A dangerous rock rapid screening and priority recheck indicating system based on unmanned aerial vehicle images

Abstract

The present application belongs to the technical field of geological disaster monitoring, and particularly relates to a dangerous rock rapid screening and priority recheck indication system based on unmanned aerial vehicle images; the system comprises: an unmanned aerial vehicle platform flies according to a pre-set inspection route, continuously collects rock wall images along the way and transmits the images to an end-side computing module; rock wall screening results processed by the end-side computing module are obtained and transmitted to a cloud-side analysis module; the end-side computing module is used for processing the rock wall images while the unmanned aerial vehicle platform is flying and inspecting, so as to obtain the rock wall screening results; the cloud-side analysis module is a ground server or a remote cloud server, and the cloud-side analysis module and the unmanned aerial vehicle platform transmit data and instructions through a wireless communication link; the cloud-side analysis module is used for comprehensively analyzing the rock wall screening results and historical data, generating a recheck plan and instructions and transmitting the recheck plan and instructions back to the unmanned aerial vehicle platform or a ground station for execution and scheduling; the present application significantly improves the identification timeliness and the scientific nature of scheduling through edge cloud cooperation and recheck closed loop.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of geological disaster monitoring, and particularly relates to a dangerous rock rapid screening and preferential re-inspection indication system based on unmanned aerial vehicle images. BACKGROUND

[0002] In the field of geological disaster prevention such as mountain slope, dangerous rock (referring to potential unstable rock mass with collapse conditions and precursors) poses a serious threat to personnel and facilities below. Traditionally, the identification and stability assessment of dangerous rock mainly rely on artificial field investigation: engineering geology personnel need to go deep into the field to map the position, size, fracture occurrence and other information of dangerous rock, and judge the stability of rock mass combined with experience. However, for high and steep slopes, artificial climbing measurement is very difficult, and it is often impossible to fully obtain the key information of dangerous rock structure surface. This leads to large errors in the stability assessment of dangerous rock, and further affects the pertinence and timeliness of the treatment measures. In addition, the artificial field investigation itself also has safety hazards. Therefore, under complex terrain such as high and steep slopes, it is of great significance to introduce unmanned aerial vehicle to carry out non-contact dangerous rock patrol: unmanned aerial vehicle carries high-definition camera and other sensors, which can obtain the image and three-dimensional information of dangerous rock body at a safe distance, overcoming the limitation that personnel cannot reach, and providing a new means for dangerous rock monitoring.

[0003] At present, unmanned aerial vehicle has been applied to dangerous rock patrol and data collection. However, in the prior art, the unmanned aerial vehicle dangerous rock patrol has the following shortcomings:

[0004] Low efficiency of artificial screening: a large number of images obtained during the patrol process usually need to be checked and screened by artificial frame by frame after the flight, which is difficult to find hidden dangers from the massive images in time. A flight of unmanned aerial vehicle may collect hundreds of slope photos, and artificial analysis not only consumes time, but also is easily affected by subjective factors, which may miss important signs.

[0005] Image data accumulation in high-risk areas: in the face of massive data accumulated by frequent patrol, the existing means lack effective real-time processing on site, and the data often accumulates in unmanned aerial vehicle or storage device waiting for offline analysis. This lag makes it difficult to quickly locate potential dangerous rock hidden dangers, reducing the timeliness of early warning.

[0006] No quantitative re-inspection decision: the existing inspection mode usually arranges subsequent re-inspection according to fixed period or experience, and lacks scientific quantitative index based on risk degree and development trend. This may lead to the following situations: the real high-risk rock mass cannot be re-inspected in time, missing the best intervention opportunity; frequent inspection of some areas with no significant changes, causing waste of manpower and material resources.

[0007] In summary, a new method is urgently needed in the current UAV dangerous rock inspection process, which can improve the efficiency of on-site data screening, timely identify high-risk rock mass, and provide objective basis for subsequent review, so as to improve the overall effect of dangerous rock monitoring and early warning. SUMMARY

[0008] In view of the deficiencies of the prior art, the present application provides a dangerous rock rapid screening and priority review indication system based on UAV images, which comprises a UAV platform, an end-side computing module and a cloud-side analysis module.

[0009] The UAV platform is used to fly according to a pre-set inspection route and continuously collect rock wall images along the way and transmit the rock wall images to the end-side computing module; the rock wall screening results processed by the end-side computing module are obtained and transmitted to the cloud-side analysis module.

[0010] The end-side computing module is installed on the UAV platform; the end-side computing module is used to process the rock wall images while the UAV platform is flying for inspection, and obtain the rock wall screening results.

[0011] The cloud-side analysis module is a ground server or a remote cloud server, and the cloud-side analysis module and the UAV end transmit data and instructions through a wireless communication link; the cloud-side analysis module comprehensively analyzes the screening results and historical data, generates a review plan and instructions, and transmits them back to the UAV platform or the ground station for execution and scheduling, and operates in a closed loop.

[0012] Preferably, the UAV platform comprises an aircraft body, a flight controller, a positioning module, a data communication unit and a camera module; the UAV platform continuously collects rock wall images along the way, which specifically comprises: the camera module takes pictures according to a fixed flight strip overlap rate, and captures rock wall images; each frame of image has a time stamp and geographical coordinates and attitude information obtained from the positioning module; the camera module transmits the rock wall images to the end-side computing module; the data communication unit is used to transmit the processing results of the end-side computing module to the cloud-side analysis module.

[0013] Preferably, the process of the end-side computing module processing the rock wall images comprises:

[0014] The rock wall images are pre-processed, including distortion correction, light equalization and noise filtering; ROI screening is adopted, focusing on the rock exposed part in the image, to obtain the pre-processed rock wall images;

[0015] An embedded deep learning model is used to process the pre-processed rock wall images to obtain dangerous rock symptom features;

[0016] The risk score is calculated according to the preset feature weight and the dangerous rock symptom features;

[0017] According to the risk score, the risk level of the rock wall image is divided; the rock wall image and the corresponding dangerous rock sign feature, the risk score, and the risk level are taken as the risk rock wall screening result.

[0018] Further, the formula for calculating the risk score is:

[0019]

[0020] wherein, represents the risk score, represents the crack density, represents the penetration degree, represents the unfavorable out-dip degree, represents the free space connectivity, represents the parallel length ratio along the slope, represents the vegetation false detection degree; 、 、 、 、 and represent the weight of the corresponding feature, respectively.

[0021] The cloud analysis module includes a data receiving and storage unit, a risk summary and evaluation unit, a historical change analysis unit, a review priority decision unit, and a task scheduling and issuing unit;

[0022] The data receiving and storage unit is used to receive the rock wall screening result and write it into the cloud database, and establish a record entry for each unmanned aerial vehicle task;

[0023] The historical change analysis unit is used to retrieve historical entry data and determine the risk development trend of the risk area according to the historical entry data;

[0024] The risk summary and evaluation unit is used to update the cloud database according to the risk development trend of the risk area;

[0025] The review priority decision unit generates a priority list according to the latest cloud database information;

[0026] The task scheduling and issuing unit generates a review plan according to the priority list; and displays the review plan or directly sends a review instruction to the unmanned aerial vehicle platform.

[0027] Further, the process of updating the cloud database includes:

[0028] Comparing the current risk area with the risk area in the original cloud database to determine whether it is a new risk area, if so, recording the dangerous rock sign feature of the corresponding risk area in the cloud database; otherwise, updating the dangerous rock sign feature of the corresponding risk area in the cloud database;

[0029] The risk parameter is corrected, and the risk score and risk level of the risk area are corrected according to the dangerous rock sign characteristics of the risk area and the corrected risk parameter.

[0030] Further, the formula for correcting the risk score of the risk area is:

[0031]

[0032]

[0033] wherein, represents the corrected risk score, represents an intermediate parameter, represents a trend score, represents a three-dimensional consistency score, represents a scene key degree, represents a quality consistency, represents a stable evidence; , , , and respectively represent the weight of the corresponding feature, represents the original risk score, represents a function.

[0034] Preferably, the formula for generating the priority list by the review priority decision unit is:

[0035]

[0036]

[0037] )

[0038] wherein, represents a priority value, represents the latest risk score, represents a facility proximity weight, represents a risk development trend factor, represents a time interval factor, represents the time since the last review, represents a normalized scale, represents a set patrol interval upper limit, represents a function, represents a minimum value; , and respectively represent a risk weight, a facility weight, and a time weight.

[0039] Preferably, the system uses a high-risk trigger and an energy adaptive strategy, including: when the risk score calculated by the end-side computing module or the priority value calculated by the cloud-side analysis module is greater than a preset threshold, the unmanned aerial vehicle platform is controlled to hover to take multiple images; only the structured results and thumbnails are reported when the link is limited, and the large images are returned after the unmanned aerial vehicle platform returns; the end-side computing module caches the failed reporting records, and the records are batched and transmitted after the unmanned aerial vehicle platform returns; when the power of the unmanned aerial vehicle platform is lower than a first power threshold, the unmanned aerial vehicle platform enters a fine shooting mode, and when the power is lower than a second power threshold, the unmanned aerial vehicle platform returns.

[0040] The beneficial effects of the present application are:

[0041] Greatly improve the on-site identification efficiency: through the deployment of intelligent algorithms on the end side, the image can be screened in real time during the cruise process, reducing a large amount of manual intervention of non-risk data, and achieving early discovery and early marking of hidden dangers; optimize data transmission and processing: edge computing filters and compresses the information amount at the source, only reports key risk data, reduces the wireless transmission pressure and cloud computing burden, and realizes efficient data utilization; realize risk grading management: innovatively introduce the dangerous rock risk score and grading mechanism, so that each hidden danger point has a quantitative evaluation, and it is convenient to manage according to the urgency of the danger;

[0042] Guide the scientific decision-making of reexamination: the cloud provides objective reexamination order suggestions according to the risk grading and evolution trend, overcoming the blindness of the previous experience-based decision-making, and ensuring timely reexamination of high-risk points without wasting energy on low-risk points;

[0043] Improve the overall early warning capability: the system forms a complete closed loop of "patrol screening-risk assessment-reexamination guidance", which can significantly enhance the initiative and timeliness of geological disaster monitoring, and plays a key role in the early warning and prevention of dangerous rock body collapse. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 The figure is a structure diagram of the dangerous rock rapid screening and priority reexamination indication system in the present application;

[0045] Figure 2 The figure is a data flow diagram of the dangerous rock rapid screening and priority reexamination indication system in the present application;

[0046] Figure 3 The figure is a task execution flowchart of the unmanned aerial vehicle in the present application. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0048] The present application provides a dangerous rock rapid screening and priority review indication system based on unmanned aerial vehicle image, as shown in the figure, the system is composed of unmanned aerial vehicle platform, end side computing module and cloud analysis module; the end side computing module is installed on the unmanned aerial vehicle platform and connected with the camera / positioning / data communication unit, the processed result is sent to the cloud analysis module through the data communication unit by wireless link, and the cloud returns the instruction and plan to the unmanned aerial vehicle / ground station, forming a closed loop of edge-cloud cooperation. Figure 1

[0049] The unmanned aerial vehicle platform is used for flying according to the preset inspection route, continuously collecting the rock wall images along the way and transmitting the rock wall images to the end side computing module; obtaining the rock wall screening result processed by the end side computing module and transmitting it to the cloud analysis module; specifically:

[0050] The unmanned aerial vehicle platform is composed of an aircraft body and a flight controller (collectively referred to as 'unmanned aerial vehicle'), a positioning module (inertial navigation unit (IMU / INS) + satellite navigation positioning module (GNSS, such as GPS / Beidou), which outputs position and attitude after fusion calculation), a data communication unit (data transmission / image transmission / control link) and a camera module (at least one high-resolution camera).

[0051] The unmanned aerial vehicle preferably adopts a multi-rotor unmanned aerial vehicle, which has hovering and low-speed cruising capability, so as to observe the mountain slope surface at close range. The unmanned aerial vehicle platform continuously collecting the rock wall images along the way specifically includes: the camera module performs aerial photography according to the fixed flight strip overlap rate, captures the rock wall images; the geographic coordinates and attitude information (latitude, longitude, elevation and attitude angle (roll, pitch and heading)) output by the positioning module are bound to each frame of image, and a unified time reference (PPS / time stamp) is used to align the camera sampling time and the IMU / INS+GNSS calculation result, and the metadata is written for subsequent spatial positioning and cross-period comparison. This ensures the positioning of each suspected dangerous rock, and provides a spatial reference for comparing the changes at different times; the images collected by the camera module are transmitted to the end side computing module through the in-machine bus for real-time processing; the key results and necessary original images after processing are reported to the cloud analysis module by the data communication unit. Preferably, the unmanned aerial vehicle platform of the present application pays attention to stability and safety in design: the aircraft is configured with obstacle avoidance sensors and executes redundant control algorithm by the flight control; when flying in steep areas, the obstacle avoidance sensing and flight control limiting strategy are combined to avoid terrain collision and ensure image stability to meet the algorithm requirements.​

[0052] The edge computing module is installed on the drone platform, such as an embedded AI computing board (with GPU acceleration) or a high-performance flight controller. This module deploys a rockfall identification algorithm specifically optimized for edge devices, including a lightweight deep learning model and a rule-based evaluation procedure. The edge computing module processes rock wall images while the drone platform is conducting flight inspections, obtaining rock wall screening results. Specifically:

[0053] Upon acquiring each new image, the edge computing module immediately performs real-time analysis, as follows:

[0054] The rock wall images undergo preprocessing, including distortion correction, illumination equalization, and noise filtering, to improve the accuracy of key feature extraction. Preferably, ROI (Region of Interest) filtering can then be used to automatically focus on the exposed rock areas in the image, reducing interference from irrelevant areas such as vegetation and sky.

[0055] An embedded deep learning model is used to process the preprocessed rock wall images to obtain the characteristics of dangerous rock formations, specifically:

[0056] The embedded deep learning model, after being trained on a large number of samples for the characteristics of dangerous rocks, can automatically detect information such as cracks, signs of impending collapse, and degree of overhang on the rock surface. Specifically, the algorithm first extracts the crack network on the rock wall image and calculates the following six normalized features, namely the characteristics of dangerous rocks, in each candidate risk area: (1) Crack density D: the total length of the crack skeleton in the area and the area (1) Area ratio (or number of lines per unit area), and linearly normalized to [0,1]. (2) Connectivity C: Whether the crack skeleton forms a connecting path that crosses the upper / lower edge or left / right edge of the area and its crossing ratio (the ratio of the length of the maximum connected component to the characteristic scale of the area), normalized to [0,1]. (3) Outward dip unfavorability O: The outward dip / suspended degree of the rock mass is estimated from the texture and geometry (such as front edge protrusion, bottom crack, etc.), and the outward dip angle or suspension index is normalized to [0,1] according to the threshold. (4) Freeway connectivity F: The contact or proximity ratio between the crack endpoint / connected component and the freeway boundary (cliff edge, free face), normalized to [0,1]. (5) Downslope parallel length ratio S: The angle between the main direction of the crack and the slope aspect is less than the set threshold. The proportion of the skeleton length to the total length of the crack is normalized to [0,1]. (6) Vegetation false detection rate V: The proportion of pixels "obscured by vegetation or misjudged as cracks" obtained by vegetation discrimination (such as color / texture / NIR index) is normalized to [0,1].

[0057] Then, a risk score is calculated based on the preset feature weights and the characteristics of dangerous rock formations. The formula for calculating the risk score is as follows:

[0058]

[0059] wherein, represents a risk score, represents a crack density, represents a penetration degree, represents an unfavorable out-tilt degree, represents an air-connection degree, represents a parallel length ratio on slope, represents a vegetation false detection degree; and respectively represent the weight of the corresponding feature.

[0060] The risk level of the rock wall image is divided according to the risk score, and preferably, it is typically divided into three levels of high, medium and low according to a preset threshold value: the score higher than the threshold value L1 is judged as high risk (I level) and needs to be paid attention to; the score between the threshold value L1 and the threshold value L2 is medium risk (II level); and the score lower than the threshold value L2 is low risk (III level).

[0061] The above weight and threshold value can be automatically calibrated according to the scene, but need to fall within the range of weight ∈ [0.05, 0.40] and threshold value ∈ [0.30, 0.80].

[0062] For images with extremely low risk and no obvious cracks, the module can be directly marked as safe. The end-side screening process is fast and efficient, and can realize the synchronization of unmanned aerial vehicle flight patrol and data analysis. For example, within a few seconds of the unmanned aerial vehicle hovering to take a picture of a rock wall, the dangerous rock identification and scoring of the image are completed. Thus, the end-side computing module produces the first round of risk rock screening results while the unmanned aerial vehicle is still on the scene; wherein the risk rock screening results include the rock wall image and the corresponding dangerous rock feature, risk score, risk level information.

[0063] In some preferred embodiments of the present application, the risk level can also be divided by using a rule-based expert system, for example, if multiple vertical penetrating cracks and rock body forward tilting are found, it is determined as high risk.

[0064] Through the design of the above-mentioned end-side module, the present application realizes the preliminary screening and quantitative evaluation of dangerous rock hazards on the end-side (unmanned aerial vehicle on-board). This on-site real-time screening significantly shortens the time from data collection to risk discovery, avoids the accumulation of inspection data, and wins valuable time for the subsequent link. In addition, since a large number of safe area images are screened out and only key risk information is uploaded, the communication and cloud processing pressure is greatly reduced.

[0065] ​​​​The cloud-based analysis module is either a ground server or a remote cloud server. It transmits data and commands to the UAV platform via a wireless communication link. The cloud-based analysis module is used to comprehensively analyze the rock face screening results and historical data, generate review plans and commands, and transmit them back to the UAV platform or ground station for execution and scheduling. Specifically:

[0066] After receiving data from the drone inspection, the cloud-based analytics module performs deeper analysis and decision generation. Its workflow is as follows: Figure 2 As shown. The cloud analytics module includes a data receiving and storage unit, a risk aggregation and assessment unit, a historical change analysis unit, a review priority decision-making unit, and a task scheduling and distribution unit.

[0067] The data receiving and storage unit is used to receive the rock wall screening results and write them into the cloud database, and to create a record entry for each UAV mission; preferably, the entry data includes mission time, risk area, and a list of suspected dangerous rocks identified by the end side.

[0068] Risk areas refer to high-risk spatial units aggregated by algorithms in space, stored as polygons (WGS-84 coordinates) or raster indexes; their geometric information (polygon, bounding rectangle, area), average S̄ / maximum risk score S_max within the area, image tile / tile list, and basic terrain attributes (average slope / aspect) are recorded.

[0069] The list of suspected dangerous rock masses identified by the end-side: refers to the set of candidate bodies (rock masses of high and medium risk levels) output in real time during the task. Each candidate body includes: a unique ID, spatial location (latitude / longitude / elevation / corresponding image fragment index), model confidence, risk score S and risk parameter vector (significant rock mass characteristics), geometric statistics (total crack length, maximum width, number of connected components, main direction), and three-dimensional terrain attributes (local slope, aspect, shortest distance to the free face), etc.

[0070] The cloud database organizes data using geospatial indexes, enabling the linked storage of observations of the same location at different times, facilitating historical comparisons.

[0071] The historical change analysis unit is used to retrieve data from previous entries and determine the risk development trend of risk areas based on this data.

[0072] For each identified dangerous rock in the cloud database, the system retrieves its characteristic data from previous inspections (crack density, length, risk score, etc., recorded over time). The algorithm compares the latest data with data from one or more previous inspections to calculate the differences: for example, by how much the crack length increased, how many new cracks appeared, and by how much the risk score increased / decreased.

[0073] Combining multi-period data can also determine the trend of change: if the score of a certain rock mass increases for two consecutive times, the trend is deteriorating; if it remains stable for multiple times, the trend is stable. The cloud generates a trend evaluation result (trend label) for each region (such as "risk continues to rise", "basically stable" or "slows down" and the like). This evolution analysis provides a scientific basis for review decisions - compared with only looking at single results, it can better reflect the development dynamics of hidden dangers.

[0074] The risk summary and evaluation unit is used to update the cloud database according to the risk development trend of the risk region. The process of updating the cloud database includes:

[0075] Compare the current risk region with the risk region in the original cloud database to determine whether it is a new risk region. If it is, record the dangerous rock sign characteristics of the corresponding risk region in the cloud database; otherwise, update the dangerous rock sign characteristics of the corresponding risk region in the cloud database.

[0076] Next, the cloud performs fine-grained evaluation for each risk region: using the strong computing power of the cloud, more complex algorithms and models can be called to verify and supplement the results on the terminal side. Specifically:

[0077] Obtain the risk correction parameter, and correct the risk score and risk level of the risk region according to the dangerous rock sign characteristics of the risk region and the risk correction parameter. Specifically:

[0078] For high-risk points, the cloud can extract multi-angle images or three-dimensional point cloud data (if the drone is equipped with multi-view images or even laser point clouds), and evaluate the specific instability mode and danger degree of the rock mass block through structure surface analysis, three-dimensional reconstruction or finite element stability simulation. These advanced analyses can help reduce false positives and improve accuracy. For example, the terminal side may judge a high risk due to crack shadows, but the cloud may lower the risk rating after confirming the crack depth through multi-angle confirmation; conversely, the cloud may find details of hidden dangers that the terminal side has not detected through fine analysis, thereby increasing the risk level.

[0079] In order to specifically quantify the adjustment, first obtain the new original risk score according to the dangerous rock sign characteristics of the risk region, and the formula is as described above. Then, obtain the risk correction parameter, including the trend score, three-dimensional consistency score, scene key degree, quality consistency and stability evidence; according to the risk correction parameter and the original risk score , the new risk score is corrected. The formula for correcting the risk score of the risk region designed by the present application is:

[0080]

[0081]

[0082] wherein, represents the revised risk score, represents the intermediate parameter; T represents the trend score, which is the slope of the score over the last several observations is normalized to: , with a value range of [−1, 1]. is the normalized scale, preferably = 0.06 / week (referring to an increase of 0.06 per 1 week). A positive value is assigned to an upward trend, and a negative value is assigned to a downward trend. The greater the slope (the greater the absolute value), the greater the magnitude of the trend score. represents the three-dimensional consistency score (multi-angle image or point cloud confirmation of the consistency of the inclination and connectivity), represents the scene criticality (distance weight from targets such as roads, pipelines, and residential buildings), represents the quality consistency (consistency of multiple models and artificial review is added points), represents stable evidence (no change or reverse recovery evidence is deducted points); , , , and represent the weights of the corresponding features, respectively, represents the original risk score, represents function.

[0083] M can be calculated by comparing the consistency of the crack geometric features extracted from the same candidate in multi-view images (or three-dimensional point clouds); K can be mapped to a weight value in the [0, 1] interval according to the actual distance of the rock mass from key targets (such as roads, pipelines, and residential buildings) (the closer the distance, the greater the weight); Q can be given by statistics of the consistency rate of multiple detection models and artificial labeling results (for example, if two models and artificial judgment exist cracks, then Q can take a high value); R can give positive / negative values according to the trend of historical risk score changes (if the score has no obvious upward trend, it is determined that there is stable evidence, and points are deducted).

[0084] The function is represented as:

[0085]

[0086] According to the updated risk score, the risk level is redefined.

[0087] The review priority decision unit generates a priority list according to the latest cloud database information. Specifically:

[0088] The cloud assigns a review priority to the dangerous rock point to be observed according to a certain strategy, and the strategy comprehensively considers the following factors: risk level (high risk of level I is prior to medium risk of level II, and level II is prior to low risk of level III), recent change range (recent deterioration is prior to no change), last inspection time (the longer the interval, the more urgent the review), and the like, and the on-site environment (for example, the hidden danger near important facilities should be more prior) is also considered if necessary. Through multi-factor decision, the system generates a priority list with weights. For example, a hidden danger at a certain place has been high risk and the score continues to rise this time, so its priority is the highest; on the contrary, a medium risk point that has existed for a long time can be appropriately reduced in review urgency if no obvious change is seen for a long time; in order to quantify the priority specifically, the formula for generating the priority list is designed:

[0089]

[0090]

[0091]

[0092] wherein, represents the priority value, represents the latest risk score, represents the facility proximity weight, represents the risk development trend factor, represents the time interval factor, represents the time from the last review, represents the normalized scale, represents the set upper limit of the patrol interval, represents function, represents the minimum value. , and are respectively the risk weight, the facility weight and the time weight, preferably, =0.5,b=0.3,c=0.2. The range of S is [0, 1], the range of H is [0, 1], the range of is [-1, 1], the range of is [0, 1], therefore, P takes the value [0, 2]. Preferably, the facility proximity weight (between 0 and 1) can be valued by the following way: key facilities (schools / hospitals / railways / highways / dense residents): ≤30 m: 1.0; 30-60 m: 0.8; 60-100 m: 0.5; 100 m: 0

[0093] important facilities (provincial and county roads / pipelines / scenic trails / base stations): ≤20 m: 0.8; 20-50 m: 0.6; 50-80 m: 0.3; 80 m: 0

[0094] General facilities (country roads / power transmission towers / scattered households / farmland): ≤20 m: 0.5; 20-40 m: 0.3; 40 m: 0 slope plus 0.1, effective protection minus 0.2.

[0095] In particular, for the value method at the same time adjacent to multiple facilities, that is, there is overlap, the value method is as follows:

[0096] The maximum weight of all categories corresponding to the overlap value (that is, the most sensitive facility type adjacent to the dangerous body) is taken to ensure the conservatism of risk assessment. For example, a dangerous point is located in the adjacent range of key facilities-highway (30 m) (weight 1.0) and important facilities-pipeline (20-50 m) (weight 0.6) at the same time, then the weight 1.0 can be taken as the comprehensive facility adjacent weight.

[0097] The segmentation is as follows:

[0098]

[0099] According to the priority value and the preset threshold, a specific priority level can be generated, for example, P≥1.20→ immediate review; 0.80≤P<1.20→ next flight priority; P<0.80→ routine inspection.

[0100] The task scheduling and issuing unit generates a review plan according to the priority list; and displays the review plan or directly sends a review instruction to the unmanned aerial vehicle platform. Specifically:

[0101] The cloud converts the review suggestions in priority order into specific inspection tasks. For example, the high-priority point list can form the key coordinates of the next unmanned aerial vehicle flight route planning, and the system automatically generates a review plan. These plans are issued to the unmanned aerial vehicle ground station or control center through the network, prompting the operator or autonomous unmanned aerial vehicle to perform targeted review cruise. For major hidden dangers that need to be reviewed immediately, the system can send an early warning notice and suggest sending an unmanned aerial vehicle or a human to verify and handle immediately. Correspondingly, low-priority areas can be monitored in routine inspection. Task issuing can be fully automatic (in an unmanned aerial vehicle cluster system, the cloud directly sends a review instruction to the unmanned aerial vehicle platform, and directs an idle unmanned aerial vehicle to review the high-priority point), or can display the review plan to the management personnel through the platform interface for prompting, and the management personnel make decisions to adopt and execute. Regardless of the way, a review management mechanism driven by risk ranking is finally established to ensure that limited patrol resources are optimally allocated: first, to ensure the safety of high-risk points, and then to consider general patrol.

[0102] For example, Figure 3As shown, the dangerous rock rapid screening and priority review indication system designed by the present application uses a high-risk trigger and energy adaptive strategy, including: when the risk score calculated by the terminal-side computing module or the priority value calculated by the cloud-side analysis module is greater than the preset threshold, the unmanned aerial vehicle platform is controlled to hover to take multiple images; when the link is limited, only the structured results and thumbnails are reported, and the large images are returned after the unmanned aerial vehicle platform returns; the terminal-side computing module buffers the failed reporting records and returns them in batches after returning; when the power of the unmanned aerial vehicle platform is lower than the first power threshold, the precise shooting mode is entered, and when the power is lower than the second power threshold, the unmanned aerial vehicle platform returns.

[0103] To sum up, through the cooperation of the above units, the present application realizes the secondary screening and overall planning of the terminal-side screening results, and truly implements the idea of "screening first and then reviewing"; the system designed by the present application significantly improves the identification timeliness and dispatching scientificity through edge-cloud cooperation and review closed loop, and has good application prospect.

[0104] The above examples further illustrate the purpose, technical solutions and advantages of the present application. It should be understood that the above examples are only preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made to the present application within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A rapid screening and priority review system for dangerous rocks based on UAV imagery, characterized in that, include: Unmanned aerial vehicle (UAV) platform, edge computing module, and cloud analytics module; The drone platform is used to fly according to a pre-set inspection route and continuously collect rock wall images along the flight path and transmit the rock wall images to the end-side computing module; Obtain the rock wall screening results processed by the end-side calculation module and transmit them to the cloud analysis module; The edge computing module is installed on the UAV platform; the edge computing module is used to process the rock wall images while the UAV platform is conducting flight inspections to obtain rock wall screening results. The process of the end-side calculation module processing the rock wall image includes: The rock wall image was preprocessed, including distortion correction, illumination equalization, and noise filtering; ROI screening was used to focus on the exposed rock parts in the image to obtain the preprocessed rock wall image. An embedded deep learning model was used to process the preprocessed rock wall images to obtain the characteristics of dangerous rock signs; A risk score is calculated based on preset feature weights and rockfall hazard indicators; the formula for calculating the risk score is as follows: ； in, Indicates risk score, Indicates crack density. Indicates the degree of penetration. Indicates the degree of extraversion disadvantage. Indicates airborne connectivity. This indicates the proportion of parallel lengths along the slope. Indicates the false detection rate of vegetation; , , , , and These represent the weights of the corresponding features; The risk level of rock wall images is determined based on the risk score; the rock wall images, along with the corresponding dangerous rock signs, risk scores, and risk levels, are used as the results of the risk rock wall screening. The cloud-based analysis module is either a ground server or a remote cloud server. It transmits data and commands to the UAV platform via a wireless communication link. The cloud-based analysis module comprehensively analyzes the rock face screening results and historical data, generates review plans and commands, and transmits them back to the UAV platform or ground station for execution and scheduling. The cloud database update process within the cloud-based analysis module includes: The current risk area is compared with the risk areas in the original cloud database to determine whether it is a new risk area. If it is, the dangerous rock sign characteristics of the corresponding risk area are recorded in the cloud database; otherwise, the dangerous rock sign characteristics of the corresponding risk area in the cloud database are updated. Obtain corrected risk parameters, and revise the risk score and risk level of the risk area based on the characteristics of dangerous rock formations and the corrected risk parameters; the formula for the corrected risk score of the risk area is expressed as: ； ； in, This indicates the revised risk score. Indicates intermediate parameters. The trend score is obtained by normalizing the slope of the scores from the most recent observations. This indicates the consistency score in three dimensions, and the consistency between outward tilt and connectivity is confirmed by multi-angle images or point clouds. Indicates the criticality of the scene, with the distance weight from targets such as roads, pipelines, and residences; This indicates quality consistency, including consistency across multiple models and consistency with manual review. This indicates stable evidence, with no change or signs of reverse recovery over time; , , , and These represent the weights of the corresponding features. This indicates the original risk score. express function.

2. The rapid screening and priority review system for dangerous rocks based on UAV images according to claim 1, characterized in that, The UAV platform includes an aircraft body, a flight controller, a positioning module, a data communication unit, and a camera module. The UAV platform continuously collects images of the rock face along its flight path, specifically through the following steps: the camera module takes aerial photos at a fixed flight path overlap rate to capture rock face images; each image frame includes a timestamp and geographic coordinates and attitude information obtained from the positioning module; the camera module transmits the rock face images to the edge computing module; and the data communication unit transmits the processing results from the edge computing module to the cloud analysis module.

3. The rapid screening and priority re-inspection system for dangerous rocks based on UAV images according to claim 1, characterized in that, The cloud-based analytics module includes a data receiving and storage unit, a risk aggregation and assessment unit, a historical change analysis unit, a review priority decision-making unit, and a task scheduling and distribution unit. The data receiving and storage unit is used to receive the rock wall screening results and write them into the cloud database, creating a record entry for each drone mission. The historical change analysis unit is used to retrieve data from previous entries and to determine the risk development trend of risk areas based on the data from previous entries; The risk aggregation and assessment unit is used to update the cloud database based on the risk development trends of risk areas; The review priority decision-making unit generates a priority list based on the latest cloud database information; The task scheduling and distribution unit generates a review plan based on the priority list; and displays the review plan or directly sends review instructions to the UAV platform.

4. The rapid screening and priority review system for dangerous rocks based on UAV images according to claim 1, characterized in that, The formula for generating the priority list by the review priority decision-making unit is expressed as follows: ； ； )； in, Indicates the priority value. This indicates the latest risk score. Indicates the proximity weight of facilities. Factors representing risk development trends Indicates the time interval factor. This indicates the time since the last follow-up examination. Indicates the normalization scale. This indicates the maximum set patrol interval. express function, This indicates finding the minimum value; , and These are risk weight, facility weight, and time weight, respectively.

5. The rapid screening and priority review system for dangerous rocks based on UAV images according to claim 1, characterized in that, The system employs a high-risk triggering and energy-adaptive strategy, including: when the risk score calculated by the edge computing module or the priority value calculated by the cloud analysis module exceeds a preset threshold, controlling the drone platform to hover and take multiple additional images; when the link is limited, only structured results and thumbnails are reported, and the drone platform returns to base with larger images; the edge computing module caches records of failed reports and retransmits them in batches after returning to base; when the drone platform's battery level is below the first battery threshold, it enters a precision shooting mode, and when the battery level is below the second battery threshold, the drone platform returns to base.

Images