calculates the straight-line distance between the centers of the two cameras. If this distance is less than a set baseline length of 20 meters, a potential overlap is identified. The proportion of pixels with overlapping geographic coordinates in the two images is counted relative to the total number of pixels in a single frame. If this proportion is greater than a set overlap matching threshold of 0.65, the two images are marked as paired frames. Within the overlapping area of ​​the paired frames, a... The Sobel operator is invoked on the center pixel of the sliding window to calculate its horizontal gradient. with vertical gradient Then through Calculate gradient magnitude If the gradient magnitude If the pixel is greater than the edge determination reference value 50 and continuously extends more than 10 pixels in a certain direction, the pixel is marked as an edge feature point, and the feature points around the feature point are further extracted The pixel gray scale matrix in the neighborhood is calculated, the gray scale difference square sum of the matrix and the candidate window matrix in the corresponding search area in the matching frame is calculated, the minimum value of the difference square sum is selected as the best matching point, and the horizontal coordinate of the feature point in the reference image is read The horizontal coordinate of the corresponding point in the matching image is read A subtraction operation is performed The pixel displacement value 50 pixels is obtained, the displacement value is filled into the matrix unit corresponding to the image pixel coordinate, and the disparity displacement matrix is generated.

[0026] S103: According to the disparity displacement matrix and the camera pose parameter, the three-dimensional coordinates of the matching points are calculated, the spatial coordinates are registered, the adjacent relationship between points is established according to the spatial distance, the height difference between adjacent points is calculated and arranged in the connection order to form a height sequence, and the surface space structure information is generated; According to the disparity displacement matrix and the camera pose parameter, the three-dimensional coordinates of the matching points are calculated, the focal length of the camera internal parameter is called (The calibration file from S101), and the adjacent shooting point position read from the high-precision inertial navigation system (INS) (integrating GPS and IMU) log carried by the unmanned aerial vehicle is calculated The disparity value of the corresponding feature point in the disparity displacement matrix is read Pixels, assuming that the pixel size is The disparity is converted into physical size The depth distance of the feature point relative to the camera optical center is calculated by using the formula of the triangulation principle The depth distance and the image plane coordinate are combined to inversely calculate the and coordinates of the feature point in the camera coordinate system, the attitude angles of the camera when shooting are read from the INS log, including the pitch angle-90 degrees, the roll angle 0 degrees, and the yaw angle 15 degrees, the rotation matrix and the translation vector are constructed The camera coordinate system coordinates are multiplied by the rotation matrix and added to the translation vector to be converted to the world coordinate system of the coal yard site The Euclidean distance between any two points is calculated, if the spatial distance between point and point is less than the neighborhood determination radius 0.5 meters, the adjacent index of ​​vertical height coordinates of the points with the points vertical height coordinates of the points , a height difference of 0.2 meters is obtained by performing subtraction, and the adjacent points and their height difference data are sequentially written into a linked list structure according to the scanning order along the positive direction of the X axis in the spatial position, to generate surface spatial structure information.

[0027] Please refer to Figure 3 The light partition information acquisition step is: S201: According to the elevation difference between adjacent points and the connection order in the surface spatial structure information, the spatial connectivity relationship between adjacent points is detected, the Euclidean distance between adjacent points is calculated, the points meeting the conditions are combined into a local triangular mesh, and the continuous patch set is formed according to the spatial position connection, to obtain a triangular mesh node set; According to the elevation difference between adjacent points and the connection order in the surface spatial structure information, the point cloud data obtained by S103 is read, and any point coordinates are extracted coordinates and coordinates The Euclidean distance between and is calculated , the operation result is meters, and the distance between and is meters, and the distance between and is meters, set a grid length threshold, which is set according to the average point distance of the point cloud 0.4 meters, specifically set to 1.5 times the average point distance, that is m, compare the three calculated distances , , with the threshold value of 0.6 meters, because 0.3, 0.245, and 0.332 are less than 0.6, so it is determined that the three points , , meet the combination condition, and the three points are combined into a local triangular mesh patch, and the vertex index is recorded, all points are traversed and the process is repeated, the triangular patches sharing the edges are connected in the topological structure, forming a continuous patch set covering the entire coal pile surface, to obtain a triangular mesh node set.

[0028] S202: Calculate the normal vector and tilt angle of the face using the coordinates of each face node in the triangular mesh node set, and obtain the azimuth and elevation angle of the sun based on the shooting time and geographical location data. Determine the angle between the sun's incident direction and the face normal, and combine the tilt angle to determine whether the face is within the sunlight illumination range, and generate the face illumination parameter set. By using the coordinates of each facet node in the triangular mesh node set, the facet in S201 can be read. The coordinates of the three vertices , , ; Calculate vectors ; with vector ; implement and vector cross product operation ; Obtain the normal vector Since the Z component is negative; Reverse it to ; Calculate the normal vector and the perpendicular Z-axis vector. dot product, ,calculate Length of the module Through inverse cosine operation The tilt angle value is obtained The time and location parameters are retrieved from the shooting time "year, month, day" and the geographical location "116.40 degrees east longitude, 39.90 degrees north latitude". The time and location parameters are input into the solar position algorithm (SPA) to calculate the solar azimuth angle as 150 degrees and the solar altitude angle as 40 degrees. Convert the solar angle into a normalized solar incidence direction vector. Calculate the solar incident vector With the normal vector of the surface dot product Through inverse cosine operation The angle between the two is obtained as degrees, the included angle The degree is compared with the 90-degree light irradiance reference value, because Determine the surface area When the surface is in backlight and not in sunlight, index the surface. Normal vector Inclination angle degrees, angle of incidence The brightness and backlight status labels "Back-lit" are combined and stored in a list to generate a set of surface lighting parameters.

[0029] S203: Based on the changes in tilt direction and elevation difference of adjacent patches in the patch illumination parameter set, calculate the shadow position and self-occlusion boundary, mark the boundary line of the illumination area, divide the continuous illumination area, record the incident direction and patch relationship of each area, and generate illumination partition information. Based on the changes in tilt direction and elevation difference between adjacent patches in the patch illumination parameter set, read... The status is "Back-lit" and its average elevation. Meters, then read their adjacent patches. The status is "Back-lit". The state is "Illuminated" (incident angle) Spend), The state is "Illuminated" (incident angle) (degree), traversing the facets in the "Illuminated" state. Obtain the coordinates of its center point. Retrieve the solar incident vector of S202 , build a self Point in the opposite direction of the sun Emitted rays (in ), regarding this ray Compared with other faces in the coal pile triangular mesh (e.g.) Performing the Möller–Trumbore ray-triangle intersection test, calculations revealed that the ray... Place and surface (Elevation) Intersection of meters, determine In In the cast shadow, update its state from "Illuminated" to "Cast-Shadow" on the opposite side. Perform the same detection; if no intersection is found, its state remains "Illuminated." Then, iterate through all faces, searching for adjacent face pairs with different states, for example... (Back-lit) and (Illuminated) shared edges, and (Cast-Shadow) and (Illuminated) Shared edges, marking these edges as the boundary lines of the illuminated region, from any facet Initially, the seed-filling algorithm is executed to find all topologically connected faces with the same "Illuminated" state, classifying them as "Illuminated Continuous Region 1". Similarly, [the algorithm is applied to other regions]. , Classified as "Illumination Continuous Region 2 (Backlight)", record the incident direction vector of "Illumination Continuous Region 1" , and list the face ID list contained therein , , generate illumination partition information.

[0030] Please refer to Figure 4 , the acquisition step of illumination correction reflection information is: S301: According to the three-dimensional nodes composed of adjacent face boundary points in the illumination partition information, extract the spatial position and boundary range of each illumination region, convert the three-dimensional coordinates of the face nodes to hyperspectral image pixel coordinates, establish the spatial correspondence between the illumination region and the hyperspectral pixel, and generate the illumination region coordinate correspondence set; According to the three-dimensional nodes composed of adjacent face boundary points in the illumination partition information, read the "Illumination Continuous Region 1" (sunlight) and its boundary node coordinate list of S203, extract the three-dimensional world coordinates of the first node , and the coordinates of the second node , and call the hyperspectral image metadata obtained by the coaxial hyperspectral imager and already completed [Co-registration] with [INS / GPS data] and S103 visible light model, obtain the world coordinates corresponding to the upper left corner of the pixel center , and its ground sampling distance meters / pixel, perform coordinate conversion calculation, and project the coordinates of to the pixel coordinate system; Calculate its column coordinates ; Calculate its row coordinates , get the pixel coordinates corresponding to ; Similarly, calculate the pixel coordinates of : , , get the pixel coordinates corresponding to , traverse all boundary nodes of "Illumination Continuous Region 1", convert all three-dimensional coordinates to two-dimensional pixel coordinates , form a closed pixel polygon, mark this polygon and all pixel coordinates contained in its interior as "sunlight region pixel set 1", read "Illumination Continuous Region 2" (backlight) and its boundary node coordinates, for example , use the same origin and​​ performing calculation, , , storing the boundary list of "backlight region pixel set 2", repeating the process until all the light partitions (sunlight, backlight, shadow) in S203 complete the pixel coordinate conversion, establishing the index of light partition label and hyperspectral pixel coordinate set, and generating the light region coordinate correspondence set.

[0031] S302: According to the light region coordinate correspondence set, separate the pixel samples of the sunlight and backlight regions, extract the pixels of the sunlight region and the backlight region, select the pixels with the same position and corresponding to the same ground point as homonymous pixels, compare the reflectance difference and mark the deviation, correct the reflection transition along the sunlight and backlight boundary line, and generate a reflection correction list; According to the light region coordinate correspondence set, read the "sunlight region pixel set" and "backlight region pixel set" of S301, and integrate the "shadow region pixel set" into the "backlight region pixel set" for unified processing, traverse each pixel in the "backlight region pixel set" , check whether its eight-neighborhood pixels are contained in the "sunlight region pixel set", if yes, mark it as a "boundary backlight pixel", for example, pixel is a "boundary backlight pixel", its adjacent is a "sunlight region pixel", select as the homonymous reference pixel of , read the reflectance of in band 1 (450nm) , read the reflectance of in the same band ; ; Calculate the reflectance ratio of the two , set a homogeneous ground reflectance ratio reference interval, which is based on the inherent texture fluctuation of coal under the same light, for example , compare the calculated ratio with the interval, not in the interval, determine that the difference is caused by light, and mark the deviation, perform the same operation on band 2 (455nm) , , , calculate the ratio , traverse all bands (for example, 200 bands) to obtain a ratio vector , traverse all "boundary backlight pixels" and their homonymous reference pixels to obtain a group of boundary ratio vectors, calculate the average vector of all boundary ratio vectors , for example ​ This average ratio vector As a global correction factor, it iterates through all pixels in the "backlight area pixel set" (including boundaries and interiors), such as interior pixels. Read its original reflectance ; Application of correction factor Perform this multiplicative correction on all its bands to obtain the corrected complete spectral vector. Save it to a list to generate a reflection correction list.

[0032] S303: Merge the correction results in the reflection correction list, generate a unified reflection curve, extract the hyperspectral pixel multi-band reflectance sequence after illumination correction, record the changes in the spectral curve, and generate illumination-corrected reflection information. Merge the correction results from the reflection correction list, retrieve the reflection correction list from S302, which contains the corrected spectra of all pixels in the original "backlight area pixel set", such as pixel... Corrected spectral vector Simultaneously retrieve the original hyperspectral image data cube And the "sunlight area pixel set" generated by S301, creating a [something] with [something]. Having the same three-dimensional dimensions (e.g.) Empty data cube (band) traversal Each pixel coordinate For example from Begin, make a judgment Does it exist in the "sunlight area pixel set"? Let's assume the query result is yes. Then from Read from The original spectral vector of the location , this vector Write directly to of Location, i.e. Processing pixels ,judge Does it exist in the "sunlight area pixel set"? If the query result is no (i.e., it is a backlight or shadow pixel), then query the reflection correction list in S302. The corresponding corrected spectral vector Assuming Write this correction vector to of Location, i.e. Read and discard The original value is used to repeat this conditional check and data write operation until... All pixel positions All the cubes are filled, at this time All the spectral curves in the cube are unified to the reflectance level under the sun illumination, extract The multi-band reflectance sequence of any pixel (for example ) in the cube, i.e. the vector , and record the sequence as a spectral curve, finally The data cube itself is the illumination-corrected reflectance information.

[0033] Please refer to Figure 5 , the acquisition step of the water absorption characteristic information is: S401: Acquire the illumination-corrected hyperspectral pixel reflectance sequence, calculate the reflectance difference between adjacent wavelength points, determine the wavelength descending interval as the identification interval according to the section of continuous decline, detect the lowest reflection point of the reflection curve in the identification interval and mark the rising points on both sides as shoulder points, determine the depth of the depression point and the shoulder width distance by using the reflectance difference, and generate a reflection change characteristic group; Acquire the illumination-corrected hyperspectral pixel reflectance sequence, read the spectral vector of the pixel in the data cube of S303 , the vector contains the reflectance of 200 bands , for example, starting from the 100th band (wavelength ), , , , , , ; Calculate the reflectance difference between adjacent wavelength points , obtain , , , , , set the minimum wavelength number reference of continuous decline as 2, detect the continuous decline of reflectance from to ( and are both negative values), determine the wavelength range containing this section as the identification interval, search for the minimum value of reflectance in this interval, determine as the lowest reflection point, i.e. the depression point, search for the position from positive to negative to the left; Determine as the left shoulder point, and the reflectance is ; Search for the position from negative to positive to the right; Sure The point on the right shoulder has a reflectivity of Calculate the average reflectivity of the shoulder ; Calculate the depth of the depression Calculate the shoulder width distance , pixel coordinates , depression location Depth of depression ,shoulder width Store it as an entry in the list to generate a group of reflection change features.

[0034] S402: Based on the reflection change feature group, combined with the slope aspect information and sunlight label, the depression depth and shoulder width distance are paired according to the slope aspect relationship, the depression depth difference and shoulder width ratio difference are calculated, and the same-direction face set is selected according to the slope aspect difference threshold to generate the slope aspect associated feature set. Based on the reflection change feature set, read the pixels in S401. Entries ( , ), query the coordinate correspondence established in S301 to determine that the pixel corresponds to the patch in S201. Retrieve from S202 The surface illumination parameter set is used to obtain its slope aspect. (Due South), Sunlight Label "Illuminated", then read the pixels. Entries ( , ), determine its corresponding facet ; Get Spend, "Illuminated" sets the aspect difference threshold to 15 degrees. This threshold is based on the natural variation of the aspect of the coal pile surface, allowing for a certain range of fluctuations, and calculates the absolute difference in aspect between the two sides. The difference of 5 degrees is compared with the threshold of 15 degrees. ,determination and Belongs to the same-direction surface, calculation and Depth difference of depressions ,calculate and The difference in shoulder width ratio, where the ratio difference is defined as... , operation Iterate through all the faces and select those with a slope difference of less than 15 degrees (e.g., ) and its corresponding dimple feature (depth, width) are put into the same set, and the slope direction associated feature set is generated.

[0035] S403: Based on the slope direction associated feature set, the dimple decay distribution is compared, and the dimple depth and shoulder width spacing are aggregated and classified according to the distribution rule. In the same type of grouping, the corresponding entries of slope direction, sun exposure label and dimple position are established, the corresponding relationship between dimple decay order and slope direction is extracted, and the water absorption feature information is obtained. Based on the slope direction associated feature set, read the same direction facet set with a slope of 180 degrees ± 15 degrees in S402, including , ) and , ), and other facets such as , ), set the classification criteria for dimple depth, which is based on the median of all dimple depths in the set Set, divide into “deep dimples” if the depth is greater than , and divide into “shallow dimples” if it is less than or equal to , set the classification criteria for shoulder width spacing, which is based on the median of all shoulder widths in the set Set, divide into “wide shoulder” if the width is greater than , and divide into “narrow shoulder” if it is less than or equal to , classify (shallow dimples), (narrow shoulder), the combined feature is “shallow-narrow”, classify (deep dimples), (wide shoulder), the combined feature is “deep-wide”, classify (shallow dimples), (narrow shoulder), the combined feature is “shallow-narrow”, aggregate and with the same combined feature “shallow-narrow” into a class, aggregate with “deep-wide” into another class, in the “shallow-narrow” class, establish the corresponding entries: (facet , slope , sun exposure label “Illuminated”, dimple position ), (facet , slope , sun exposure label “Illuminated”, dimple position ​​​​​​), according to the physical meaning (usually the deep and wide absorption feature corresponds to higher moisture), set "deep-wide" as "strong attenuation" (order 1), "shallow-narrow" as "weak attenuation" (order 2), extract the attenuation order and the corresponding relationship of the slope direction , and obtain the moisture absorption feature information.

[0036] Please refer to Figure 6 , the acquisition step of the coal surface moisture prediction result is: S501: Call the corresponding items of the depression point position and the slope in the moisture absorption feature information, combine the face sheet spatial coordinates, calculate the relative height difference of each depression point along the slope direction, detect the continuous change section according to the height difference distribution, judge the coincidence relationship of the extension direction of the slope line and the projection position of the depression point, determine the continuous interval of the height difference along the slope direction, and generate a slope height difference distribution group; Call the corresponding items of the depression point position and the slope in the moisture absorption feature information, read the items of S403, for example, the face sheet (slope , depression point ) and the face sheet (slope , depression point ), and call the face sheet center point spatial coordinates and center point coordinates generated in S103, for example, construct an extension line along the slope (north) direction of , calculate the relative height difference of the point to any point (e.g. ) on its extension line , the here is obtained by performing inverse distance weighting (IDW) interpolation on the nearest 3 vertices in the S103 point cloud around the position , for example, the interpolation calculation result is meters, then meters, traverse the sampling points with an equal interval of 1 meter along the direction, calculate the relative height difference sequence of each sampling point , set a height difference change threshold of 0.03 meters / meter, which is based on the minimum effective slope of the coal pile surface moisture flow, when the relative height difference change rate of adjacent sampling points is less than the threshold, it is determined that the height difference change is continuous, for example is greater than , it is determined as discontinuous, and is less than , it is determined as continuous, based on this continuous change determination, determine the continuous height difference interval from the point along the The section extending 1 meter in the direction is a continuous elevation difference section. center point of the dough On the slope extension line The projection position on, and depression Spatial overlap determination is performed. Since the depression feature coincides with the geometric center point of the patch in spatial position, it is determined to be overlapped. Point 1 is used as the starting point of the slope line, and its continuous elevation difference interval is recorded, i.e., along... Direction Elevation, continuing until A 1-meter interval at an elevation of meters will The surface patch is combined with its continuous elevation difference intervals to generate an elevation difference distribution group.

[0037] S502: Based on the slope aspect elevation difference distribution group, detect the slope aspect vector change rate at the intersection of each depression point and the slope aspect line of the adjacent area, identify the elevation gradient abrupt change point as the convergence node, connect the depression point, convergence node and transition node on the slope aspect line in spatial order, establish a water migration link along the slope aspect direction, compare and analyze the spectral reflectance change trend of each node in the link according to the reflectance attenuation order, determine the accumulation attribute or outflow attribute of the node, and generate a water migration link set accordingly. Based on the slope aspect elevation difference distribution group, read the surface patch in S501. and its adjacent face The slope aspect elevation difference distribution data, and retrieved and Intersecting common edge ,exist geometric center point ,calculate slope vector (Point to )and slope vector (Point to The angle between them: The rate of change of the aspect vector is defined as A threshold of 0.15 was set for elevation gradient abrupt change. This threshold was used to screen points where water flow direction changed significantly. A threshold higher than this value would be considered a significant change. It was determined to be a gradient mutation, because ,determination Not a point of abrupt change in elevation gradient, not a convergence node, check. and Intersecting common edge center point Assuming its slope vector change rate is calculated as ,because ,Will Mark as a convergence node, and the patch Treat it as a migration segment, Starting from the slope, along the direction Direction, towards Extend the establishment of water migration links According to S403 The reflection attenuation order is "weak attenuation" (order 2), and the "outflow" determination criterion is set as the "weak attenuation" feature defined in S403 (e.g., depression depth). ),because Depth of depression If this "outflow" criterion is met, then the judgment is made. The attribute is "outflow", which will aggregate the node. The unified label is "accumulation". As the endpoint, record Link, and The point is assigned the "outflow" attribute. Points are assigned the "accumulation" attribute to generate a set of water migration links. S503: Based on the moisture migration link set, the relationship between the intersection nodes of adjacent links is summarized, the connection density and aspect angle distribution of adjacent links are calculated, continuous areas are divided according to the connection density threshold, the proportion of the number of clustered nodes in each area is counted and assigned a moisture level label, the surface moisture label is sorted out and rasterized to obtain the prediction result of coal surface moisture. Based on the water migration link set, read the data generated by S502. (Outflow to accumulation) and adjacent links (Outflow to accumulation) and (Accumulation to accumulation), summarizing and converging nodes , calculation Given a circle with a radius of 1 meter and the center of the circle, calculate the number of intersections of all links within that circle. Assuming there are 3 intersecting links, calculate the intersection of the links and the number of intersections. The total length of the common side of the points that intersect is 3m; Calculate the connection density The connection density threshold is set to 0.8. This threshold is used to distinguish areas where water flow is highly concentrated. and Comparison, because ,determination The area in question is a highly continuous region; therefore, this highly continuous region is named... ,statistics The attributes of all link nodes within, such as There are a total of 20 nodes, including 15 accumulation nodes and 5 outflow nodes; Calculating the proportion of the number of accumulated nodes , set the proportion threshold of the water content level label: “low water content”, “medium water content”, “high water content”, compare with the threshold interval, determine the water content level label of the region is “high water content”, set the water content label list of all patches (for example ) in “high water content”, convert the patch-level label list to a raster image with a resolution of 0.1 meters / pixel, set the pixel value of the “high water content” region to 3, “medium water content” to 2, and “low water content” to 1, to generate the coal surface moisture prediction result.

[0038] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A coal surface moisture prediction method with 3D reconstruction and light adaptation, characterized in that, The method comprises the following steps: S1: obtaining image frames of the coal pile area and numbering to form an image sequence, detecting matching feature points in the overlapping area, calculating parallax information to recover three-dimensional coordinates, completing spatial alignment and reconstructing the surface morphology of the coal pile, extracting the elevation difference and connection order of adjacent points, and generating surface spatial structure information; S2: establishing a local triangular mesh to form a patch set according to the elevation difference and connection order of adjacent points in the surface spatial structure information, fitting the patch normal to determine the tilt direction, combining with the solar azimuth to determine the sun and backlight, identifying the shadow and occlusion boundary, and generating illumination partition information; S3: based on the three-dimensional nodes composed of adjacent patch boundary points in the illumination partition information, extracting the spatial position and boundary of the illumination area and establishing the spatial correspondence with the hyperspectral pixels, separating the sun and backlight samples, comparing the reflection differences of the same pixels and correcting the transition, and generating illumination corrected reflection information; S4: according to the corrected spectral curve in the illumination corrected reflection information, analyzing the reflectivity change trend, determining the continuous descending interval and detecting the depth of the depression point and the shoulder width, establishing the reflection attenuation order and associating the slope direction and the sun label, and generating moisture absorption characteristic information by clustering similar regions.

2. The coal surface moisture prediction method with 3D reconstruction and light adaptation according to claim 1, characterized in that: The surface spatial structure information includes a set of spatial point coordinates, adjacency topological relationship, elevation difference parameters and boundary connectivity attributes, the illumination partition information includes illumination area range, incident angle parameters, occlusion boundary features and patch index information, the illumination corrected reflection information includes pixel coordinate mapping table, reflectivity correction curve, multi-band reflectivity sequence and illumination area correspondence, and the moisture absorption characteristic information includes depression point spectral position, shoulder distance feature, slope direction association parameter and sun classification label.

3. The coal surface moisture prediction method with 3D reconstruction and light adaptation according to claim 1, characterized in that, The surface spatial structure information acquisition step is: S101: obtaining the flight line parameters and geographic coordinates of the coal pile operation area, setting the flight height, flight direction and image overlap rate, using a UAV to carry a visible light camera to continuously collect image frames along the planned flight line, numbering to form an image sequence according to the shooting order, detecting the brightness difference of the image frames and adjusting the gray scale distribution according to the brightness balance threshold, and correcting the image edge offset according to the lens distortion coefficient, to generate a set of brightness balanced images; S102: based on the set of brightness balanced images, detecting the overlapping area between adjacent images, screening paired frames with an overlap rate reaching an overlap matching threshold, extracting texture gradient and edge corner information in the overlapping area, calculating feature point gray difference and performing matching comparison, calculating parallax distance according to the pixel displacement of matching points, and generating a parallax displacement matrix; S103: calculating the three-dimensional coordinates of the matching points according to the parallax displacement matrix and camera pose parameters, registering the spatial coordinates and establishing the adjacency relationship between points according to the spatial distance, calculating the elevation difference of adjacent points and arranging them in connection order to form an elevation sequence, and generating surface spatial structure information.

4. The coal surface moisture prediction method with 3D reconstruction and light adaptation according to claim 1, characterized in that, The illumination partition information acquisition step is: S201: detecting a spatial connection relationship between adjacent points according to a height difference and a connection order of the adjacent points in the surface spatial structure information, calculating a Euclidean distance of the adjacent points, combining points meeting a condition into a local triangular grid, connecting and forming a continuous patch set according to spatial positions, and obtaining a triangular mesh node set; S202: calculating a normal vector and an inclination angle value of a patch according to a coordinate of each patch node in the triangular mesh node set, obtaining a solar azimuth and an altitude angle according to shooting time and geographical position data, judging an included angle between a solar incident direction and a patch normal, judging whether the patch is in a sunlight irradiation range according to the inclination angle, and generating a patch illumination parameter set; S203: calculating a shadow position and a self-shading boundary according to a change in an inclination direction and a height difference of adjacent patches in the patch illumination parameter set, marking a boundary line of an illumination area, dividing a continuous illumination area, recording an incident direction and a patch relationship of each area, and generating illumination partition information.

5. The coal surface moisture prediction method with 3D reconstruction and light adaptation according to claim 1, characterized in that, The acquisition step of the illumination correction reflection information is: S301: extracting a spatial position and a boundary range of each illumination area according to three-dimensional nodes composed of adjacent patch boundary points in the illumination partition information, converting three-dimensional coordinates of patch nodes into hyperspectral image pixel coordinates, establishing a spatial correspondence relationship between an illumination area and a hyperspectral pixel, and generating an illumination area coordinate correspondence set; S302: separating pixel samples of a sunlit area and a backlit area according to the illumination area coordinate correspondence set, extracting pixels of the sunlit area and the backlit area, selecting pixels with the same position and corresponding to the same ground point as homonymous pixels, comparing reflectivity differences and marking deviations, performing reflection transition correction along a sunlit area and a backlit area boundary, and generating a reflection correction list; S303: merging correction results in the reflection correction list, generating a unified reflection curve, extracting a hyperspectral pixel multi-band reflectivity sequence after illumination correction, recording spectral curve changes, and generating illumination correction reflection information.

6. The coal surface moisture prediction method with 3D reconstruction and light adaptation according to claim 1, characterized in that, The acquisition step of the water absorption characteristic information is: S401: acquiring the hyperspectral pixel reflectivity sequence after illumination correction, calculating a reflectivity difference of adjacent wavelength points, determining a wavelength drop interval as an identification interval according to a continuously descending section, detecting a lowest reflection point of a reflection curve in the identification interval and marking two side rising points as shoulder points, determining a depression depth and a shoulder width distance by using a reflectivity difference, and generating a reflection change characteristic group; S402: according to the reflection change characteristic group, combining patch slope direction information and a sunlit label, pairing the depression depth and the shoulder width distance according to a slope direction relationship, calculating a depression depth difference and a shoulder width proportion difference, screening a same-direction patch set according to a slope direction difference threshold, and generating a slope direction correlation characteristic set; S403: based on the slope direction correlation characteristic set, comparing a depression attenuation distribution, classifying the depression depth and the shoulder width distance according to a distribution rule, establishing a corresponding item of a slope direction, a sunlit label and a depression position in a same-class grouping, extracting a corresponding relationship between a depression attenuation order and a slope direction, and obtaining water absorption characteristic information.

7. The coal surface moisture prediction method with 3D reconstruction and light adaptation according to claim 1, characterized in that, The method further comprises: S5: calling the corresponding entries of the depression position and aspect in the water absorption characteristic information, obtaining the relative height difference distribution of each depression along the aspect by combining the spatial coordinates of the facets, establishing the water migration link, assigning the accumulated outflow attribute according to the reflection attenuation order, summarizing the adjacent link intersection relationship to form a continuous area water content level label, outputting the facet-level water content annotation list and rasterized summary, and generating the coal surface water content prediction result; The coal surface water content prediction result includes a water content level label, a facet-level water content annotation table and a rasterized water content distribution map.

8. The coal surface moisture prediction method with 3D reconstruction and light adaptation according to claim 7, characterized in that, The acquisition steps of the coal surface water content prediction result are: S501: calling the corresponding entries of the depression position and aspect in the water absorption characteristic information, combining the spatial coordinates of the facets, calculating the relative height difference of each depression along the aspect direction, detecting the continuous change section according to the height difference distribution, judging the coincidence relationship between the extension direction of the aspect line and the projection position of the depression, determining the continuous interval of the height difference along the aspect direction, and generating the aspect height difference distribution group; S502: according to the aspect height difference distribution group, detecting the aspect vector change rate at the intersection position of each depression and the adjacent facet aspect line, identifying the height difference gradient mutation point as a convergence node, connecting the depressions, convergence nodes and transition nodes on the aspect line in spatial order in turn, establishing the water migration link along the aspect direction, comparing and analyzing the spectral reflection change trend of each node in the link according to the reflection attenuation order, determining the accumulation attribute or outflow attribute of the node, and generating the water migration link set accordingly; S503: based on the water migration link set, summarizing the intersection node relationship of adjacent links, calculating the connection density and aspect angle distribution of adjacent links, dividing the continuous area according to the connection density threshold, counting the proportion of accumulated node number in each area and assigning the water content level label, arranging the facet-level water content annotation and performing rasterized summary, and obtaining the coal surface water content prediction result.

9. The coal surface moisture prediction method with 3D reconstruction and light adaptation according to claim 8, characterized in that, The process of calculating the relative height difference of each depression along the aspect direction is specifically: After obtaining the spatial coordinates of the facets, a spatial interpolation algorithm based on Euclidean distance is used to calculate the height difference between adjacent facet nodes by weighting, the stability of the aspect line extension trend is determined by a preset height change rate threshold, and the height difference gradient sequence is extracted in the continuous change interval of the height difference to form the aspect height difference distribution group; The process of detecting the aspect vector change rate is specifically: The normal vectors of adjacent nodes in the aspect height difference distribution group are analyzed for similarity, and when the similarity is lower than a preset change threshold, the aspect mutation point is determined, and the spatial position of the mutation point is identified as the convergence node.

10. The coal surface moisture prediction method with 3D reconstruction and light adaptation according to claim 8, characterized in that, The process of determining the accumulation or outflow attribute of the link node according to the reflection attenuation order is specifically: According to the combination parameters of the depression depth difference and shoulder width proportion difference in the water absorption characteristic information, the accumulation or outflow attribute of the link node is judged according to a preset reflection attenuation threshold, the nodes with reflection attenuation values exceeding the threshold are marked as accumulated nodes, and the nodes with reflection attenuation values not exceeding the threshold are marked as outflow nodes; The process of dividing the continuous area according to the connection density threshold is specifically: The connection number of each convergence node and the adjacent slope aspect distribution are calculated in the water migration link set, when the node connection number is greater than the preset connection density threshold and the adjacent slope aspect distribution difference is less than the preset angle threshold, the corresponding area is defined as the same water content grade area, and the coal surface water content prediction result is generated according to the convergence node proportion.

Images