Deep Learning Geological Disaster Intelligent Identification Method and System Based on UAV Remote Sensing Data

Abstract

The present invention discloses a deep learning-based intelligent geological disaster identification method and system for drone remote sensing data, which relates to the technical field of intelligent disaster identification. The key points of its technical solution include the following steps: using the remote sensing equipment carried by the drone based on the flight position of the drone to obtain the remote sensing data to be identified in different geological regions, analyzing and screening the remote sensing data to be identified to obtain the remote sensing data to be identified with suspected geological disaster characteristics and marking them as abnormal data, and matching the corresponding geological disaster type information from the geological disaster information database according to the abnormal data; dividing the geological region where the abnormal data is located into a geological horizontal region and a geological vertical region; processing and analyzing the change status of the topographic and geomorphic features in the geological horizontal region to obtain horizontal influence feature data; the effect is that this application can improve the judgment accuracy of geological disasters.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent disaster identification, and more specifically, to an intelligent geological disaster identification method and system based on deep learning of unmanned aerial vehicle (UAV) remote sensing data. Background Art

[0002] With the rapid development of the economic society, human engineering activities have become increasingly frequent. Such as large-scale urban construction, mining, road construction, etc., which have had a significant impact on the geological environment and increased the probability of geological disasters. At the same time, global climate change has led to an increase in extreme climate events, and inducing factors such as heavy rain and earthquakes have also made the geological disaster situation increasingly severe.

[0003] Geological disasters such as landslides, debris flows, and ground collapses not only seriously threaten people's lives and property safety, but also cause huge damage to the ecological environment and infrastructure. Accurately and timely identifying and warning geological disasters is of great importance. Traditional geological disaster identification methods mainly rely on manual field surveys, simple instrument monitoring, etc. Manual field surveys are limited by conditions such as terrain and traffic, and it is difficult to conduct comprehensively and timely in mountainous areas, remote areas, etc., and the efficiency is low and the labor cost is high; the spatial coverage of simple instrument monitoring is limited and the data acquisition is discontinuous, so it is difficult to conduct real-time and dynamic monitoring of large areas.

[0004] In recent years, remote sensing technology has been widely used in the field of geological disaster monitoring due to its advantages of large-scale, periodic, and non-contact data acquisition. However, the original remote sensing data is easily interfered by various factors such as topography, geological structure, and atmospheric conditions, resulting in problems such as noise and distortion in the data, which affects the accuracy of geological disaster identification. At the same time, relying solely on remote sensing data for geological disaster identification lacks the comprehensive analysis of multi-source information such as geological structure and physical properties of rock and soil masses, so it is difficult to accurately judge and identify geological disasters. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technology, the purpose of the present invention is to provide an intelligent geological disaster identification method and system based on deep learning of unmanned aerial vehicle (UAV) remote sensing data.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] An intelligent geological disaster identification method based on deep learning of unmanned aerial vehicle (UAV) remote sensing data, the method includes the following steps:

[0008] S1. Based on the flight position of the drone, use the remote sensing equipment carried by the drone to obtain the remote sensing data to be identified in different geological regions. Analyze and screen the remote sensing data to be identified to obtain the remote sensing data to be identified that is suspected to have geological disaster characteristics, mark it as abnormal data, and match the corresponding geological disaster type information from the geological disaster information database;

[0009] S2. Divide the geological region where the abnormal data is located into a geological horizontal region and a geological vertical region;

[0010] The geological horizontal region here refers to the geological scope extended in the horizontal direction, often involving the distribution of geological bodies, topographic and geomorphic features, etc. on the same altitude or similar altitude levels, along the horizontal directions such as east-west and north-south. For example, the distribution range of different rock layers in the horizontal direction, the extension of the mountain range trend on the horizontal plane, etc.

[0011] The geological vertical region refers to the geological scope in the vertical direction, that is, the part extending from the surface to the deep underground, covering the stratigraphic structure at different depths, the change of rock types, the vertical manifestation of geological structures, etc. For example, the depth and characteristic differences of different sedimentary layers and bedrock layers from the surface downwards.

[0012] S3. Process and analyze the change situation of the topographic and geomorphic features in the geological horizontal region to obtain the horizontal influence feature data;

[0013] S4. Process and analyze the change situation of the topographic and geomorphic features in the geological vertical region to obtain the vertical influence feature data;

[0014] S5. Compare the horizontal change stability of the horizontal influence feature data and the vertical change stability of the vertical influence feature data, and then select a geological position set in the geological horizontal region or the geological vertical region;

[0015] S6. Extract the remote sensing data to be identified corresponding to the geological position set, and perform interference removal processing on the remote sensing data to be identified corresponding to the geological position set to obtain a remote sensing data processing set;

[0016] S7. Judge whether there is a geological disaster in the geological region according to the geological disaster type information and the remote sensing data processing set.

[0017] Preferably, analyze and screen the remote sensing data to obtain the data suspected to have geological disaster characteristics and mark it as abnormal data. Specifically:

[0018] Compare the texture features of the remote sensing data to be identified with the texture features of the preset normal geological remote sensing data, and screen out the remote sensing data to be identified whose texture features are different from those of the preset normal geological remote sensing data, and mark it as abnormal data.

[0019] Preferably, step S3 specifically includes:

[0020] Combining the topographic and geomorphic feature change conditions and historical topographic and geomorphic data within the geological horizontal region to predict the change trend of the geological structure within the geological horizontal region, obtaining the horizontal geological structure change trend;

[0021] Predicting the change of the physical properties of the rock and soil mass within the geological horizontal region based on the topographic and geomorphic feature change conditions and the horizontal geological structure change trend, obtaining the horizontal rock and soil mass physical property change trend;

[0022] Combining the horizontal geological structure change trend and the horizontal rock and soil mass physical property change situation into horizontal influence feature data.

[0023] Preferably, step S4 specifically includes:

[0024] Combining the topographic and geomorphic feature change conditions and historical topographic and geomorphic data within the geological longitudinal region to predict the change trend of the geological structure within the geological longitudinal region, obtaining the longitudinal geological structure change trend;

[0025] Predicting the change of the physical properties of the rock and soil mass within the geological longitudinal region based on the topographic and geomorphic feature change conditions and the longitudinal geological structure change trend, obtaining the longitudinal rock and soil mass physical property change trend;

[0026] Combining the longitudinal geological structure change trend and the longitudinal rock and soil mass physical property change situation into longitudinal influence feature data.

[0027] Preferably, step S5 specifically includes:

[0028] When the horizontal change stability of the horizontal influence feature data is not lower than the longitudinal change stability of the longitudinal influence feature data, a first position locking region is divided within the geological horizontal region with the projection point of the UAV flight position on the ground as the origin, and a geological position set is selected from the first position locking region according to the abnormal position points corresponding to the abnormal data;

[0029] When the horizontal change stability of the horizontal influence feature data is lower than the longitudinal change stability of the longitudinal influence feature data, a second position locking region is divided within the geological longitudinal region with the projection point of the UAV flight position on the ground as the symmetry point, and a geological position set is selected from the second position locking region according to the abnormal position points corresponding to the abnormal data;

[0030] Wherein, the geological position set includes at least three analysis position points.

[0031] Preferably, step S6 specifically includes:

[0032] Extract the remote sensing data to be identified corresponding to the set of geological positions in the lateral geological region. Estimate the first data interference coefficient based on the changes in the lateral topographic and geomorphic features and the lateral influence feature data. Adjust and process this remote sensing data according to the first data interference coefficient to obtain the first type of adjusted remote sensing data set;

[0033] Extract the remote sensing data to be identified corresponding to the set of geological positions in the longitudinal geological region. Estimate the second data interference coefficient based on the changes in the longitudinal topographic and geomorphic features and the longitudinal influence feature data. Adjust and process this remote sensing data according to the second data interference coefficient to obtain the second type of adjusted remote sensing data set;

[0034] Among them, the remote sensing data processing set includes the first type of adjusted remote sensing data set and the second type of adjusted remote sensing data set.

[0035] Preferably, judge whether there is a geological disaster in the geological region according to the geological disaster type information and the remote sensing data processing set, which specifically includes the following steps:

[0036] Extract the historical remote sensing data of the same type of geological disasters at different location points in the historical period and the corresponding historical geological disaster actual situation information. Establish a deep learning geological disaster recognition model according to the historical remote sensing data and the historical geological disaster actual situation information;

[0037] Input the geological disaster type information and the first type of adjusted remote sensing data set into the deep learning geological disaster recognition model to obtain the first type of disaster recognition result set; Input the geological disaster type information and the second type of adjusted remote sensing data set into the deep learning geological disaster recognition model to obtain the second type of disaster recognition result set;

[0038] Judge whether there is a geological disaster in the geological region through the first type of disaster recognition result set and the second type of disaster recognition result set.

[0039] Preferably, judge whether there is a geological disaster in the geological region through the first type of disaster recognition result set and the second type of disaster recognition result set, which specifically includes the following steps:

[0040] If each disaster recognition result in the first type of disaster recognition result set shows the existence of a disaster, it is determined that there is a geological disaster in this geological region. If each disaster recognition result in the second type of disaster recognition result set shows the existence of a disaster, it is determined that there is a geological disaster in this geological region;

[0041] If there is one disaster recognition result in the first type of disaster recognition result set that shows the non-existence of a disaster, it is determined that there is no geological disaster in this geological region. If there is one disaster recognition result in the second type of disaster recognition result set that shows the non-existence of a disaster, it is determined that there is no geological disaster in this geological region.

[0042] The intelligent geological disaster identification system based on UAV remote sensing data further includes:

[0043] A marker matching module, which uses the remote sensing equipment carried by the UAV based on the flight position of the UAV to obtain the remote sensing data to be identified in different geological regions, analyzes and screens the remote sensing data to be identified to obtain the remote sensing data to be identified with suspected geological disaster characteristics and marks it as abnormal data, and matches the corresponding geological disaster type information from the geological disaster information database according to the abnormal data;

[0044] A division module, which divides the geological region where the abnormal data is located into a geological horizontal region and a geological vertical region;

[0045] A first processing and analysis module, which processes and analyzes the change status of the topographic and geomorphic features in the geological horizontal region to obtain horizontal influence feature data;

[0046] A second processing and analysis module, which processes and analyzes the change status of the topographic and geomorphic features in the geological vertical region to obtain vertical influence feature data;

[0047] A comparison and selection module, which compares the horizontal change stability of the horizontal influence feature data and the vertical change stability of the vertical influence feature data, and then selects a geological position set in the geological horizontal region or the geological vertical region;

[0048] A processing module, which extracts the remote sensing data to be identified corresponding to the geological position set, and performs interference removal processing on the remote sensing data to be identified corresponding to the geological position set to obtain a remote sensing data processing set;

[0049] A judgment module, which judges whether there is a geological disaster in the geological region according to the geological disaster type information and the remote sensing data processing set.

[0050] Preferably, an electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the intelligent geological disaster identification method based on UAV remote sensing data.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] By finely comparing the texture features of the remote sensing data to be identified with the texture features of the preset normal geological remote sensing data, this application can accurately screen out the abnormal data with suspected geological disaster characteristics. This method greatly improves the acquisition efficiency of geological disaster data and the accuracy of screening compared with traditional manual investigation or single means monitoring, laying a solid foundation for accurately judging geological disasters subsequently.

[0053] The geological area where abnormal data is located is innovatively divided into a geological horizontal area and a geological vertical area, and in-depth analysis is carried out from multiple dimensions. During the analysis of the horizontal and vertical areas, not only the current topographic and geomorphic feature changes are combined, but also historical topographic and geomorphic data is incorporated to predict the trend of geological structure changes; at the same time, according to the topographic and geomorphic feature changes and the predicted geological structure change trend, the trend of physical property changes of rock and soil masses is further predicted. This multi-dimensional and comprehensive analysis method comprehensively and meticulously obtains geological area information, can capture potential geological hazard risk factors that are easily overlooked by single-dimensional analysis, thus more accurately judging the possibility of geological disasters occurring and enhancing the reliability of geological disaster early warning.

[0054] When processing the remote sensing data to be identified corresponding to the geological position set, the interference factors existing in the actual environment are fully considered. By scientifically estimating the data interference coefficient according to the topographic and geomorphic feature changes and influence feature data of each region, and carrying out targeted adjustment processing on the corresponding remote sensing data according to the interference coefficient. This anti-interference processing method effectively reduces the interference of external factors on remote sensing data, makes the data entering the subsequent analysis link more pure and accurate, ensures the reliability of geological disaster judgment based on these data, and avoids misjudgment or missed judgment caused by data interference.

[0055] Based on a large amount of historical remote sensing data of the same type of geological disasters at different location points in historical periods and the corresponding actual geological disaster situation information, a deep learning geological disaster identification model is constructed. The processed remote sensing data set and geological disaster type information are input into this model for intelligent analysis to obtain a disaster identification result set. This intelligent judgment method based on big data and deep learning can simulate and learn complex geological disaster patterns, and is more scientific and objective than traditional empirical judgment, thus being able to accurately judge whether there are geological disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic flowchart of the deep learning geological disaster intelligent identification method based on UAV remote sensing data proposed by the present invention;

[0057] Figure 2 It is a schematic diagram of the modules of the deep learning geological disaster intelligent identification system based on UAV remote sensing data proposed by the present invention;

[0058] Figure 3 It is a schematic structural diagram of the electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0059] The embodiments further illustrate the deep learning geological disaster intelligent identification method and system based on UAV remote sensing data proposed by the present invention.

[0060] As Figure 1 shown, a deep learning-based intelligent geological disaster identification method for drone remote sensing data, the method comprising the following steps:

[0061] S1. Based on the flight position of the drone, use the remote sensing equipment carried by the drone to obtain the remote sensing data to be identified in different geological regions, analyze and screen the remote sensing data to be identified to obtain the remote sensing data to be identified suspected of having geological disaster characteristics and mark them as abnormal data, and match the corresponding geological disaster type information from the geological disaster information database according to the abnormal data;

[0062] S2. Divide the geological region where the abnormal data is located into a geological horizontal region and a geological vertical region;

[0063] S3. Process and analyze the change status of the topographic and geomorphic features in the geological horizontal region to obtain horizontal influence feature data;

[0064] S4. Process and analyze the change status of the topographic and geomorphic features in the geological vertical region to obtain vertical influence feature data;

[0065] S5. After comparing the horizontal change stability of the horizontal influence feature data and the vertical change stability of the vertical influence feature data, select a geological position set in the geological horizontal region or the geological vertical region;

[0066] S6. Extract the remote sensing data to be identified corresponding to the geological position set, and perform interference removal processing on the remote sensing data to be identified corresponding to the geological position set to obtain a remote sensing data processing set;

[0067] S7. Determine whether there is a geological disaster in the geological region according to the geological disaster type information and the remote sensing data processing set.

[0068] The drone in this application can flexibly fly over different geological regions. Taking its own flight position as a reference, it uses the carried remote sensing equipment to collect data from different geological regions below according to the set parameters and scanning modes, and obtains the remote sensing data to be identified. These data contain various information such as the topography, geomorphology, and surface texture of the geological region. Subsequently, analyze and screen the remote sensing data to be identified to obtain the remote sensing data to be identified suspected of having geological disaster characteristics and mark them as abnormal data. If there is abnormal data, it indicates that there may be geological disaster characteristics in this region.

[0069] In the process of constructing the geological disaster information database, first widely collect various data related to geological disasters through multiple channels, including but not limited to: conducting field investigations in different regions to record the topographic changes (slope, elevation difference, etc.), geotechnical properties (soil type, rock hardness, etc.) of landslides, the gully morphology and deposit characteristics of debris flow occurrence areas, the stratigraphic structure and collapse depth of ground collapse areas, and other detailed information.

[0070] Use various monitoring devices (such as displacement sensors, inclinometers, rain gauges, etc.) to collect data on the displacement, tilt angle, rainfall, etc. of geological bodies in real-time or regularly. These data can reflect the dynamic changes of geological bodies and are crucial for judging whether geological disasters may occur.

[0071] Collect information on geological disaster events that have occurred in history, including information such as the time, location, scale, impact range, and losses caused by the disasters. At the same time, organize relevant environmental data before and after the disasters, such as meteorological conditions, seismic activities, etc.

[0072] Classify and store the collected data according to different types of geological disasters (such as landslides, debris flows, collapses, ground subsidence, ground fissures, etc.). Further subdivide the data dimensions for each disaster type. For example, landslide disasters include morphological parameters (length, width, thickness) of the landslide body, material composition (rock-soil ratio), inducing factors (rainfall, earthquake, human engineering activities, etc.). At the same time, encode the data for rapid retrieval and identification by computers.

[0073] Select a suitable database management system (such as relational databases MySQL, Oracle, etc.), and build a geological disaster information database based on the organized data structure. Establish data indexes to optimize query performance and ensure that data can be accessed and called quickly and accurately.

[0074] After obtaining abnormal data suspected of having geological disaster characteristics, it is necessary to extract its characteristics for matching with the information database. Use image processing technology to analyze the texture information of remote sensing images in the abnormal data. For example, the texture of the landslide disaster area appears rough and irregular, forming a sharp contrast with the smooth texture of the surrounding normal terrain; the debris flow area shows striped or mixed texture characteristics. Quantitatively describe the texture characteristics by calculating parameters such as the roughness, contrast, and directionality of the texture.

[0075] Extract topographic and geomorphic related characteristics from the abnormal data, such as slope, aspect, elevation change, etc. For example, landslides usually occur in areas with steeper slopes. Judge whether it meets the topographic conditions for landslide occurrence by calculating parameters such as the average slope and maximum slope of the abnormal area. For ground subsidence, there may be a sudden decrease in local elevation, and such characteristics can be captured by analyzing elevation data.

[0076] If the abnormal data contains information related to rock and soil bodies (such as underground rock and soil body structure data obtained by means of geological radar, etc.), analyze the characteristics of the type (such as clay, sand, rock, etc.), thickness, density, etc. of the rock and soil bodies. Different types of geological disasters are closely related to the characteristics of rock and soil bodies. For example, loose sand is more likely to trigger debris flow disasters, while thicker soft soil layers may increase the risk of ground subsidence.

[0077] Combine with abnormal data to obtain environmental information, such as rainfall, seismic activity, and the intensity of human engineering activities, etc. For example, during periods of heavy rainfall, the possibility of rainfall-induced landslides, debris flows and other disasters needs to be considered. If there are recent seismic activity records in this area, attention should be paid to whether the impact of the earthquake on the stability of geological bodies may trigger disasters such as collapses.

[0078] Match the extracted abnormal data features with the data in the geological disaster information database, and use a variety of similarity calculation methods, such as Euclidean distance, cosine similarity, etc., to calculate the similarity between the abnormal data features and the typical features of various geological disasters in the information database. For example, for landslide disasters, compare the features such as slope and texture roughness of the abnormal data with the typical feature parameters of landslides in the information database, and calculate the similarity value between the two.

[0079] Sort the types of geological disasters in the information database according to the similarity calculation results. The higher the similarity, the higher the ranking of the disaster type. For example, if the similarity between the abnormal data and the features of landslide disasters is calculated to be the highest, followed by debris flows, then landslides rank first and debris flows rank second in the sorting results.

[0080] Set a similarity threshold. When the similarity of a certain type of geological disaster exceeds this threshold, it is considered that the abnormal data matches successfully with this type of geological disaster, and it is determined as the corresponding geological disaster type information. For example, set the threshold to 0.7. If the similarity calculation result of the landslide disaster is 0.8, then it is determined that the geological disaster type corresponding to the abnormal data is a landslide.

[0081] The system divides the geological area where the abnormal data is located into a geological horizontal area and a geological vertical area according to the spatial dimension. When analyzing the geological horizontal area, collect real-time data on the changes in topographic and geomorphic features in the current area, such as changes in surface slope and rock exposure degree, etc.; at the same time, call historical topographic and geomorphic data to predict the changing trend of geological structures in this area through methods such as time series analysis. For example, whether there are signs of horizontal movement or deformation in the strata. Then, according to the changes in topographic and geomorphic features and the predicted changing trend of geological structures, predict the changing trend of the physical properties of horizontal rock and soil masses, such as changes in parameters such as the compressive strength and shear strength of rock and soil masses, and integrate the changing trend of horizontal geological structures and the physical properties of horizontal rock and soil masses to obtain horizontal impact feature data. Similarly, in the geological vertical area, according to the same process and method, collect real-time data on the changes in longitudinal topographic and geomorphic features and combine historical data to predict the changing trend of longitudinal geological structures to obtain the changing trend of the physical properties of longitudinal rock and soil masses, and finally combine to form longitudinal impact feature data.

[0082] Quantify and compare the change stabilities of both the lateral influence feature data and the vertical influence feature data, so as to select a set of geological positions within the geological lateral region or the geological vertical region.

[0083] Respectively extract the remotely sensed data to be identified corresponding to the set of geological positions within the geological lateral region and the geological vertical region. In the geological lateral region, estimate the first data interference coefficient according to the change condition of the lateral topographic and geomorphic features and the lateral influence feature data. Based on the first data interference coefficient, adjust and process the remotely sensed data to be identified in the geological lateral region through data processing algorithms such as filtering and correction. After removing or reducing the influence of interference factors, obtain the first type of adjusted remotely sensed data set. Using the same principle and method, estimate the second data interference coefficient in the geological vertical region according to the change condition of the vertical topographic and geomorphic features and the vertical influence feature data, and adjust and process the remotely sensed data in the geological vertical region to obtain the second type of adjusted remotely sensed data set. The two types of adjusted remotely sensed data sets together constitute the remotely sensed data processing set, providing high-quality data support for subsequent geological disaster judgment, so as to facilitate judging whether there are geological disasters in the geological region according to the geological disaster type information and the remotely sensed data processing set.

[0084] Analyze and screen the remotely sensed data to obtain data suspected of having geological disaster characteristics and mark them as abnormal data. Specifically:

[0085] Compare the texture features of the remotely sensed data to be identified with the texture features of the preset normal geological remotely sensed data, and screen out the remotely sensed data to be identified whose texture features are different from those of the preset normal geological remotely sensed data and mark them as abnormal data.

[0086] In the field of geological disaster monitoring, remote sensing technology can obtain a large amount of data on the surface conditions. Pre-collect and organize the remotely sensed data under normal geological conditions and extract their texture features to construct a texture feature library of preset normal geological remotely sensed data. These normal texture features reflect the texture patterns presented by the geological surface on the remote sensing image in the absence of geological disasters, such as the texture details and distribution rules of rocks, soil, vegetation, etc. under normal conditions.

[0087] Extract the texture features of the remotely sensed data to be identified, and then carefully compare these newly extracted texture features with the texture features of the preset normal geological remotely sensed data. Consider multiple parameters such as the shape, direction, density, and repeatability of the texture during the comparison process. For example, the texture of a normal mountain surface presents a certain regular pattern on the remote sensing image. If a geological disaster such as a landslide occurs, the mountain surface is damaged and the texture features will change significantly.

[0088] When the texture features of the remotely sensed data to be recognized show differences beyond a certain threshold in some parameters from the preset normal texture features, it is determined that the remotely sensed data to be recognized is abnormal. Such differences may be manifested as texture disorders, absences, newly added specific texture patterns, etc. If a difference is determined to exist, the remotely sensed data to be recognized is screened out and marked as abnormal data, and these abnormal data are considered as data suspected of having geological disaster characteristics.

[0089] Step S3 specifically includes:

[0090] Combining the changes in topographic and geomorphic features within the lateral geological region and historical topographic and geomorphic data to predict the changing trend of geological structures within the lateral geological region to obtain the lateral geological structure change trend;

[0091] Based on the changes in topographic and geomorphic features and the lateral geological structure change trend, predicting the changes in the physical properties of rock and soil masses within the lateral geological region to obtain the lateral rock and soil mass physical property change trend;

[0092] Combining the lateral geological structure change trend and the lateral rock and soil mass physical property change situation into lateral influence feature data.

[0093] This application uses optical sensors carried by satellites or aircraft to obtain surface reflected light information in different bands to form images, and obtains the changes in topographic and geomorphic features through multi - period image comparison, such as changes in vegetation coverage, water body range changes, and subtle deformations on the mountain surface. For example, when monitoring a landslide area, the differences in vegetation coverage range and mountain contour on the images before and after the landslide can be clearly captured.

[0094] Interferometric synthetic aperture radar (InSAR) technology emits radar waves to the ground and receives the echoes, and obtains surface micro - deformation information according to the echo phase difference to monitor millimeter - level terrain displacement. Therefore, it can be used to monitor the changes in topographic and geomorphic features such as fault activities, ground subsidence, and landslides.

[0095] Reasonably arrange GPS monitoring stations within the lateral geological region, continuously record the three - dimensional coordinate information of the stations, and obtain the surface displacement data of the locations where the stations are located, thereby reflecting the slow changes in regional topographic and geomorphic features, such as the uplift or subsidence of the terrain caused by plate movement.

[0096] Use drones equipped with high - resolution cameras, lidar and other devices to obtain regional topographic and geomorphic data at low altitude. It can be flexibly targeted at small - scale key areas for high - precision mapping, and quickly generate results such as digital orthophoto maps (DOM) and digital elevation models (DEM). By comparing the UAV aerial survey results at different times, the local changes in topographic and geomorphic features, such as small - scale mountain collapses and gully development, can be detailedly grasped.

[0097] Use a three-dimensional laser scanner to scan the geological horizontal area and quickly obtain high-precision three-dimensional point cloud data of the ground surface. It can accurately restore the details of the terrain and landforms, such as the texture of the rock surface, the height and slope of the steep bank, etc. Comparing the data of multiple scans can clearly present the short-term changes in the terrain and landforms, which is often used to monitor the terrain changes in areas such as open-pit mines and road slopes.

[0098] The Geographic Information System (GIS) can uniformly store and manage topographic and geomorphic data from different sources and in different formats (such as current high-precision topographic survey data, digitized data of historical topographic maps) and geological-related data (such as the distribution of geological faults, formation lithology data). By overlaying the topographic and geomorphic layers of different periods, it can intuitively display the changes in topographic and geomorphic features, such as changes in terrain elevation, movement of the boundaries of geomorphic units, etc. For example, by comparing the contour layers of different years, the erosion or accumulation of the mountain body can be clearly seen, and then it can assist in judging the shaping effect of geological tectonic activities on the terrain.

[0099] Establish a geological structure change model in combination with geological structure theories (such as the plate tectonics theory). Input the change data of topographic and geomorphic features and historical data to simulate the evolution process of geological structures on different time scales, and then predict the future trend of geological structure changes. For example, according to the relative displacement amount and direction of the terrain on both sides of the fault, simulate the future activity trend of the fault.

[0100] Optical remote sensing images can reflect the macroscopic geomorphic features of the ground surface. By automatically classifying and identifying the changes in topographic and geomorphic elements in remote sensing images of different periods through computer, such as river course changes, changes in the location of landslides, etc. These changes may be related to geological tectonic activities. For example, newly emerged linear landforms may be manifestations of fault activities. Radar remote sensing (such as InSAR technology) can accurately measure the minute deformations of the ground surface and obtain terrain displacement information at the millimeter level. By analyzing multi-period InSAR data, the activities of geological structures such as faults and folds can be monitored, and combined with historical topographic and geomorphic data, predict their future change trends.

[0101] Fuse remote sensing data with other geological data (such as gravity and magnetic data) to comprehensively judge the characteristics of geological structures. For example, combine the density differences of underground substances reflected by gravity data and the topographic changes shown in remote sensing images to infer the changes in underground geological structures, and then predict the trend of geological structure changes.

[0102] Establish a Global Positioning System (GPS) monitoring network within the geological horizontal area to continuously monitor the changes in the three-dimensional coordinates of the points. After summarizing the displacement speed and direction of the points through the long-term accumulated high-precision coordinate data, determine the crustal movement situation of the area, and combine historical topographic and geomorphic data to infer the activity intensity and trend of geological structures. For example, the displacement data monitored by GPS stations near the plate boundary can help predict the geological structure changes caused by the relative movement of the plates.

[0103] Obtain the change data of the topographic elevation in the area through high-precision leveling. Repeatedly measure the elevation of the leveling points at different times to obtain the elevation change amount, so as to judge the vertical movement of the crust in combination with the historical topographic and geomorphic data, and then predict the change trend of the geological structure. For example, the leveling measurement data can provide important basis when studying the relationship between regional subsidence or uplift and fault activities.

[0104] Based on the principles of geomechanics, use finite element software to establish a numerical model of the geological structure. Input the change data of topographic and geomorphic features, rock mechanics parameters (such as elastic modulus, Poisson's ratio), and historical geological structure activity information to simulate the deformation and evolution process of the geological structure under stress. By adjusting the model parameters and boundary conditions, predict the change trend of the geological structure in different scenarios to provide reference for geological disaster assessment and engineering construction.

[0105] For geological structures composed of discrete media (such as loose rock masses), the discrete element method is used for simulation. Discretize the geological body into multiple units, consider the interaction and movement between the units, simulate the change process of the geological structure under the action of gravity, tectonic stress, etc., and combine the topographic and geomorphic change data and historical information to predict the long-term change trend of the geological structure.

[0106] Within the geological lateral area, through comprehensive analysis of the changes in the current topographic and geomorphic features, such as changes in the morphology of mountains, the trend of river valleys, the undulation of plains, etc., and combine with historical topographic and geomorphic data (topographic and geomorphic information recorded in topographic maps, remote sensing images, etc. at different times). For example, if historical data shows that a fault in a certain area had signs of slow movement in the past, combined with the characteristic changes such as tiny surface dislocations in the current topographic and geomorphic features, the future activity trend of the fault can be predicted.

[0107] Based on the change status of topographic and geomorphic features and the change trend of the lateral geological structure. Consider that geological structure changes (such as compression, tension, etc.) will produce stress on the rock and soil masses, as well as the influence of topographic and geomorphic changes (such as erosion, deposition) on the rock and soil masses. For example, when the geological structure activity causes the rock and soil masses to be compressed, it may lead to an increase in the density of the rock and soil masses and a decrease in porosity, thus affecting its physical properties such as compressive strength.

[0108] When the geological structure changes or the physical properties of the rock and soil masses change, the density of the underground substances will change, which will in turn cause gravity anomalies. Combine the gravity anomaly data with the topographic and geomorphic changes (such as mountain uplift or subsidence may be accompanied by changes in the density of the rock and soil masses) to infer the change trend of the physical properties such as the density of the rock and soil masses. For example, if there is an abnormal increase in the gravity value in the fault activity area, it may indicate that high-density rock and soil masses have invaded due to tectonic movement.

[0109] Due to the different magnetic characteristics of different rock and soil masses, geological tectonic movements will lead to the redistribution of rock and soil masses containing magnetic minerals. By combining magnetic force data with topographic and geologic structural changes, the changing trends of physical properties such as the content of magnetic minerals in rock and soil masses can be obtained. For example, fold tectonic changes may cause the originally dispersed magnetic minerals to aggregate, resulting in local magnetic field anomalies.

[0110] By applying an electric field underground and measuring parameters such as the potential difference at different positions, the physical properties of rock and soil masses, such as porosity, water content, and lithology, will affect their electrical conductivity. Based on the electrical conductivity changes caused by geological tectonic activities (such as faults conducting groundwater) and topographic and geomorphic changes (such as changes in drainage conditions), the changing trends of physical properties such as the porosity and water content of rock and soil masses can be predicted.

[0111] Use a pressure device to evenly press a probe into the rock and soil mass to measure the resistance received by the probe. This resistance is closely related to the mechanical properties of the rock and soil mass. By combining the stress changes caused by geological tectonic activities (such as tectonic stress compacting or stretching the rock and soil mass) and the static cone penetration test data at different times of topographic and geomorphic changes (such as filling and excavation projects), the changing trends of physical properties such as the compressive strength and deformation modulus of the rock and soil mass can be predicted.

[0112] Based on geotechnical mechanics theory, use finite element software to establish a rock and soil mass model. Input data on topographic and geomorphic feature changes (such as terrain undulation and slope changes), trends in geological structural changes (such as fault displacement and fold deformation), and initial physical property parameters of the rock and soil mass (such as density and elastic modulus). By simulating the mechanical responses of the rock and soil mass under different boundary conditions and loads, predict the changing trends of physical properties of the rock and soil mass (such as stress, strain, pore water pressure, etc.) over time. For example, simulate the stress redistribution and physical property changes during a landslide.

[0113] For discrete granular rock and soil masses, the discrete element method is used for simulation. The rock and soil mass is regarded as composed of a large number of discrete particles. By simulating the interactions between particles (such as frictional force and cohesive force) and geological tectonic movements and topographic and geomorphic changes, the movement and deformation of the rock and soil mass under complex stress conditions can be simulated, and then the changing trends of physical properties such as the porosity and permeability of the rock and soil mass can be predicted. For example, simulate the liquefaction process and physical property evolution of loose sand under earthquake action.

[0114] Fuse the topographic and geomorphic change information obtained by remote sensing (such as changes in vegetation cover reflecting changes in the water content and stability of rock and soil masses) with the physical property data of rock and soil masses obtained by geophysical exploration, in-situ testing, etc., and geological structural change data. Utilize the powerful spatial analysis function of geographic information system (GIS) to establish a comprehensive database, judge the correlations between different data, and then construct a prediction model to obtain the lateral changing trends of physical properties of rock and soil masses.

[0115] Perform time - series analysis on topographic and geomorphic features, geological structures, and geotechnical physical property data from different periods based on a GIS platform. By exploring the variation patterns of data over time and combining with geological process mechanisms, predict the variation trends of geotechnical physical properties over a certain period in the future.

[0116] Integrate the predicted lateral geological structure change trend information (such as the degree of intensification or slowdown of fault activities, the direction of fold deformation, etc.) with the lateral geotechnical physical property change conditions (such as the increase or decrease of geotechnical density and strength, etc.) into lateral influence characteristic data. These data can comprehensively reflect the change trends of geological structures and geotechnical physical properties in the lateral geological region, providing basic data support for subsequent geological disaster assessment, engineering construction planning, etc.

[0117] Step S4 specifically includes:

[0118] Predict the longitudinal geological structure change trend by combining the topographic and geomorphic feature change conditions and historical topographic and geomorphic data in the longitudinal geological region;

[0119] Predict the longitudinal geotechnical physical property change trend based on the topographic and geomorphic feature change conditions and the longitudinal geological structure change trend in the longitudinal geological region;

[0120] Combine the longitudinal geological structure change trend and the longitudinal geotechnical physical property change conditions into longitudinal influence characteristic data.

[0121] The method for predicting the longitudinal geological structure change trend is the same as that for predicting the lateral geological structure change trend, and the method for predicting the longitudinal geotechnical physical property change trend is the same as that for predicting the lateral geotechnical physical property change trend, so it will not be repeated here.

[0122] Step S5 specifically includes:

[0123] When the lateral change stability of the lateral influence characteristic data is not lower than the longitudinal change stability of the longitudinal influence characteristic data, divide the first position - locking area with the projection point of the UAV flight position on the ground as the origin in the lateral geological region, and select the geological position set from the first position - locking area according to the abnormal position points corresponding to the abnormal data.

[0124] This application uses a variety of sensors and monitoring equipment to collect data. For example, geological sensors are used to monitor the displacement and stress changes of geological structures; rock and soil sensors obtain the physical property parameters of rock and soil, such as density, porosity, elastic modulus, etc. Topographic image data is obtained through remote sensing satellites and drone aerial surveys, and then feature data is extracted through image processing. At the same time, combined with the global positioning system (GPS) monitoring station to obtain surface displacement data to provide basic information for lateral and longitudinal impact feature data.

[0125] For example, statistical methods are used to identify and remove outliers, and interpolation methods are used to fill missing values ​​to preprocess the raw data. Then, data from different sources and formats are integrated according to spatial location and time sequence to form a unified data set for subsequent analysis of horizontal and vertical change stability.

[0126] Arrange the horizontal impact characteristic data and the vertical impact characteristic data in chronological order to form a time series. For example, the horizontal geological structure displacement data of a certain geological area are arranged in order according to the daily, monthly or annual monitoring values.

[0127] Calculate the mean and standard deviation of time series data. The mean reflects the average level of the data, and the standard deviation measures the dispersion of the data. The smaller the standard deviation, the smaller the fluctuation of the data around the mean, that is, the higher the stability of the change. For example, if the standard deviation of the horizontal impact feature data is small over a period of time, it means that its horizontal change stability is good.

[0128] If the data still has strong autocorrelation over a long time interval, it means that the data changes are relatively stable and there are no drastic changes. For example, the autocorrelation coefficient of the longitudinal impact feature data is high in multiple time intervals, indicating that its longitudinal changes are relatively stable.

[0129] Use linear regression and other methods to fit the trend line of time series data. If the trend line is relatively flat, it means that the data change trend is stable; if the slope of the trend line changes greatly, it means that the data change stability is poor. For example, the slope of the trend line of the horizontally affecting characteristic data fluctuates less in different time periods, indicating that its horizontal change stability is high.

[0130] When the monitoring points of the lateral influencing characteristic data are unevenly distributed, spatial interpolation methods (such as Kriging interpolation, inverse distance weighted interpolation, etc.) are used to expand the discrete monitoring data into a continuous spatial data surface in order to more comprehensively analyze the lateral changes.

[0131] Compare the horizontal and vertical impact characteristic data of different time periods to observe the data change trend and fluctuation. For example, compare the vertical impact characteristic data of the past five years and the past year to determine whether the vertical change stability has changed.

[0132] When the lateral change stability of the lateral influence feature data is lower than the longitudinal change stability of the longitudinal influence feature data, a second position locking area is divided in the geological longitudinal area with the projection point of the UAV flight position on the ground as the symmetry point, and a geological position set is selected from the second position locking area according to the abnormal position point corresponding to the abnormal data;

[0133] Among them, the geological position set includes at least three analysis position points.

[0134] This application divides the position locking area based on the change stability of the lateral and longitudinal influence feature data. When the lateral change stability is high, the first position locking area is divided in the geological lateral area with the UAV projection point as the origin; when the lateral stability is low, the second position locking area is divided in the longitudinal area with the projection point as the symmetry point. This way of dividing the area according to the data characteristics can closely surround the abnormal position point corresponding to the abnormal data and accurately locate the area where geological problems may exist.

[0135] A geological position set is selected from the divided position locking area, and it is required that the geological position set contains at least three analysis position points.

[0136] Step S6 specifically includes:

[0137] Extract the remotely sensed data to be identified corresponding to the geological position set in the geological lateral area, estimate the first data interference coefficient according to the lateral topographic and geomorphic feature change condition and the lateral influence feature data, and perform adjustment processing on this remotely sensed data according to the first data interference coefficient to obtain the first type of adjusted remotely sensed data set;

[0138] Extract the remotely sensed data to be identified corresponding to the geological position set in the geological longitudinal area, estimate the second data interference coefficient according to the longitudinal topographic and geomorphic feature change condition and the longitudinal influence feature data, and perform adjustment processing on this remotely sensed data according to the second data interference coefficient to obtain the second type of adjusted remotely sensed data set;

[0139] Among them, the remotely sensed data processing set includes the first type of adjusted remotely sensed data set and the second type of adjusted remotely sensed data set.

[0140] In the geological lateral area and the longitudinal area, corresponding position locking areas (such as the first position locking area and the second position locking area) have been divided in advance by comparing the change stability of the lateral and longitudinal influence feature data, etc., and geological position sets have been selected from these areas. The selection of these position sets comprehensively considers various factors such as geological structure and topographic and geomorphic features, aiming to cover the key geological information points to ensure that the remotely sensed data extracted subsequently can effectively reflect the geological regional characteristics.

[0141] Based on a determined set of geological locations, using the spatial query function of Geographic Information System (GIS) and combining with the geographic coordinate information of remote sensing data, accurately locate and extract the remote sensing data to be identified corresponding to the set of geological locations within the lateral geological area. These data record the surface features of the corresponding locations in the form of images or numerical values, including but not limited to land cover types, vegetation coverage, surface reflectivity, etc. In the longitudinal geological area, follow the same principles and methods to extract the remote sensing data to be identified corresponding to the set of geological locations within this area.

[0142] Deeply analyze the changes in the lateral topographic and geomorphic features. With the help of data such as Digital Terrain Model (DTM) and Digital Elevation Model (DEM), identify the morphological evolution of topographic units such as mountains, valleys, and plains in the lateral direction, such as the erosion of mountains, the widening or diversion of valleys, etc. These changes will affect the reflection and scattering characteristics of the surface to remote sensing signals. For example, a landslide causes a change in the terrain undulation, making the originally regular terrain reflection pattern complex and interfering with the accuracy of remote sensing data.

[0143] Integrate the lateral influence characteristic data such as the change trend of lateral geological structures (such as fault activities, fold deformations, etc.) and the change of physical properties of lateral rock and soil masses (such as the change of density and porosity of rock and soil masses). Use multidisciplinary knowledge such as geomechanics and geophysics, and combine with statistics to construct a data correlation model. For example, when fault activities cause the fragmentation of rock and soil masses, the change of their density and porosity will affect the penetration and reflection of remote sensing signals. Quantify this influence through the model to estimate the first data interference coefficient, which reflects the interference degree of lateral geological factors on remote sensing data.

[0144] Regarding the changes in the longitudinal topographic and geomorphic features, study the ups and downs, displacements, etc. of the strata in the vertical direction. Use data such as borehole data and geological profiles to obtain the changes in stratum thickness and the vertical displacement of stratum interfaces. For example, the vertical uplift or subsidence of strata will change the surface elevation, and thus affect the elevation measurement and object recognition of remote sensing data.

[0145] Integrate the longitudinal influence characteristic data such as the longitudinal geological structure changes (such as vertical fault activities, vertical deformations of stratum folds) and the changes in physical properties of longitudinal rock and soil masses (such as differences in physical parameters of rock and soil masses at different depths). Adopt a method similar to the lateral one to establish a longitudinal data correlation model to evaluate the interference of longitudinal geological factors on remote sensing data and obtain the second data interference coefficient, which is used to characterize the interference level of longitudinal geological conditions on remote sensing data.

[0146] According to the estimated first data interference coefficient, a series of data processing techniques are used to adjust the remotely sensed data to be identified in the geological horizontal area. If the interference coefficient indicates that there is a spectral signal shift caused by geological tectonic activities, a spectral correction algorithm can be used to adjust the disturbed spectral bands by comparing with the standard spectral library to restore their true ground object reflection characteristics. For the geometric deformation caused by topographic and geomorphic changes, a geometric correction method is used to geometrically correct the remotely sensed image based on known topographic control points and mathematical transformation models, eliminating the image deformation caused by factors such as terrain undulation and sensor attitude, and finally forming the first type of adjusted remotely sensed data set.

[0147] According to the second data interference coefficient, targeted processing is carried out on the remotely sensed data to be identified in the geological vertical area. If the physical properties of the longitudinal rock and soil masses change, resulting in noise interference in the remotely sensed data, filtering algorithms such as median filtering and Gaussian filtering are used to remove the noise and smooth the data. For the abnormal characteristics of the remotely sensed data caused by longitudinal geological tectonic activities, a feature repair algorithm is used to repair and supplement the abnormal characteristics in combination with geological prior knowledge to obtain the second type of adjusted remotely sensed data set.

[0148] Integrate the first type of adjusted remotely sensed data set and the second type of adjusted remotely sensed data set obtained through the above processing. During the integration process, ensure the spatial consistency and attribute integrity of the data, and establish a unified data storage structure and index mechanism to form the final remotely sensed data processing set. This data set removes or reduces the interference of geological factors on the original remotely sensed data, and more accurately presents the actual situation of the geological horizontal and vertical areas, providing a high-quality data basis for subsequent geological disaster monitoring, geological resource exploration, topographic and geomorphic evolution research, etc.

[0149] Judge whether there are geological disasters in the geological area according to the geological disaster type information and the remotely sensed data processing set, which specifically includes the following steps:

[0150] Extract the historical remotely sensed data of the same type of geological disasters at different location points in the historical period and the corresponding historical geological disaster actual situation information, and establish a deep learning geological disaster recognition model according to the historical remotely sensed data and the historical geological disaster actual situation information;

[0151] Input the geological disaster type information and the first type of adjusted remotely sensed data set into the deep learning geological disaster recognition model to obtain the first type of disaster recognition result set; input the geological disaster type information and the second type of adjusted remotely sensed data set into the deep learning geological disaster recognition model to obtain the second type of disaster recognition result set;

[0152] Judge whether there are geological disasters in the geological area through the first type of disaster recognition result set and the second type of disaster recognition result set.

[0153] This application widely collects historical remote sensing data of the same type of geological disasters (such as landslides, debris flows, surface deformations caused by earthquakes, etc.) at different location points during the historical period. These data cover images obtained from multiple remote sensing platforms, including optical remote sensing images, radar remote sensing images, etc., and record the spectral, texture, geometric and other characteristic information of the surface before and after the occurrence of geological disasters. At the same time, collect the corresponding historical information on the actual situation of geological disasters, such as the accurate location, scale, damage degree and other detailed information of the disaster occurrence, providing an accurate factual basis for subsequent analysis.

[0154] Using deep learning algorithms (such as convolutional neural network CNN, recurrent neural network RNN and its variants), taking historical remote sensing data as input and the historical information on the actual situation of geological disasters as annotations (i.e., the expected output). By continuously adjusting the weights and parameters of the neural network, the model learns the mapping relationship between historical remote sensing data and the actual situation of geological disasters. For example, when training a landslide disaster recognition model, let the model learn the correlation between the changes in terrain, landform, vegetation cover and other features in the remote sensing image during the occurrence of a landslide and the actual scale and location of the landslide, so as to establish a deep learning geological disaster recognition model that can accurately identify specific types of geological disasters.

[0155] Historical remote sensing data may contain noise, outliers and missing values. For noise, filtering techniques such as median filtering, Gaussian filtering, etc. are used to remove random noise; for outliers, they are identified and corrected or removed through statistical methods (such as box plot analysis); for missing values, interpolation methods (such as linear interpolation, spline interpolation) are used to fill them to ensure the integrity and accuracy of the data.

[0156] Remote sensing data from different sources may have differences in numerical range and dimension. To make the data comparable, normalization methods such as min-max normalization (mapping the data to the [0,1] interval) or Z-score normalization (making the data have a distribution with a mean of 0 and a standard deviation of 1) are used to process the historical remote sensing data.

[0157] Annotate the historical remote sensing data in combination with the historical information on the actual situation of geological disasters. Clearly mark the location, type (such as landslide, debris flow, earthquake, etc.), scale and other information of the occurrence of geological disasters, and convert it into a label form recognizable by a computer, providing accurate supervision information for model training.

[0158] Convolutional neural networks are good at processing data with a grid structure, such as remote sensing images. Its convolutional layer can automatically extract local features (such as texture, shape) in the image, and the pooling layer can reduce the data dimension and retain key information. For example, in landslide disaster recognition, CNN can learn the unique terrain texture and landform shape features of the landslide area through convolutional operations.

[0159] If the evolution information of geological disasters over time is considered, recurrent neural networks and their variants are more applicable. LSTM (Long Short-Term Memory) and GRU (Gated Recurrent Unit) can effectively handle the long-term dependence problems in long sequence data and can be used to analyze the changes of multi-temporal remote sensing data over time to identify the development trends of geological disasters.

[0160] When the data volume is limited, generative adversarial networks can be used for data augmentation. The generator generates samples similar to real remote sensing data, and the discriminator distinguishes between real data and generated data. Through adversarial training, the quality of the generated data is improved, the training dataset is expanded, and the generalization ability of the model is enhanced.

[0161] The preprocessed and labeled historical remote sensing data is divided into a training set, a validation set, and a test set. Generally, the training set is used for learning and adjusting the model parameters, the validation set is used to evaluate the model performance and adjust hyperparameters (such as learning rate, number of iterations, etc.) during the training process, and the test set is used to finally evaluate the generalization ability of the model on unseen data.

[0162] For classification tasks (judging whether there are geological disasters and the types of disasters), the cross-entropy loss function is commonly used; for regression tasks (predicting continuous values such as the scale of disasters), the mean squared error loss function is more appropriate. The model parameters are continuously optimized by minimizing the loss function.

[0163] Optimization algorithms are used to update the model parameters, such as Stochastic Gradient Descent (SGD) and its improved versions (such as Adagrad, Adadelta, Adam, etc.). These algorithms calculate the gradients of the loss function with respect to the model parameters and gradually adjust the parameter values to continuously reduce the loss of the model on the training set.

[0164] Multiple evaluation metrics are used to measure the model performance, such as accuracy (the ratio of the number of correctly identified samples to the total number of samples), recall (the ratio of the number of correctly identified positive samples to the actual number of positive samples), F1 value (the harmonic mean of accuracy and recall), etc. In geological disaster identification, these metrics can help evaluate the model's ability to identify different types of geological disasters.

[0165] According to the evaluation results of the validation set and the test set, the model is optimized. If the model shows overfitting (performing well on the training set but poorly on the test set), regularization methods (such as L1, L2 regularization), Dropout technology, etc. can be used to reduce the model complexity; if the model shows underfitting (performing poorly on both the training set and the test set), try to increase the number of model layers, the number of neurons, or adjust the hyperparameters, and retrain the model until the model performance reaches a satisfactory level.

[0166] Input the currently obtained geological disaster type information (specifying which type of geological disaster, such as debris flow) and the first type of adjusted remote sensing data set (remote sensing data adjusted for geological lateral area interference factors) into the established deep learning geological disaster recognition model. The model judges the data in the first type of adjusted remote sensing data set based on the knowledge and patterns learned during the training stage and outputs the first type of disaster recognition result set. This result set includes whether there is a target geological disaster in the geological lateral area.

[0167] Similarly, input the geological disaster type information and the second type of adjusted remote sensing data set (remote sensing data adjusted for geological longitudinal area interference factors) into the deep learning geological disaster recognition model. The model processes the remote sensing data of the longitudinal area and outputs the second type of disaster recognition result set, giving the recognition and judgment information about geological disasters in the geological longitudinal area.

[0168] Comprehensively analyze the first type of disaster recognition result set and the second type of disaster recognition result set. Judge whether there is a geological disaster in the entire geological area.

[0169] Judge whether there is a geological disaster in the geological area through the first type of disaster recognition result set and the second type of disaster recognition result set, specifically including the following steps:

[0170] If each disaster recognition result in the first type of disaster recognition result set shows the existence of a disaster, then it is determined that there is a geological disaster in this geological area. If each disaster recognition result in the second type of disaster recognition result set shows the existence of a disaster, then it is determined that there is a geological disaster in this geological area;

[0171] If there is one disaster recognition result in the first type of disaster recognition result set that shows the non-existence of a disaster, then it is determined that there is no geological disaster in this geological area. If there is one disaster recognition result in the second type of disaster recognition result set that shows the non-existence of a disaster, then it is determined that there is no geological disaster in this geological area.

[0172] Since the geological location set includes at least three analysis location points, there are at least three remote sensing data to be recognized corresponding to the geological location set in the geological lateral area. Therefore, the first type of adjusted remote sensing data set includes at least three remote sensing data to be recognized, and the second type of adjusted remote sensing data set includes at least three remote sensing data to be recognized. Therefore, the first type of disaster recognition result set includes at least three disaster recognition results, and the second type of disaster recognition result set includes at least three disaster recognition results.

[0173] When in the first type of disaster recognition result set, each disaster recognition result indicates the existence of a disaster, it means that the entire area exhibits the characteristics of geological disasters. Based on this, it can be determined that there are geological disasters in this geological area. Similarly, if each result in the second type of disaster recognition result set also shows the existence of a disaster, it means that from the longitudinal area, it also meets the characteristics of geological disasters occurring, and it can also be determined that there are geological disasters in this geological area. This determination method is based on the consideration of overall consistency. As long as a certain type of result set points to the existence of a disaster in all results, it can be determined that there are disasters in the geological area.

[0174] If only one disaster recognition result in the first type of disaster recognition result set shows the non-existence of a disaster, this indicates that there are some areas in the horizontal area that do not meet the characteristics of geological disasters occurring. From an overall perspective, it is determined that there are no geological disasters in this geological area. The same is true for the second type of disaster recognition result set. As long as one of the results shows the non-existence of a disaster, it means that there are parts in the longitudinal area without disaster characteristics, and then it is determined that there are no geological disasters in this geological area.

[0175] As Figure 2 shown, the intelligent geological disaster identification system based on UAV remote sensing data further includes:

[0176] The marker matching module uses the remote sensing equipment carried by the UAV based on the flight position of the UAV to obtain the remote sensing data to be identified in different geological areas, analyzes and screens the remote sensing data to be identified to obtain the remote sensing data to be identified suspected of having geological disaster characteristics and marks it as abnormal data, and matches the corresponding geological disaster type information from the geological disaster information database according to the abnormal data;

[0177] The division module divides the geological area where the abnormal data is located into a geological horizontal area and a geological longitudinal area;

[0178] The first processing and analysis module processes and analyzes the change situation of the topographic and geomorphic features in the geological horizontal area to obtain the horizontal influence feature data;

[0179] The second processing and analysis module processes and analyzes the change situation of the topographic and geomorphic features in the geological longitudinal area to obtain the longitudinal influence feature data;

[0180] The comparison and selection module compares the horizontal change stability of the horizontal influence feature data and the longitudinal change stability of the longitudinal influence feature data, and then selects a geological position set in the geological horizontal area or the geological longitudinal area;

[0181] The processing module extracts the remote sensing data to be identified corresponding to the geological position set, and performs interference removal processing on the remote sensing data to be identified corresponding to the geological position set to obtain a remote sensing data processing set;

[0182] A judgment module that determines whether there is a geological disaster in a geological area based on geological disaster type information and a remote sensing data processing set.

[0183] An electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements a deep learning intelligent identification method for geological disasters based on drone remote sensing data.

[0184] As Figure 3 shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630, and a communication bus 640. Among them, the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 can call the logical instructions in the memory 630 to execute a deep learning intelligent identification method for geological disasters based on drone remote sensing data.

[0185] In addition, when the logical instructions in the above-mentioned memory 630 are implemented in the form of software functional units and sold or used as an independent product, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical disks, and other various media that can store program codes.

[0186] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute a deep learning intelligent identification method for geological disasters based on drone remote sensing data.

[0187] On yet another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is implemented to execute a deep learning intelligent identification method for geological disasters based on drone remote sensing data.

[0188] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative work.

[0189] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the above technical solution, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0190] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A deep learning method for intelligent identification of geological hazards based on UAV remote sensing data, characterized in that: The method comprises the following steps: S1. Based on the flight position of the UAV, the remote sensing equipment carried by the UAV is used to obtain the remote sensing data to be identified in different geological areas, and the remote sensing data to be identified are analyzed and screened to obtain the remote sensing data to be identified that are suspected to have geological disaster characteristics and marked as abnormal data, and the corresponding geological disaster type information is matched from the geological disaster information database according to the abnormal data; S2. Divide the geological area where the abnormal data is located into a geological transverse area and a geological longitudinal area; the geological transverse area refers to the geological range extending in the horizontal direction, and the geological longitudinal area refers to the geological range in the vertical direction; S3. Process and analyze the changes in topographic and geomorphic characteristics within the geological transverse region and historical topographic and geomorphic data to obtain transverse impact characteristic data; S4. Process and analyze the change status of topographic and geomorphic characteristics in the geological longitudinal region and the historical topographic and geomorphic data to obtain longitudinal impact characteristic data; S5. After comparing the lateral change stability of the lateral impact characteristic data and the longitudinal change stability of the longitudinal impact characteristic data, a geological position set is selected within the geological lateral region or the geological longitudinal region; When the lateral change stability of the lateral influencing characteristic data is not lower than the longitudinal change stability of the longitudinal influencing characteristic data, a first position locking area is divided in the geological lateral area with the projection point of the UAV flight position on the ground as the origin, and a geological position set is selected from the first position locking area according to the abnormal position points corresponding to the abnormal data; When the lateral change stability of the lateral influencing characteristic data is lower than the longitudinal change stability of the longitudinal influencing characteristic data, a second position locking area is divided in the geological longitudinal area with the projection point of the UAV flight position on the ground as a symmetrical point, and a geological position set is selected from the second position locking area according to the abnormal position point corresponding to the abnormal data; Wherein, the geological location set includes at least three analysis location points; S6, extracting the remote sensing data to be identified corresponding to the geological location set, and performing interference removal processing on the remote sensing data to be identified corresponding to the geological location set to obtain a remote sensing data processing set; S7. Determine whether there are geological disasters in the geological area based on the geological disaster type information and remote sensing data processing set.

2. The deep learning method for intelligent identification of geological hazards based on UAV remote sensing data according to claim 1 is characterized in that: The remote sensing data is analyzed and screened to obtain data suspected of having geological disaster characteristics and marked as abnormal data, specifically: The texture features of the remote sensing data to be identified are compared with the texture features of the preset normal geological remote sensing data, and the remote sensing data to be identified whose texture features are different from those of the preset normal geological remote sensing data are screened out and marked as abnormal data.

3. The deep learning method for intelligent identification of geological hazards based on UAV remote sensing data according to claim 2 is characterized in that: The S3 step specifically includes: Combine the changes in topographic and geomorphic features in the geological transverse area with historical topographic and geomorphic data to predict the geological structure change trend in the geological transverse area and obtain the horizontal geological structure change trend; According to the change of topographic features and the change trend of lateral geological structure, the change of physical properties of rock and soil in the lateral geological area is predicted to obtain the change trend of lateral rock and soil physical properties; The lateral geological structure change trend and the lateral change of rock and soil physical properties are combined into lateral impact characteristic data.

4. The deep learning method for intelligent identification of geological hazards based on UAV remote sensing data according to claim 3 is characterized in that: The S4 step specifically includes: Combine the changes in topographic and geomorphic features in the geological longitudinal region with historical topographic and geomorphic data to predict the trend of geological structure changes in the geological longitudinal region and obtain the longitudinal geological structure change trend; According to the change of topographic features and the change trend of longitudinal geological structure, the change of physical properties of rock and soil in the longitudinal geological area is predicted to obtain the change trend of physical properties of longitudinal rock and soil; The longitudinal geological structure change trend and the longitudinal rock and soil physical property change situation are combined into longitudinal impact characteristic data.

5. The deep learning method for intelligent identification of geological hazards based on UAV remote sensing data according to claim 4 is characterized in that: Step S6 specifically includes: Extracting remote sensing data to be identified corresponding to a geological position set within a geological transverse region, estimating a first data interference coefficient according to a change in transverse topographic and geomorphic features and transverse impact feature data, and adjusting the remote sensing data according to the first data interference coefficient to obtain a first type of adjusted remote sensing data set; Extracting the remote sensing data to be identified corresponding to the geological position set within the geological longitudinal region, estimating the second data interference coefficient according to the longitudinal topographic and geomorphic feature change status and the longitudinal impact feature data, and adjusting the remote sensing data according to the second data interference coefficient to obtain a second type of adjusted remote sensing data set; The remote sensing data processing set includes a first type of adjusted remote sensing data set and a second type of adjusted remote sensing data set.

6. The deep learning method for intelligent identification of geological hazards based on UAV remote sensing data according to claim 5 is characterized in that: According to the geological disaster type information and remote sensing data processing set, it is judged whether there is a geological disaster in the geological area, which specifically includes the following steps: Extract historical remote sensing data of the same type of geological disasters at different locations in the historical period and the corresponding actual situation information of historical geological disasters, and establish a deep learning geological disaster identification model based on the historical remote sensing data and the actual situation information of historical geological disasters; Input the geological disaster type information and the first type of adjusted remote sensing data set into the deep learning geological disaster identification model to obtain the first type of disaster identification result set; input the geological disaster type information and the second type of adjusted remote sensing data set into the deep learning geological disaster identification model to obtain the second type of disaster identification result set; The first type of disaster identification result set and the second type of disaster identification result set are used to determine whether there are geological disasters in the geological area.

7. The deep learning method for intelligent identification of geological hazards based on UAV remote sensing data according to claim 6 is characterized in that: Judging whether there are geological disasters in the geological area through the first type of disaster identification result set and the second type of disaster identification result set specifically includes the following steps: If each disaster identification result in the first type of disaster identification result set shows that a disaster exists, it is determined that a geological disaster exists in the geological area; if each disaster identification result in the second type of disaster identification result set shows that a disaster exists, it is determined that a geological disaster exists in the geological area; If one of the disaster identification results in the first type of disaster identification result set shows that there is no disaster, it is determined that there is no geological disaster in the geological area. If one of the disaster identification results in the second type of disaster identification result set shows that there is no disaster, it is determined that there is no geological disaster in the geological area.

8. A deep learning geological disaster intelligent identification system based on UAV remote sensing data, applied to a deep learning geological disaster intelligent identification method based on UAV remote sensing data as claimed in any one of claims 1 to 7, characterized in that: Also includes: The tag matching module uses the remote sensing equipment carried by the drone to obtain the remote sensing data to be identified in different geological areas based on the flight position of the drone, analyzes and screens the remote sensing data to be identified to obtain the remote sensing data suspected of having geological disaster characteristics and marks them as abnormal data, and matches the corresponding geological disaster type information from the geological disaster information database based on the abnormal data; A division module divides the geological area where the abnormal data is located into a geological horizontal area and a geological vertical area; The first processing and analysis module processes and analyzes the changes in topographic and geomorphic features within the geological transverse region to obtain transverse impact feature data; The second processing and analysis module processes and analyzes the change of topographic and geomorphic features in the geological longitudinal area to obtain longitudinal impact feature data; A comparison and selection module, which selects a geological position set within a geological transverse region or a geological longitudinal region after comparing the lateral change stability of the lateral impact characteristic data and the longitudinal change stability of the longitudinal impact characteristic data; A processing module extracts the remote sensing data to be identified corresponding to the geological location set, and performs interference removal processing on the remote sensing data to be identified corresponding to the geological location set to obtain a remote sensing data processing set; The judgment module judges whether there are geological disasters in the geological area based on the geological disaster type information and remote sensing data processing set.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, it implements the deep learning geological disaster intelligent identification method based on drone remote sensing data as described in any one of claims 1 to 7.

Images