Unmanned aerial vehicle based regional geological survey method, system, device and medium

Abstract

The application relates to the technical field of geological exploration. A method, a system, equipment and a medium for regional geological survey based on a UAV are provided, wherein the method comprises the following steps: performing geological entity identification on an intrinsic reflectivity image by using a pre-trained interpretation model to obtain an initial geological interpretation result; performing confidence evaluation based on the initial geological interpretation result to generate a confidence evaluation result; triggering data reacquisition according to the confidence evaluation result; performing observation condition interference separation on a reacquired image sequence based on reacquired flight metadata and a digital surface model to generate optimized data; performing model fine-tuning on the pre-trained interpretation model based on the optimized data to generate an updated model; and reinterpreting the optimized data by using the updated model and outputting a geological map to achieve the technical effects of improving image radiation correction accuracy, enhancing the generalization ability of a geological interpretation model, optimizing low-confidence area identification efficiency and realizing the collaborative optimization of data acquisition and model updating.

Description

Technical Field

[0001] This invention relates to the field of geological exploration technology, and in particular to methods, systems, equipment and media for regional geological surveys based on unmanned aerial vehicles (UAVs). Background Technology

[0002] With the rapid development of UAV remote sensing technology, it has shown great potential in the field of regional geological surveys. Traditional geological surveys rely on manual fieldwork, which has drawbacks such as low efficiency, high cost, and high risk. UAV geological surveys, by acquiring surface information through high-resolution imagery, have become a key means to improve survey efficiency.

[0003] Traditional techniques use static radiometric correction models to process UAV imagery. However, these models cannot adapt to dynamically changing observation conditions, easily leading to uneven shadow distribution and radiometric distortion, which in turn affects the accuracy of geological information extraction. When facing the interpretation needs of complex geological environments, traditional techniques use pre-trained geological interpretation models. However, these models have poor generalization ability, easily resulting in errors in geological entity identification, reliance on manual topological repair of fault structures, and interference from similar spectra in lithological boundary segmentation. Furthermore, traditional techniques lack a data-model co-optimization mechanism in geological information extraction and interpretation, leading to reliance on manual interpretation for confidence assessment of interpretation results, failure to automatically trigger re-acquisition in low-confidence areas, and inefficient model updates that require training with the entire dataset. Summary of the Invention

[0004] Therefore, it is necessary to provide UAV-based regional geological survey methods, systems, equipment, and media to address the aforementioned technical issues, in order to improve the accuracy of image radiometric correction, enhance the generalization ability of geological interpretation models, optimize the efficiency of low-confidence area identification, and achieve the technical effect of synergistic optimization of data acquisition and model updating.

[0005] In the first aspect, this application provides a regional geological survey method based on unmanned aerial vehicles (UAVs), the method comprising:

[0006] The original time-series image sequence, digital surface model and flight metadata of the survey area are acquired simultaneously through the optical sensors and positioning devices carried by the UAV.

[0007] Based on flight metadata and digital surface models, observation condition interference separation is performed on the original time-series image sequence to generate intrinsic reflectance images.

[0008] Using a pre-trained interpretation model, geological entities are identified in intrinsic reflectance images to obtain initial geological interpretation results.

[0009] Confidence assessment is performed based on the initial geological interpretation results to generate confidence assessment results; data reacquisition is triggered based on the confidence assessment results; observation condition interference is separated on the reacquisition image sequence based on the reacquisition flight metadata and digital surface model to generate optimized data; the pre-trained interpretation model is fine-tuned based on the optimized data to generate an updated model; the optimized data is reinterpreted using the updated model to output geological maps.

[0010] In one embodiment, based on flight metadata and a digital surface model, observation condition interference separation is performed on the original time-series image sequence to generate an intrinsic reflectance image, including:

[0011] Based on the solar azimuth and digital surface model in the flight metadata, dynamic shadow projection modeling is performed to obtain a spatiotemporal continuous shadow distribution map.

[0012] Illumination consistency correction was performed on the original temporal image sequence using a spatiotemporally continuous shadow distribution map to obtain a radiometrically normalized image.

[0013] Based on a pre-defined multi-scale atmospheric transport model, atmospheric scattering effects are eliminated from the radiation-normalized image to generate an atmospherically corrected image.

[0014] An anisotropic diffusion filtering algorithm is used to suppress sensor noise in atmospheric correction images, resulting in intrinsic reflectance images.

[0015] In one embodiment, illumination consistency correction is performed on the original temporal image sequence using a spatiotemporally continuous shadow distribution map to obtain a radiometrically normalized image, including:

[0016] The original temporal image sequence is segmented into shadow regions using a spatiotemporally continuous shadow distribution map to generate shadow region masks.

[0017] Based on a pre-defined solar-sensor geometric constraint model, radiative transmission compensation is performed on the image area covered by the shadow mask to obtain a radiative compensation image.

[0018] Extracting unmasked regions from the original temporal image sequence using shadow region masks;

[0019] A nonlinear photometric alignment algorithm is used to perform spectral consistency fusion between radiometrically compensated images and unmasked area images to generate radiometrically normalized images.

[0020] In one embodiment, an anisotropic diffusion filtering algorithm is used to suppress sensor noise in atmospheric correction images to obtain intrinsic reflectance images, including:

[0021] Gradient features are extracted from atmospheric corrected images to obtain gradient magnitude maps;

[0022] Based on the preset edge preservation constraint function, the diffusion coefficient of the gradient magnitude map is calculated to generate an anisotropic diffusion coefficient map.

[0023] Multi-scale diffusion processing of atmospheric correction images is performed using anisotropic diffusion coefficient maps to generate denoised images;

[0024] When the change in the denoised image between adjacent iterations is less than the preset convergence threshold, the intrinsic reflectance image is output.

[0025] In one embodiment, a pre-trained interpretation model is used to identify geological entities in the intrinsic reflectance image to obtain initial geological interpretation results, including:

[0026] Multi-scale feature pyramid extraction is performed on the intrinsic reflectance image to obtain a spatial-spectral joint feature map. The expression for the spatial-spectral joint feature map is as follows:

[0027]

[0028] in, Represents the joint spatial-spectral feature map. Indicates intrinsic reflectance image. This represents the Gaussian pyramid operator. This represents the convolution operation. Indicates feature cascading, Indicates feature weighting, Represents the spectral weight matrix. This represents the bias vector. Indicates the number of pyramid levels. Indicates a corrected linear unit. Indicates scale index;

[0029] By employing a geological entity attention mechanism, we model land cover associations using the spatial-spectral joint feature map, generating a geological context feature map. The expression for the geological context feature map is as follows:

[0030]

[0031] in, Represents geological context features. Represents the joint spatial-spectral feature map. This represents the Softmax function. Representing feature dimension, Indicates the weight of geological features. This represents the convolution operation. Represents a linear unit with Gaussian error. Represents the query matrix. Indicates the first The key matrix of each attention head. Indicates the first The value matrix of each attention head, Indicates the attention head index, Indicates the total number of attention heads. This represents the matrix transpose operator;

[0032] Based on a pre-defined fault-rock stratum topology rule set, the geological context feature map is segmented by structural constraints to generate an initial geological segmentation map.

[0033] The initial geological interpretation results are obtained by performing topological repair on the fracture structures in the initial geological segmentation map using a morphological fault connection algorithm.

[0034] In one embodiment, a morphological fault connectivity algorithm is used to perform topological repair on the fault structures in the initial geological segmentation map to obtain the initial geological interpretation results, including:

[0035] The fracture structure framework is extracted from the initial geological segmentation map to obtain the fracture framework line map;

[0036] Based on the preset fracture ductility criterion, spatial attraction domain analysis is performed on the fracture endpoints in the fracture skeleton line diagram to generate candidate fracture connection point pairs. The expression for the candidate fracture connection point pairs is as follows:

[0037]

[0038] in, Indicates a pair of candidate points for broken connections. This represents the distance vector between endpoints. Indicates the directional consistency threshold. Represents the angle between vectors. Indicates the maximum acceptable angle. Indicates the coordinates of the fracture endpoints. Indicates endpoint The direction vector at that location, Indicates endpoint The direction vector at that location, Indicates endpoint The structural stress vector at that location, Indicates endpoint The structural stress vector at the location;

[0039] An adaptive morphological path reconstruction algorithm is used to perform geological topological connections on candidate fracture connection points to generate a fracture network map for repair.

[0040] The non-fractured areas in the repaired fault network map and the initial geological segmentation map are merged to generate the initial geological interpretation results.

[0041] In one embodiment, based on a preset fault-stratum topology rule set, structural constraint segmentation is performed on the geological context feature map to generate an initial geological segmentation map, including:

[0042] Using a fault orientation constraint segmentation algorithm, the fault structures in the geological context feature map are segmented with consistent orientation to generate a fault region map.

[0043] Based on the rock layer contact relationship rules, the lithological boundaries in the geological context feature map are segmented by attitude constraints to generate a lithological unit map;

[0044] A topological conflict resolution algorithm is used to fuse the fault region map and lithological unit map to achieve geological rule consistency, resulting in an initial geological segmentation map. The expression for the initial geological segmentation map is as follows:

[0045]

[0046] in, This represents the initial geological segmentation map. Map showing the fractured region. Represents a lithological unit diagram. This represents the topological conflict resolution operator. Indicates feature fusion operation, Indicates the conflict correction factor. Indicates the conflict indicator function. Represents the set of geological boundaries. Represents the set of fracture boundaries. This represents the set of lithological boundaries.

[0047] Secondly, this application also provides a UAV-based regional geological survey system, which includes:

[0048] The UAV data acquisition module is used to simultaneously acquire the original time-series image sequence, digital surface model and flight metadata of the survey area through the optical sensors and positioning devices carried by the UAV.

[0049] The observation condition interference separation module is used to separate observation condition interference from the original time-series image sequence based on flight metadata and digital surface model, and generate intrinsic reflectance image;

[0050] The geological entity recognition module is used to identify geological entities in intrinsic reflectance images through a pre-trained interpretation model to obtain initial geological interpretation results.

[0051] The interpretation result optimization module is used to perform confidence assessment based on the initial geological interpretation results and generate confidence assessment results; trigger data reacquisition based on the confidence assessment results; perform observation condition interference separation on the reacquisition image sequence based on the reacquisition flight metadata and digital surface model to generate optimized data; fine-tune the pre-trained interpretation model based on the optimized data to generate an updated model; and reinterpret the optimized data using the updated model to output geological maps.

[0052] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods in the first aspect of this application.

[0053] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods in the first aspect of this application.

[0054] This application provides a method, system, equipment, and medium for regional geological surveys based on unmanned aerial vehicles (UAVs). The method includes: simultaneously acquiring the original time-series image sequence, digital surface model, and flight metadata of the survey area using optical sensors and positioning devices mounted on the UAV, providing a basic support for spatiotemporal matching for subsequent geological information processing; combining flight metadata and digital surface model to separate observation condition interferences from the original time-series image sequence, dynamically eliminating the effects of shadow projection and atmospheric scattering, and improving the accuracy of image radiometric correction; and relying on a pre-trained interpretation model to identify geological entities in intrinsic reflectance images, combining multi-scale spatial-spectral feature extraction and geological entity attention mechanisms to enhance the generalization ability of the geological interpretation model and reduce the probability of geological entity identification deviation and lithological boundary segmentation being interfered with by similar spectra.

[0055] Confidence assessment is performed based on the initial geological interpretation results. Data reacquisition of low-confidence areas is triggered based on the assessment results. Optimized data is then generated by combining the reacquisitioned flight metadata with the digital surface model, and the pre-trained interpretation model is fine-tuned to optimize the identification efficiency of low-confidence areas. At the same time, a closed loop is formed between data acquisition and model updates, achieving synergistic optimization of data acquisition and model updates. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1This is a flowchart of a UAV-based regional geological survey method according to one embodiment of the present invention;

[0058] Figure 2 This is a flowchart of a method for performing illumination consistency correction on an original temporal image sequence using a spatiotemporally continuous shadow distribution map to obtain a radiometrically normalized image in one embodiment of the present invention.

[0059] Figure 3 This is a structural diagram of a UAV-based regional geological survey system according to one embodiment of the present invention. Detailed Implementation

[0060] To make the above-mentioned objects, features, and advantages of this application more apparent and understandable, the specific embodiments of this application will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0061] First, the application scenarios of the embodiments of this application are described. In the embodiments of this application, a method, system, equipment and medium for regional geological survey based on UAVs are provided, which are applicable to scenarios such as mineral resource exploration in mountainous areas, geological disaster investigation in the permafrost zone of the Qinghai-Tibet Plateau, and geological environment monitoring of coastal mudflats.

[0062] In illustrative purposes, the UAV-based regional geological survey methods, systems, equipment, and media provided in this application embodiment can also be applied to other application scenarios such as geological assessment of large-scale engineering site selection, early warning of landslide risks around cities, and background investigation of farmland soil geology. These are merely illustrative examples and do not limit the specific application scenarios.

[0063] like Figure 1 As shown, this application provides a regional geological survey method based on unmanned aerial vehicles (UAVs), the method comprising:

[0064] S101: Through the optical sensors and positioning devices carried by the UAV, it simultaneously acquires the original time-series image sequence, digital surface model and flight metadata of the survey area.

[0065] For example, the UAV geological survey terminal arrives at the designated survey area and maintains a stable flight attitude according to the preset regional geological survey plan. The UAV geological survey terminal uses its onboard optical sensor to continuously capture images of the surface of the survey area, and at the same time activates its onboard positioning device to record the position information and flight attitude information that are synchronized with the image capture action in the time dimension in real time, ensuring that the image acquisition process of the optical sensor and the information recording process of the positioning device form a precise spatiotemporal correspondence.

[0066] The UAV geological survey terminal uses continuous image data captured by optical sensors, combined with synchronous position information and flight attitude information recorded by the positioning device, to reconstruct the three-dimensional terrain structure and generate a digital surface model of the survey area. Throughout the entire process of image acquisition, positioning information recording, and three-dimensional terrain reconstruction, the UAV geological survey terminal synchronously records flight metadata, including flight parameters, environmental parameters, and equipment operating status parameters.

[0067] S102: Based on flight metadata and digital surface models, observation condition interference is separated from the original time-series image sequence to generate intrinsic reflectance images.

[0068] For example, the UAV geological survey terminal performs spatiotemporal registration of flight metadata, digital surface models, and original time-series image sequences to ensure consistency of spatiotemporal information among the three. The UAV geological survey terminal extracts observation condition-related information from the flight metadata and, combined with the terrain features of the digital surface model, identifies observation condition interference information in the original time-series image sequences.

[0069] The UAV geological survey terminal strips the observation condition interference information from the original time-series image sequence, separates the interference information from the effective image information, and performs radiometric correction and fidelity processing on the image after interference stripping to generate intrinsic reflectance image.

[0070] S103: Using a pre-trained interpretation model, geological entities are identified in the intrinsic reflectance image to obtain initial geological interpretation results.

[0071] For example, the UAV geological survey terminal retrieves a pre-trained interpretation model and intrinsic reflectance imagery, performs standardized preprocessing on the intrinsic reflectance imagery to adapt it to the input requirements of the pre-trained interpretation model. Using the pre-trained interpretation model, the UAV geological survey terminal extracts spatial structure features and spectral response features from the pre-processed intrinsic reflectance imagery, and combines these features to determine the type and spatial location of geological entities in the imagery, clarifying the boundaries and distribution range of different geological entities.

[0072] The UAV geological survey terminal performs topological consistency verification between the judgment results of geological entities and the positioning information, corrects misidentification problems, integrates the verified information, and obtains the initial geological interpretation results.

[0073] S104: Based on the initial geological interpretation results, a confidence assessment is performed to generate a confidence assessment result; based on the confidence assessment result, data reacquisition is triggered; based on the reacquisition flight metadata and digital surface model, observation condition interference is separated from the reacquisition image sequence to generate optimized data; based on the optimized data, the pre-trained interpretation model is fine-tuned to generate an updated model; the updated model is used to reinterpret the optimized data to output geological maps.

[0074] For example, the UAV geological survey terminal combines evaluation criteria such as consistency of geological entity identification and accuracy of boundary delineation to conduct a confidence assessment of the initial geological interpretation results and generate a confidence assessment result. Based on the confidence assessment result, the UAV geological survey terminal identifies interpretation areas that do not meet the preset reliability level, triggers data re-acquisition for the corresponding areas, and after the re-acquisition is completed, retrieves the re-acquisition flight metadata, the re-acquisition digital surface model, and the re-acquisition image sequence, processes the re-acquisition image sequence according to the observation condition interference separation process, removes interference information, and generates optimized data.

[0075] The UAV geological survey terminal uses optimized data as a basis to iteratively optimize and fine-tune the parameters of the pre-trained interpretation model, adapting it to the geological characteristics of the re-collected area and generating an updated model. The UAV geological survey terminal then calls the updated model to reinterpret the optimized data, integrates various types of geological information, organizes it according to standardized formats, and outputs geological maps.

[0076] One embodiment of this application provides a regional geological survey method based on unmanned aerial vehicles (UAVs), comprising: synchronously acquiring the original time-series image sequence, digital surface model, and flight metadata of the survey area through optical sensors and positioning devices mounted on the UAV, providing a basic support for spatiotemporal matching for subsequent geological information processing; combining flight metadata and digital surface model to separate observation condition interferences in the original time-series image sequence, dynamically eliminating the influence of shadow projection and atmospheric scattering, and improving the image radiometric correction accuracy; relying on a pre-trained interpretation model to identify geological entities in intrinsic reflectance images, and combining multi-scale spatial-spectral feature extraction and geological entity attention mechanism to enhance the generalization ability of the geological interpretation model and reduce the probability of geological entity identification deviation and lithological boundary segmentation being interfered with by similar spectra.

[0077] Confidence assessment is performed based on the initial geological interpretation results. Data reacquisition of low-confidence areas is triggered based on the assessment results. Optimized data is then generated by combining the reacquisitioned flight metadata with the digital surface model, and the pre-trained interpretation model is fine-tuned to optimize the identification efficiency of low-confidence areas. At the same time, a closed loop is formed between data acquisition and model updates, achieving synergistic optimization of data acquisition and model updates.

[0078] In one embodiment, based on flight metadata and a digital surface model, observation condition interference separation is performed on the original time-series image sequence to generate an intrinsic reflectance image, including:

[0079] (1) Based on the solar azimuth and digital surface model in the flight metadata, dynamic shadow projection modeling is performed to obtain a spatiotemporal continuous shadow distribution map.

[0080] For example, the UAV geological survey terminal retrieves flight metadata and digital surface models, performs terrain elevation normalization processing on the digital surface models to eliminate the impact of terrain benchmark differences on shadow simulation, and then extracts complete solar azimuth information from the flight metadata and matches the solar azimuth information with the normalized digital surface models in terms of spatiotemporal dimensions.

[0081] The UAV geological survey terminal constructs a dynamic shadow projection model by simulating the projection path and shading relationship of sunlight on the terrain surface. It continuously iterates and optimizes the model parameters to fit the light change pattern at different times, generating a spatiotemporal continuous shadow distribution map covering the entire survey area at all times.

[0082] Among them, the solar azimuth information includes the horizontal azimuth angle and vertical altitude angle data of the sun, and the spatiotemporal continuous shadow distribution map is a continuous image map that can reflect the changes in the shadow coverage range and intensity at different time points.

[0083] (2) The original temporal image sequence was corrected for illumination consistency by using the spatiotemporal continuous shadow distribution map to obtain the radiation normalized image.

[0084] For example, the UAV geological survey terminal retrieves the spatiotemporal continuous shadow distribution map and the original time-series image sequence. Based on the shadow boundary information in the spatiotemporal continuous shadow distribution map, it performs pixel-level shadow region segmentation on the original time-series image sequence to generate a shadow region mask. Then, based on the solar-sensor geometric relationship model, it calculates the amount of missing radiation energy in the shadow region and performs pixel-by-pixel radiation compensation processing on the image part covered by the shadow region mask according to the amount of missing energy.

[0085] The UAV geological survey terminal adjusts the photometric consistency of areas not covered by shadows in the original time-series image sequence. It then uses a nonlinear fusion algorithm to seamlessly stitch together the radiation-compensated shadow area image with the photometric-adjusted non-shadow area image to obtain a radiation-normalized image with unified illumination conditions.

[0086] Among them, the shadow area mask is a binary image that marks the shadow coverage area in the image, the radiation compensation processing is a targeted processing method that supplements the corresponding radiation energy according to the degree of light loss, and the nonlinear fusion algorithm is a fusion technology that ensures a natural transition between images of different regions.

[0087] (3) Based on the preset multi-scale atmospheric transmission model, atmospheric scattering effect is eliminated in the radiation normalized image to generate atmospheric corrected image.

[0088] For example, the UAV geological survey terminal calls a preset multi-scale atmospheric transmission model to perform atmospheric parameter inversion on the radiation normalized image, obtain key parameters such as atmospheric optical thickness and aerosol concentration in different areas of the survey area, and then divides the radiation normalized image into different atmospheric influence levels according to the above parameters.

[0089] The UAV geological survey terminal uses a scattering elimination algorithm corresponding to a multi-scale atmospheric transport model to perform layered stripping of atmospheric scattering components in the images, targeting the atmospheric characteristics at each level. During the process, the authenticity of the image radiation information is verified in real time to avoid over-processing that could lead to distortion of geological features, and atmospheric correction images with atmospheric scattering interference eliminated are generated.

[0090] Among them, the multi-scale atmospheric transport model is a simulation model that covers the transport laws of different atmospheric levels such as the near-surface layer and the troposphere, while atmospheric parameter inversion is a calculation process that infers key atmospheric physical parameters from image radiation information.

[0091] (4) Sensor noise suppression is performed on atmospheric correction images by an anisotropic diffusion filtering algorithm to obtain intrinsic reflectance images.

[0092] For example, the UAV geological survey terminal uses an anisotropic diffusion filtering algorithm to extract global gradient features from atmospheric correction images, distinguish geological edge features from sensor noise signals in the images, and calculate an adaptive diffusion coefficient based on the extracted gradient features, so that the diffusion coefficient exhibits a differentiated distribution as the gradient changes.

[0093] The UAV geological survey terminal applies the diffusion coefficient to the multi-scale diffusion processing of atmospheric correction images, increases the diffusion intensity in areas with concentrated noise, and retains diffusion constraints in geological edge areas. Through multiple iterations, the sensor noise signal is reduced to a preset range without loss of geological detail features, resulting in an intrinsic reflectivity image that can truly reflect the geological reflectivity characteristics of the surface.

[0094] Among them, the anisotropic diffusion filtering algorithm is a filtering technique that achieves noise suppression and edge preservation by dynamically adjusting the diffusion intensity. The diffusion coefficient is a parameter that changes dynamically according to the image gradient characteristics and is used to control the degree of diffusion in different regions.

[0095] like Figure 2 As shown, illumination consistency correction is performed on the original temporal image sequence using a spatiotemporal continuous shadow distribution map to obtain a radiometrically normalized image, including:

[0096] S201: The original temporal image sequence is segmented into shadow regions using a spatiotemporally continuous shadow distribution map to generate a shadow region mask.

[0097] For example, the UAV geological survey terminal retrieves the spatiotemporal continuous shadow distribution map and the original time-series image sequence, performs edge enhancement processing on the spatiotemporal continuous shadow distribution map to highlight the boundary contour features between the shadow area and the non-shadow area, and extracts the grayscale threshold and distribution pattern of the shadow pixels in the spatiotemporal continuous shadow distribution map.

[0098] The UAV geological survey terminal preprocesses the original time-series image sequence to eliminate slight noise interference in the image to improve segmentation accuracy. Then, based on the grayscale threshold of shadow pixels and boundary contour features, it performs pixel-by-pixel shadow determination on the original time-series image sequence to clarify whether each pixel belongs to the shadow area or the non-shadow area. Based on the determination results, a shadow area mask is generated to mark the pixel position, boundary and coverage of the shadow area.

[0099] The shadow region mask includes binarized image data that marks the position, boundary, and coverage of shadow region pixels in the original temporal image sequence. The binarized image data uses 0 and 1 to represent non-shadow region pixels and shadow region pixels, respectively.

[0100] S202: Based on the preset solar-sensor geometric constraint model, radiative transmission compensation is performed on the image area covered by the shadow mask to obtain a radiative compensation image.

[0101] For example, the UAV geological survey terminal retrieves a preset solar-sensor geometric constraint model, a shadow area mask, and an original time-series image sequence. It then extracts the image area covered by the shadow area mask from the original time-series image sequence and obtains basic attribute data such as the radiance value and image texture features of that area.

[0102] The UAV geological survey terminal uses a solar-sensor geometric constraint model to analyze the angle of sunlight, the sensor's shooting posture, and the relative positional relationship between the two. Combined with the influence of atmospheric transmission on radiation energy, it calculates the radiation transmission attenuation and radiation energy loss in the shaded area covered by the mask, determines the corresponding radiation compensation coefficient, and applies the radiation compensation coefficient to the shaded area image on a pixel-by-pixel basis to complete the radiation energy replenishment and transmission characteristic correction, thus obtaining a radiation-compensated image.

[0103] Among them, the solar-sensor geometric constraint model includes a geometric calculation model that reflects the relationship between the angle of sunlight, the angle of sensor shooting and the law of image radiative transmission, and the radiative transmission compensation includes image processing methods for supplementing radiative energy and correcting radiative transmission characteristics for shadowed areas.

[0104] S203: Extract unmasked areas from the original time-series image sequence using a shadow area mask.

[0105] For example, the UAV geological survey terminal retrieves the shadow area mask and the original time-series image sequence, performs morphological processing on the shadow area mask, smooths the shadow area boundary through dilation and erosion operations, and fills the tiny holes in the mask to ensure a more accurate division between the shadow area and the non-shadow area.

[0106] The UAV geological survey terminal establishes region extraction rules based on the processed shadow area mask, clarifies the pixel judgment criteria for non-shadow areas, and then performs region filtering operation frame by frame on the original time-series image sequence, removing the shadow area image portion marked by the shadow area mask, and completely retaining and integrating all pixel data of non-shadow areas in the original time-series image sequence. After verifying the completeness of the extraction, the unmasked area image is obtained.

[0107] Among them, the unshaded area image includes complete unshaded area image data obtained after regional extraction from the original time-series image sequence that was not covered by shadows. This image data retains all the geological feature information of the unshaded areas in the original time-series image sequence.

[0108] S204: A nonlinear photometric alignment algorithm is used to perform spectral consistency fusion between radiometrically compensated images and unmasked area images to generate radiometrically normalized images.

[0109] For example, the UAV geological survey terminal retrieves radiation-compensated images and images of unshaded areas, extracts key parameters such as spectral response characteristics, mean light intensity, and contrast of the radiation-compensated images and images of unshaded areas respectively, and establishes a parameter comparison matrix for the two types of images.

[0110] The UAV geological survey terminal adopts a nonlinear photometric alignment algorithm to construct a photometric mapping relationship between the radiometric compensated image and the unshaded area image based on the parameter comparison matrix. It adaptively adjusts the spectral response characteristics and illumination intensity of the radiometric compensated image to keep it consistent with the photometric benchmark of the unshaded area image. Then, it performs transition processing on the edge overlapping areas of the two types of images, performs a global spectral consistency fusion operation, verifies the illumination uniformity and geological feature integrity of the fused image, and generates a radiometric normalized image.

[0111] Among them, the nonlinear photometric alignment algorithm includes an image processing algorithm that achieves unified matching of photometric levels and spectral characteristics of different images through nonlinear transformation, and the radiometric normalized image includes standardized surface image data after completing illumination consistency correction and spectral fusion.

[0112] In one embodiment, an anisotropic diffusion filtering algorithm is used to suppress sensor noise in atmospheric correction images to obtain intrinsic reflectance images, including:

[0113] (1) Extract gradient features from atmospheric corrected images to obtain gradient amplitude maps.

[0114] For example, the UAV geological survey terminal retrieves atmospheric correction images, performs global grayscale standardization on the atmospheric correction images to eliminate gradient calculation interference caused by grayscale differences in different regions, and performs a slight smoothing operation on the atmospheric correction images to weaken the influence of minor texture fluctuations on gradient features.

[0115] The UAV geological survey terminal uses a preset gradient extraction operator to perform pixel-by-pixel gradient calculation on the standardized atmospheric correction image. It obtains the gradient components of each pixel in the horizontal, vertical and diagonal directions, integrates the gradient components in each direction to calculate the pixel comprehensive gradient value, and generates a gradient magnitude map covering the entire area based on the comprehensive gradient value of all pixels.

[0116] The gradient magnitude map includes image data that reflects the degree of grayscale change between each pixel and its neighboring pixels in the atmospheric correction image, as well as the gradient differences between geological edges and noise areas.

[0117] (2) Based on the preset edge preservation constraint function, the diffusion coefficient of the gradient magnitude map is calculated to generate an anisotropic diffusion coefficient map.

[0118] For example, the UAV geological survey terminal retrieves the preset edge preservation constraint function and gradient magnitude map, performs full-domain statistical analysis on the gradient magnitude map, divides the strong and weak intervals of gradient magnitude, and clarifies the high gradient interval corresponding to the geological edge area and the low gradient interval corresponding to the sensor noise area.

[0119] The UAV geological survey terminal substitutes the gradient values ​​of each pixel in the gradient magnitude map into a preset edge-preserving constraint function, and dynamically calculates the diffusion coefficient of each pixel according to the interval where the gradient value is located. This results in a smaller diffusion coefficient for geological edge areas with high gradients and a larger diffusion coefficient for noise areas with low gradients. The diffusion coefficients of all pixels are then integrated to generate an anisotropic diffusion coefficient map.

[0120] Among them, the edge preservation constraint function includes a mathematical constraint model that can dynamically adjust the diffusion coefficient according to the gradient magnitude and take into account both noise suppression and edge preservation, and the anisotropic diffusion coefficient map includes a coefficient distribution image that records the diffusion intensity corresponding to each pixel in the atmospheric correction image.

[0121] (3) Use the anisotropic diffusion coefficient map to perform multi-scale diffusion processing on the atmospheric correction image to generate a denoised image.

[0122] For example, the UAV geological survey terminal retrieves anisotropic diffusion coefficient maps and atmospheric correction images to determine the basic scale and iteration level of multi-scale diffusion processing, and clarifies the diffusion processing range and intensity weight of different levels.

[0123] The UAV geological survey terminal applies the diffusion coefficients in the anisotropic diffusion coefficient map to the atmospheric correction image layer by layer according to the preset iteration level. It performs strong diffusion processing on low gradient noise areas to weaken the noise signal and weak diffusion processing on high gradient geological edge areas to preserve edge details. Through multi-scale iteration, it completes global diffusion denoising and generates denoised images.

[0124] Multi-scale diffusion processing includes a noise suppression method that proceeds step by step according to different scale levels and is executed differently based on the diffusion coefficient. Denoising images include image data with weakened sensor noise after multi-scale diffusion processing.

[0125] (4) When the change in adjacent iterations of the denoised image is less than the preset convergence threshold, output the intrinsic reflectance image.

[0126] For example, the UAV geological survey terminal records the denoised image generated in the current iteration and the denoised image generated in the previous iteration, calculates the difference between the two types of denoised images in terms of global pixel gray value and gradient features, and quantifies the change in adjacent iterations.

[0127] The UAV geological survey terminal compares the calculated changes in adjacent iterations with a preset convergence threshold. If the changes in adjacent iterations are greater than or equal to the preset convergence threshold, it returns to the multi-scale diffusion processing step to continue iterating. If the changes in adjacent iterations are less than the preset convergence threshold, it indicates that the denoising effect has reached a stable state, and the iteration stops and the intrinsic reflectance image is output.

[0128] Among them, the change in adjacent iterations includes the quantized value of the difference in pixel features and gradient distribution between the denoised images generated in the two iterations, and the preset convergence threshold includes the standard threshold parameter used to determine whether the denoising process has reached a stable state.

[0129] In one embodiment, a pre-trained interpretation model is used to identify geological entities in the intrinsic reflectance image to obtain initial geological interpretation results, including:

[0130] (1) Multi-scale feature pyramid extraction is performed on the intrinsic reflectance image to obtain a spatial-spectral joint feature map. The expression of the spatial-spectral joint feature map is:

[0131]

[0132] in, Represents the joint spatial-spectral feature map. Indicates intrinsic reflectance image. This represents the Gaussian pyramid operator. This represents the convolution operation. Indicates feature cascading, Indicates feature weighting, Represents the spectral weight matrix. This represents the bias vector. Indicates the number of pyramid levels. Indicates a corrected linear unit. Indicates scale index.

[0133] For example, the UAV geological survey terminal retrieves the intrinsic reflectance image, performs grayscale normalization on the intrinsic reflectance image to eliminate the interference of radiation differences in different regions on feature extraction, and then calls the Gaussian pyramid operator to perform multi-scale downsampling decomposition on the normalized intrinsic reflectance image to generate image layers corresponding to different resolutions.

[0134] The UAV geological survey terminal performs convolution operations on each resolution image layer to extract spatial texture features and spectral response features of each layer. It calls the spectral weight matrix to adjust the extracted features at the channel level, and after superimposing the bias vector, it inputs the modified linear unit for nonlinear activation. The activated features of each layer are concatenated according to the channel dimension to generate a spatial-spectral joint feature map.

[0135] Among them, the Gaussian pyramid operator includes a multi-scale decomposition tool that generates multi-resolution image layers through downsampling and Gaussian filtering; the convolution operation includes a computation method that extracts local spatial and spectral features of the image through learnable convolution kernels; the feature concatenation includes an operation that splices and integrates feature tensors of different scales along the channel dimension; and the corrected linear unit includes a nonlinear activation unit that can retain positive eigenvalues ​​and suppress negative eigenvalues.

[0136] (2) By using the geological entity attention mechanism, the spatial-spectral joint feature map is used to model the land cover association and generate a geological context feature map. The expression of the geological context feature map is:

[0137]

[0138] in, Represents geological context features. Represents the joint spatial-spectral feature map. This represents the Softmax function. Representing feature dimension, Indicates the weight of geological features. This represents the convolution operation. Represents a linear unit with Gaussian error. Represents the query matrix. Indicates the first The key matrix of each attention head. Indicates the first The value matrix of each attention head, Indicates the attention head index, Indicates the total number of attention heads. This represents the matrix transpose operator.

[0139] For example, the UAV geological survey terminal retrieves the spatial-spectral joint feature map, inputs the spatial-spectral joint feature map into the geological entity attention mechanism, calculates the initial attention weight of each attention head by querying the matrix, the key matrix of the kth attention head and the value matrix of the kth attention head, and then normalizes the initial attention weight by the Softmax function to obtain the attention distribution reflecting the correlation strength between different ground features.

[0140] The UAV geological survey terminal performs convolution operations on the spatial-spectral joint feature map, adjusts the convolutional feature tensor through geological feature weights, inputs it into the Gaussian error linear unit for smooth nonlinear transformation, weights and fuses the transformed feature tensor with the attention distribution, and then stitches and integrates the fusion results of all attention heads to generate a geological context feature map that includes global relational information of ground features.

[0141] The attention head includes a parallel attention calculation unit capable of independently calculating the local feature relationships; the Softmax function includes a normalization function capable of converting weight values ​​into a probability distribution; the Gaussian error linear unit includes a nonlinear activation unit capable of smoothing and suppressing negative eigenvalues ​​while retaining positive eigenvalues; and the geological feature weights include a learnable parameter matrix used to adjust the contribution of geological features.

[0142] (3) Based on the preset fault-rock topology rule set, the geological context feature map is segmented by structural constraints to generate an initial geological segmentation map.

[0143] For example, the UAV geological survey terminal retrieves a preset fault-rock stratum topology rule set, which includes geological prior rules such as the spatial adjacency relationship between faults and rock strata, morphological extension constraints, and hierarchical stacking logic. Then, it performs global feature mapping on the geological context feature map, mapping the pixels in the feature map to the corresponding geological semantic candidate categories.

[0144] The UAV geological survey terminal applies the fault-rock stratum topology rule set to geological semantic candidate categories, performs topology consistency verification on the category of each pixel, eliminates candidate categories that do not conform to the topology rules, and then performs regional segmentation on geological entities based on the verification results, distinguishes the boundaries and distribution range of different geological entities such as faults and rock strata, and generates an initial geological segmentation map.

[0145] Among them, the fault-strata topology rule set includes a set of topology constraint rules constructed based on regional geological survey experience, the structural constraint segmentation includes a semantic segmentation method of combining topology rules to feature maps, and the initial geological segmentation map includes multi-valued image data that marks different geological entity regions.

[0146] (4) The fault structures in the initial geological segmentation map are repaired by topological repair using the morphological fault connection algorithm to obtain the initial geological interpretation results.

[0147] For example, the UAV geological survey terminal retrieves the initial geological segmentation map, uses a morphological fault connection algorithm to identify the faults in the initial geological segmentation map, extracts the endpoints and fracture areas of the fault line segments through edge detection, and analyzes the topological continuity, morphological orientation and extension trend of the fracture area.

[0148] The UAV geological survey terminal performs topological repair on the fault based on the analysis results. It connects the fault segments through morphological expansion and corrosion operations, fills the tiny cavities in the fault area, and then verifies the topological integrity of the repaired segmentation map to ensure that the spatial structure of the geological entity conforms to geological laws and obtains the initial geological interpretation results.

[0149] Among them, the morphological fault connection algorithm includes a professional algorithm for repairing faulted geological structures based on morphological operations, the topology repair includes processing operations to restore the spatial continuity of faulted geological entities, and the initial geological interpretation results include the geological entity identification and segmentation results after the topology repair is completed.

[0150] In one embodiment, a morphological fault connectivity algorithm is used to perform topological repair on the fault structures in the initial geological segmentation map to obtain the initial geological interpretation results, including:

[0151] (1) Extract the fracture structure skeleton from the initial geological segmentation map to obtain the fracture skeleton line map.

[0152] For example, the UAV geological survey terminal retrieves the initial geological segmentation map, performs full-domain binarization processing on the initial geological segmentation map, sets the pixels of the fault structure area to the first gray value, and sets the pixels of the non-fault structure area to the second gray value, thereby enhancing the pixel difference between the two types of areas.

[0153] The UAV geological survey terminal performs morphological opening operations on the binarized initial geological segmentation map to eliminate tiny isolated noise pixels and burr structures in the map. Then, a thinning algorithm is used to peel off the fracture structure area layer by layer, retaining the single-pixel width, central axis and endpoint coordinate information of the fracture structure, and generating a fracture skeleton line map.

[0154] The fracture structure skeleton extraction includes a thinning process that transforms planar fracture structures into single-pixel-width central axes. The fracture skeleton line map includes single-pixel-width image data that records the direction of the central axis of the fracture structure, the position of the endpoints, and the branching relationships.

[0155] (2) Based on the preset fracture ductility criterion, spatial attraction domain analysis is performed on the fracture endpoints in the fracture skeleton line diagram to generate fracture connection candidate point pairs. The expression for the fracture connection candidate point pairs is:

[0156]

[0157] in, Indicates a pair of candidate points for broken connections. This represents the distance vector between endpoints. Indicates the directional consistency threshold. Represents the angle between vectors. Indicates the maximum acceptable angle. Indicates the coordinates of the fracture endpoints. Indicates endpoint The direction vector at that location, Indicates endpoint The direction vector at that location, Indicates endpoint The structural stress vector at that location, Indicates endpoint The structural stress vector at that location.

[0158] For example, the UAV geological survey terminal retrieves the fracture skeleton line map and the preset fracture ductility criterion, extracts the coordinate information of all fracture endpoints from the fracture skeleton line map, obtains the direction vector at each fracture endpoint through local gradient calculation, and obtains the tectonic stress vector at each fracture endpoint by combining regional geological stress data.

[0159] The UAV geological survey terminal calculates the distance vector between the endpoints for each pair of fault endpoints. It obtains the directional consistency parameter and the angle parameter between the tectonic stress vector through vector operation. The two types of parameters are compared with the directional consistency threshold and the maximum acceptable angle, respectively. Fault endpoint pairs that meet both conditions are selected to generate candidate fault connection point pairs.

[0160] Among them, the fracture ductility criterion includes a set of fracture connection judgment rules based on the extension law of geological structure and stress distribution characteristics; the spatial attraction domain analysis includes an analysis method that quantitatively evaluates the connection possibility between fracture endpoints through vector parameters; and the fracture connection candidate point pairs include combinations of potential connectable fracture endpoints that meet the fracture ductility criterion.

[0161] (3) By using the adaptive morphological path reconstruction algorithm, geological topological connections are made between candidate points of fracture connection to generate a fracture network diagram for repair.

[0162] For example, the UAV geological survey terminal retrieves candidate points of fracture connection and an adaptive morphological path reconstruction algorithm. For each candidate point of fracture connection, it analyzes the fracture strike, tectonic stress direction and surrounding geological features at the endpoints to generate a smooth connection path that conforms to geological laws.

[0163] The UAV geological survey terminal adopts an adaptive morphological path reconstruction algorithm. It performs topological connection operation on each candidate fracture connection point pair, optimizes the width and continuity of the connection path through multiple rounds of morphological expansion and erosion operations, eliminates jagged edges and breaks in the path, and then integrates all connected fracture structures to generate a repair fracture network map.

[0164] Among them, the adaptive morphological path reconstruction algorithm includes a morphological algorithm that can dynamically adjust the shape of the connection path according to geological features, the geological topology connection includes a topology repair operation to restore the spatial continuity of the fault structure, and the repaired fault network map includes a complete fault structure network image after the topology connection is completed.

[0165] (4) The non-fractured areas in the repaired fracture network map and the initial geological segmentation map are merged to generate the initial geological interpretation results.

[0166] For example, the UAV geological survey terminal retrieves the repair fault network map and the initial geological segmentation map, extracts the image data of the non-faulted areas from the initial geological segmentation map, and preserves the original geological texture and boundary features of the non-faulted areas.

[0167] The UAV geological survey terminal stitches the repaired fracture network map with the image data of the non-fracture area at the pixel level. It processes the boundary pixels of the two types of areas through an edge smoothing algorithm to ensure a natural transition. Then, it performs topological consistency verification on the fused image to confirm that the spatial structure of the geological entities conforms to geological laws and generates the initial geological interpretation results.

[0168] The non-fractured areas include geological entities such as rock strata and rock masses that are not covered by fault structures in the initial geological segmentation map. The fusion operation includes a processing method that integrates the repaired fault network map and the image data of the non-fractured areas into a unified geological image. The initial geological interpretation results include the geological entity identification and segmentation results after the fault structure topology repair is completed.

[0169] In one embodiment, based on a preset fault-stratum topology rule set, structural constraint segmentation is performed on the geological context feature map to generate an initial geological segmentation map, including:

[0170] (1) By using the fault strike constraint segmentation algorithm, the fault structures in the geological context feature map are segmented in a consistent direction to generate a fault area map.

[0171] For example, the UAV geological survey terminal retrieves the geological context feature map, performs global gradient feature extraction on the geological context feature map, identifies potential fault regions with linear extension characteristics by calculating the gray-level change rate between pixels, and then extracts fault strike constraint rules from the preset fault-rock topology rule set to clarify the extension direction range and continuity requirements of different fault structures.

[0172] The UAV geological survey terminal uses a fault strike constraint segmentation algorithm to perform pixel-by-pixel direction consistency verification on potential fault areas, eliminate pseudo-fault areas with discrete strikes and discontinuous extensions, retain fault structures that meet the strike constraints, and perform regional contour fitting to generate a fault area map containing fault structure location, strike, and boundary information.

[0173] Among them, the fault strike constraint segmentation algorithm includes a professional algorithm that combines fault extension direction features with geological prior rules to perform semantic segmentation, and the fault area map includes binary segmented image data that marks the spatial location, strike trend and boundary contour of the fault structure.

[0174] (2) Based on the rock layer contact relationship rules, the lithological boundaries in the geological context feature map are divided by attitude constraint to generate a lithological unit map.

[0175] For example, the UAV geological survey terminal retrieves the geological context feature map and the preset rock layer contact relationship rules, extracts the lithological spectral features and texture features of the geological context feature map, identifies potential boundary areas with lithological changes by analyzing the differences in pixel spectral response and texture structure, and then extracts the attitude constraints from the rock layer contact relationship rules to clarify the range of attitude parameters such as dip angle and strike of different lithological units.

[0176] The UAV geological survey terminal performs attitude consistency verification on potential boundary areas, distinguishes between real lithological boundaries that meet attitude constraints and false boundaries formed by noise or interference, divides the lithological boundaries that meet the conditions into closed regions, and generates a lithological unit map that includes the location, boundary and distribution information of different lithological units.

[0177] Among them, the rock strata contact relationship rules include a set of lithological boundary determination rules based on geological contact laws and lithological occurrence characteristics, and the lithological unit map includes multi-valued segmented image data that marks the spatial location, boundary outline and distribution range of different lithological units.

[0178] (3) Using a topological conflict resolution algorithm, the fault region map and the lithological unit map are fused to achieve geological rule consistency, resulting in an initial geological segmentation map. The expression for the initial geological segmentation map is:

[0179]

[0180] in, This represents the initial geological segmentation map. Map showing the fractured region. Represents a lithological unit diagram. This represents the topological conflict resolution operator. Indicates feature fusion operation, Indicates the conflict correction factor. Indicates the conflict indicator function. Represents the set of geological boundaries. Represents the set of fracture boundaries. This represents the set of lithological boundaries.

[0181] For example, the UAV geological survey terminal retrieves fault area maps, lithological unit maps, and topological conflict resolution algorithms, extracts the set of fault boundaries in the fault area map and the set of lithological boundaries in the lithological unit map, performs spatial overlay analysis on the two types of boundary sets, and identifies conflict areas with topological contradictions by comparing the spatial position and orientation relationship of the boundaries.

[0182] The UAV geological survey terminal uses a topological conflict resolution operator to correct conflict areas, marks conflict locations through a conflict indicator function, and adjusts the boundary morphology of conflict areas by combining conflict correction coefficients to make the spatial relationship between fault boundaries and lithological boundaries conform to geological rules. Then, feature fusion operations are performed on the fault area map and the lithological unit map to integrate the regional and boundary information of the two types of segmentation maps and generate an initial geological segmentation map that conforms to geological rules.

[0183] Among them, the topological conflict resolution algorithm includes a fusion algorithm that can correct topological contradictions at geological boundaries and ensure the consistency of geological structures after fusion; the topological conflict resolution operator includes a computational unit for adjusting the shape of conflict boundaries and optimizing spatial relationships; and the feature fusion operation includes the operation of integrating the boundary and regional information of the fault area map and the lithological unit map into a unified segmented image.

[0184] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0185] In one embodiment, such as Figure 3 As shown, this application also provides a UAV-based regional geological survey system 300, which includes:

[0186] The UAV data acquisition module 301 is used to simultaneously acquire the original time-series image sequence, digital surface model and flight metadata of the survey area through the optical sensors and positioning devices carried by the UAV.

[0187] The observation condition interference separation module 302 is used to separate the observation condition interference from the original time-series image sequence based on flight metadata and digital surface model, and generate intrinsic reflectance image;

[0188] The geological entity recognition module 303 is used to perform geological entity recognition on the intrinsic reflectance image through a pre-trained interpretation model to obtain the initial geological interpretation result;

[0189] The interpretation result optimization module 304 is used to perform confidence assessment based on the initial geological interpretation results and generate confidence assessment results; trigger data reacquisition based on the confidence assessment results; perform observation condition interference separation on the reacquisition image sequence based on the reacquisition flight metadata and digital surface model to generate optimized data; fine-tune the pre-trained interpretation model based on the optimized data to generate an updated model; and reinterpret the optimized data using the updated model to output geological maps.

[0190] Specifically, the UAV geological survey terminal includes: a UAV data acquisition module 301, an observation condition interference separation module 302, a geological entity identification module 303, and an interpretation result optimization module 304.

[0191] The UAV data acquisition module plans its flight path according to a pre-set regional geological survey plan, controls the UAV to reach the designated survey area, and maintains a stable flight attitude. The UAV data acquisition module activates its onboard optical sensor to continuously capture images of the survey area's surface, and simultaneously activates its onboard positioning device to record position and flight attitude information in real time, ensuring a precise spatiotemporal correspondence between the image acquisition by the optical sensor and the information recorded by the positioning device.

[0192] The UAV data acquisition module uses continuous image data captured by optical sensors, combined with synchronous position information and flight attitude information recorded by the positioning device, to perform three-dimensional terrain structure reconstruction processing and generate a digital surface model of the survey area. At the same time, throughout the entire process of image acquisition, positioning information recording, and three-dimensional terrain reconstruction, flight metadata, including flight parameters, environmental parameters, and equipment operating status parameters, is recorded synchronously.

[0193] The UAV data acquisition module includes optical sensors, positioning devices, and data processing units mounted on the UAV. The optical sensors include professional imaging equipment for capturing surface images. The positioning devices include satellite positioning and inertial measurement equipment for acquiring position and attitude information. The original time-series image sequence includes continuous surface image data arranged in chronological order. The digital surface model includes three-dimensional model data reflecting the terrain elevation and surface morphology of the survey area. The flight metadata includes a set of parameters that record the UAV's flight status and equipment operation.

[0194] The observation condition interference separation module retrieves flight metadata, digital surface model, and original time-series image sequence, and performs spatiotemporal registration processing on the three to ensure that the spatiotemporal dimension information of flight metadata, digital surface model, and original time-series image sequence is consistent.

[0195] The observation condition interference separation module extracts observation condition-related information such as solar azimuth angle from flight metadata, and constructs a dynamic shadow projection model by combining it with the terrain undulation features of the digital surface model. It generates a spatiotemporal continuous shadow distribution map, and performs illumination consistency correction on the original time series image sequence based on the map to obtain a radiometrically normalized image.

[0196] The observation condition interference separation module calls the preset multi-scale atmospheric transmission model to perform atmospheric scattering effect elimination processing on the radiation normalized image, generating an atmospheric correction image. Then, an anisotropic diffusion filtering algorithm is used to perform sensor noise suppression processing on the atmospheric correction image, generating an intrinsic reflectance image.

[0197] The observation condition interference separation module includes a specialized processing unit for removing image interference; the spatiotemporal registration module includes a processing method for aligning the spatiotemporal information of multi-source data; the dynamic shadow projection model includes an illumination model that simulates shadow distribution; the multi-scale atmospheric transmission model includes a specialized model that simulates atmospheric scattering effects; the anisotropic diffusion filtering algorithm includes a filtering algorithm that balances noise suppression and edge preservation; and the intrinsic reflectance image includes standardized surface image data after removing observation interference.

[0198] The geological entity recognition module retrieves the pre-trained interpretation model and intrinsic reflectance image, performs standardization preprocessing on the intrinsic reflectance image to adapt it to the input requirements of the pre-trained interpretation model, and then performs multi-scale feature pyramid extraction on the pre-processed intrinsic reflectance image to obtain a spatial-spectral joint feature map.

[0199] The geological entity recognition module uses a geological entity attention mechanism to model the relationship between spatial and spectral joint feature maps, generating a geological context feature map. Then, based on a pre-defined fault-stratum topology rule set, it performs structural constraint segmentation on the geological context feature map to generate an initial geological segmentation map. The module then uses a morphological fault connectivity algorithm to perform topological repair on the fault structures in the initial geological segmentation map, integrating the repaired geological entity recognition and segmentation information to obtain the initial geological interpretation result.

[0200] The geological entity recognition module includes an algorithm processing unit for geological target recognition, a pre-trained interpretation model including a recognition model trained on geological images, a spatial-spectral joint feature map including feature data integrating multi-scale spatial and spectral features, a geological entity attention mechanism including an algorithm mechanism for modeling the relationship between ground features, a fault-stratum topology rule set including topological constraint rules based on geological priors, a morphological fault connection algorithm including a professional algorithm for repairing fault structures, and initial geological interpretation results including preliminary geological entity recognition and segmentation data.

[0201] The interpretation result optimization module combines evaluation criteria such as consistency of geological entity identification and accuracy of boundary delineation to conduct confidence assessment of the initial geological interpretation results, systematically analyze the reliability of interpretation results in each region, and generate confidence assessment results.

[0202] Based on the confidence assessment results, the interpretation result optimization module identifies areas where the interpretation reliability does not meet the preset requirements, triggers the corresponding area's data reacquisition command, and after the reacquisition is completed, it retrieves the reacquisition flight metadata, the reacquisition digital surface model, and the reacquisition image sequence, processes the reacquisition image sequence according to the observation condition interference separation process, and generates optimized data.

[0203] The interpretation result optimization module uses optimized data as a basis to iteratively optimize and fine-tune the parameters of the pre-trained interpretation model, so that the pre-trained interpretation model can be adapted to the geological features of the re-collected area, generate an updated model, and then call the updated model to reinterpret the optimized data. The various geological information obtained from the reinterpretation is integrated, and after being organized according to the standard format of geological maps, the geological maps are output.

[0204] The interpretation result optimization module includes a processing unit for interpretation result evaluation and iterative optimization. The confidence evaluation results include evaluation data reflecting the reliability of the interpretation results. The optimized data includes high-quality image data after interference separation of reacquired data. The updated model includes an interpretation model with improved accuracy after fine-tuning. The geological maps include professional map data that presents geological information in accordance with specifications.

[0205] The observation condition interference separation module 302 is also used for:

[0206] Based on the solar azimuth and digital surface model in the flight metadata, dynamic shadow projection modeling is performed to obtain a spatiotemporal continuous shadow distribution map.

[0207] Illumination consistency correction was performed on the original temporal image sequence using a spatiotemporally continuous shadow distribution map to obtain a radiometrically normalized image.

[0208] Based on a pre-defined multi-scale atmospheric transport model, atmospheric scattering effects are eliminated from the radiation-normalized image to generate an atmospherically corrected image.

[0209] An anisotropic diffusion filtering algorithm is used to suppress sensor noise in atmospheric correction images, resulting in intrinsic reflectance images.

[0210] The observation condition interference separation module 302 is also used for:

[0211] The original temporal image sequence is segmented into shadow regions using a spatiotemporally continuous shadow distribution map to generate shadow region masks.

[0212] Based on a pre-defined solar-sensor geometric constraint model, radiative transmission compensation is performed on the image area covered by the shadow mask to obtain a radiative compensation image.

[0213] Extracting unmasked regions from the original temporal image sequence using shadow region masks;

[0214] A nonlinear photometric alignment algorithm is used to perform spectral consistency fusion between radiometrically compensated images and unmasked area images to generate radiometrically normalized images.

[0215] The observation condition interference separation module 302 is also used for:

[0216] Gradient features are extracted from atmospheric corrected images to obtain gradient magnitude maps;

[0217] Based on the preset edge preservation constraint function, the diffusion coefficient of the gradient magnitude map is calculated to generate an anisotropic diffusion coefficient map.

[0218] Multi-scale diffusion processing of atmospheric correction images is performed using anisotropic diffusion coefficient maps to generate denoised images;

[0219] When the change in the denoised image between adjacent iterations is less than the preset convergence threshold, the intrinsic reflectance image is output.

[0220] The geological entity recognition module 303 is also used for:

[0221] Multi-scale feature pyramid extraction is performed on the intrinsic reflectance image to obtain a spatial-spectral joint feature map. The expression for the spatial-spectral joint feature map is as follows:

[0222]

[0223] in, Represents the joint spatial-spectral feature map. Indicates intrinsic reflectance image. This represents the Gaussian pyramid operator. This represents the convolution operation. Indicates feature cascading, Indicates feature weighting, Represents the spectral weight matrix. This represents the bias vector. Indicates the number of pyramid levels. Indicates a corrected linear unit. Indicates scale index;

[0224] By employing a geological entity attention mechanism, we model land cover associations using the spatial-spectral joint feature map, generating a geological context feature map. The expression for the geological context feature map is as follows:

[0225]

[0226] in, Represents geological context features. Represents the joint spatial-spectral feature map. This represents the Softmax function. Representing feature dimension, Indicates the weight of geological features. This represents the convolution operation. Represents a linear unit with Gaussian error. Represents the query matrix. Indicates the first The key matrix of each attention head. Indicates the first The value matrix of each attention head, Indicates the attention head index, Indicates the total number of attention heads. This represents the matrix transpose operator;

[0227] Based on a pre-defined fault-rock stratum topology rule set, the geological context feature map is segmented by structural constraints to generate an initial geological segmentation map.

[0228] The initial geological interpretation results are obtained by performing topological repair on the fracture structures in the initial geological segmentation map using a morphological fault connection algorithm.

[0229] The geological entity recognition module 303 is also used for:

[0230] The fracture structure framework is extracted from the initial geological segmentation map to obtain the fracture framework line map;

[0231] Based on the preset fracture ductility criterion, spatial attraction domain analysis is performed on the fracture endpoints in the fracture skeleton line diagram to generate candidate fracture connection point pairs. The expression for the candidate fracture connection point pairs is as follows:

[0232]

[0233] in, Indicates a pair of candidate points for broken connections. This represents the distance vector between endpoints. Indicates the directional consistency threshold. Represents the angle between vectors. Indicates the maximum acceptable angle. Indicates the coordinates of the fracture endpoints. Indicates endpoint The direction vector at that location, Indicates endpoint The direction vector at that location, Indicates endpoint The structural stress vector at that location, Indicates endpoint The structural stress vector at the location;

[0234] An adaptive morphological path reconstruction algorithm is used to perform geological topological connections on candidate fracture connection points to generate a fracture network map for repair.

[0235] The non-fractured areas in the repaired fault network map and the initial geological segmentation map are merged to generate the initial geological interpretation results.

[0236] The geological entity recognition module 303 is also used for:

[0237] Using a fault orientation constraint segmentation algorithm, the fault structures in the geological context feature map are segmented with consistent orientation to generate a fault region map.

[0238] Based on the rock layer contact relationship rules, the lithological boundaries in the geological context feature map are segmented by attitude constraints to generate a lithological unit map;

[0239] A topological conflict resolution algorithm is used to fuse the fault region map and lithological unit map to achieve geological rule consistency, resulting in an initial geological segmentation map. The expression for the initial geological segmentation map is as follows:

[0240]

[0241] in, This represents the initial geological segmentation map. Map showing the fractured region. Represents a lithological unit diagram. This represents the topological conflict resolution operator. Indicates feature fusion operation, Indicates the conflict correction factor. Indicates the conflict indicator function. Represents the set of geological boundaries. Represents the set of fracture boundaries. This represents the set of lithological boundaries.

[0242] In one embodiment, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0243] In one embodiment, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described method embodiments.

[0244] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0245] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A regional geological survey method based on unmanned aerial vehicles (UAVs), characterized in that, The method includes: The original time-series image sequence, digital surface model and flight metadata of the survey area are acquired simultaneously through the optical sensors and positioning devices carried by the UAV. Based on the flight metadata and the digital surface model, observation condition interference separation is performed on the original time-series image sequence to generate intrinsic reflectance images; The geological entity identification is performed on the intrinsic reflectance image using a pre-trained interpretation model to obtain the initial geological interpretation results. A confidence assessment is performed based on the initial geological interpretation results to generate a confidence assessment result. Data reacquisition is triggered based on the confidence assessment result. Based on the reacquisitioned flight metadata and digital surface model, observation condition interference is separated from the reacquisitioned image sequence to generate optimized data. The pre-trained interpretation model is fine-tuned based on the optimized data to generate an updated model. The updated model is used to reinterpret the optimized data to output geological maps.

2. The regional geological survey method based on unmanned aerial vehicles according to claim 1, characterized in that, The process of separating observation condition interference from the original time-series image sequence based on the flight metadata and the digital surface model to generate intrinsic reflectance images includes: Based on the solar azimuth angle in the flight metadata and the digital surface model, dynamic shadow projection modeling is performed to obtain a spatiotemporal continuous shadow distribution map; The original temporal image sequence is subjected to illumination consistency correction using the spatiotemporal continuous shadow distribution map to obtain a radiometrically normalized image. Based on a preset multi-scale atmospheric transport model, atmospheric scattering effects are eliminated from the radiation-normalized image to generate an atmospherically corrected image. The atmospheric correction image is subjected to sensor noise suppression using an anisotropic diffusion filtering algorithm to obtain the intrinsic reflectance image.

3. The regional geological survey method based on unmanned aerial vehicles according to claim 2, characterized in that, The step of performing illumination consistency correction on the original temporal image sequence using the spatiotemporal continuous shadow distribution map to obtain a radiometrically normalized image includes: The original temporal image sequence is segmented into shadow regions using the spatiotemporal continuous shadow distribution map to generate a shadow region mask. Based on a preset solar-sensor geometric constraint model, radiative transmission compensation is performed on the image area covered by the shadow area mask to obtain a radiative compensation image. Unmasked region images are extracted from the original temporal image sequence using the shadow region mask; The radiometrically compensated image and the unmasked area image are spectrally consistent fused using a nonlinear photometric alignment algorithm to generate the radiometrically normalized image.

4. The regional geological survey method based on unmanned aerial vehicles according to claim 2, characterized in that, The process of performing sensor noise suppression on the atmospheric correction image using an anisotropic diffusion filtering algorithm to obtain the intrinsic reflectance image includes: Gradient feature extraction is performed on the atmospheric correction image to obtain a gradient magnitude map; Based on the preset edge preservation constraint function, the diffusion coefficient of the gradient magnitude map is calculated to generate an anisotropic diffusion coefficient map. The atmospheric correction image is subjected to multi-scale diffusion processing using the anisotropic diffusion coefficient map to generate a denoised image. When the change in the adjacent iterations of the denoised image is less than a preset convergence threshold, the intrinsic reflectance image is output.

5. The regional geological survey method based on unmanned aerial vehicles according to claim 1, characterized in that, The process of using a pre-trained interpretation model to identify geological entities in the intrinsic reflectance image to obtain initial geological interpretation results includes: Multi-scale feature pyramid extraction is performed on the intrinsic reflectance image to obtain a spatial-spectral joint feature map. The expression of the spatial-spectral joint feature map is as follows: in, Represents the joint spatial-spectral feature map. Indicates intrinsic reflectance image. This represents the Gaussian pyramid operator. This represents the convolution operation. Indicates feature cascading, Indicates feature weighting, Represents the spectral weight matrix. This represents the bias vector. Indicates the number of pyramid levels. Indicates a modified linear unit. Indicates scale index; By employing a geological entity attention mechanism, land cover association modeling is performed on the spatial-spectral joint feature map to generate a geological context feature map. The expression for the geological context feature map is as follows: in, Represents geological context features. Represents the joint spatial-spectral feature map. This represents the Softmax function. Representing feature dimension, Indicates the weight of geological features. This represents the convolution operation. Represents a linear unit with Gaussian error. Represents the query matrix. Indicates the first The key matrix of each attention head. Indicates the first The value matrix of each attention head, Indicates the attention head index, Indicates the total number of attention heads. This represents the matrix transpose operator; Based on a preset fault-rock stratum topology rule set, the geological context feature map is segmented by structural constraints to generate an initial geological segmentation map; The initial geological interpretation results are obtained by performing topological repair on the fracture structures in the initial geological segmentation map using a morphological fault connection algorithm.

6. The regional geological survey method based on unmanned aerial vehicles according to claim 5, characterized in that, The initial geological interpretation results are obtained by performing topological repair on the fault structures in the initial geological segmentation map using a morphological fault connection algorithm, including: The initial geological segmentation map is subjected to fracture structure skeleton extraction to obtain fracture skeleton line map; Based on a preset fracture ductility criterion, spatial attraction domain analysis is performed on the fracture endpoints in the fracture skeleton diagram to generate candidate fracture connection point pairs. The expression for the candidate fracture connection point pairs is as follows: in, Indicates a pair of candidate points for broken connections. This represents the distance vector between endpoints. Indicates the directional consistency threshold. Represents the angle between vectors. Indicates the maximum acceptable angle. Indicates the coordinates of the fracture endpoints. Endpoints The direction vector at that location, Indicates endpoint The direction vector at that location, Indicates endpoint The structural stress vector at that location, Indicates endpoint The structural stress vector at the location; An adaptive morphological path reconstruction algorithm is used to perform geological topological connections on the candidate fracture connection point pairs to generate a repair fracture network diagram. The non-fractured areas in the repaired fracture network map and the initial geological segmentation map are merged to generate the initial geological interpretation result.

7. The regional geological survey method based on unmanned aerial vehicles according to claim 5, characterized in that, The method, based on a preset fault-stratum topology rule set, performs structural constraint segmentation on the geological context feature map to generate an initial geological segmentation map, including: Using a fault strike constraint segmentation algorithm, the fault structures in the geological context feature map are segmented with consistent orientation to generate a fault region map. Based on the rock strata contact relationship rules, the lithological boundaries in the geological context feature map are segmented by attitude constraints to generate a lithological unit map; The initial geological segmentation map is obtained by performing geological rule consistency fusion on the fracture region map and the lithological unit map through a topological conflict resolution algorithm.

8. A regional geological survey system based on unmanned aerial vehicles (UAVs), characterized in that, The system includes: The UAV data acquisition module is used to simultaneously acquire the original time-series image sequence, digital surface model and flight metadata of the survey area through the optical sensors and positioning devices carried by the UAV. The observation condition interference separation module is used to separate the observation condition interference from the original time-series image sequence based on the flight metadata and the digital surface model, and generate intrinsic reflectance images. The geological entity recognition module is used to identify geological entities in the intrinsic reflectance image through a pre-trained interpretation model to obtain initial geological interpretation results. The interpretation result optimization module is used to perform confidence assessment based on the initial geological interpretation results and generate a confidence assessment result; trigger data reacquisition based on the confidence assessment result; perform observation condition interference separation on the reacquisition image sequence based on the reacquisition flight metadata and digital surface model to generate optimized data; fine-tune the pre-trained interpretation model based on the optimized data to generate an updated model; and reinterpret the optimized data using the updated model to output geological maps.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the UAV-based regional geological survey method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the UAV-based regional geological survey method as described in any one of claims 1 to 7.

Images