Harmful object identification method fusing multi-source information acquired by unmanned aerial vehicle

Abstract

The invention relates to a hazardous substance identification method fusing multi-source information collected by an unmanned aerial vehicle, and belongs to the technical field of hazardous substance detection. Optical detection data and magnetic detection data collected by the unmanned aerial vehicle for the first time are identified, invalid data are eliminated, and an optical detection position point and confidence and a magnetic detection position point and confidence are obtained; corresponding target points are obtained through matching on the basis that all distance combinations of the optical detection position points and the magnetic detection position points are smaller than a first distance threshold value, a batch of hazardous object target points are found out through a first preset threshold value, secondary detection within a narrowing range is carried out on the other target points, and the remaining hazardous object target points are found out; the problems that in the prior art, existing detection efficiency and reliability are not high, and analysis and processing of detection results are complex are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of hazard detection, and in particular to a hazard identification method that integrates multi-source information collected by unmanned aerial vehicles. Background Art

[0002] How to quickly locate and eliminate this type of hidden hazards is a necessary condition to ensure the normal work and life of residents. Common hazard detection methods include optical detection and magnetic detection. Among them, the optical detection method can use artificial intelligence recognition methods to identify surface targets in pictures or videos, and the magnetic detection method can accurately detect ferromagnetic metal hazards underground or covered by vegetation through magnetic anomalies caused by ferromagnetic objects. Non-ferromagnetic hazards cannot be detected. Combining the two is currently a popular practice, which plays to their respective strengths and complements the shortcomings of the two detection methods.

[0003] The UAV underground hazard detection system is usually used for hazard detection at various shooting ranges and airports, as well as detection of mines and ferromagnetic obstacles in shallow waters. The operator operates at a safe distance, overcoming the disadvantage that the operator of the traditional handheld detection device must be close to the hazard, and can realize unmanned rapid exploration of a large area, which is an upgrade of detection means and equipment technology.

[0004] However, in the existing technology, the use of drones to conduct optical detection combined with magnetic detection of hazards has not achieved effective integration, and the analysis and processing of detection results are complicated, resulting in low efficiency and reliability in hazard detection. Summary of the invention

[0005] In view of the above analysis, an embodiment of the present invention aims to provide a hazard identification method that integrates multi-source information collected by a drone, so as to solve the problems of low efficiency and reliability of existing detection and complex analysis and processing of detection results.

[0006] On the one hand, an embodiment of the present invention provides a method for identifying hazards by integrating multi-source information collected by a drone, and the method specifically includes:

[0007] S1: The drone detects the target area and collects light detection data, magnetic detection data, and drone data corresponding to the light detection data and magnetic detection data respectively;

[0008] S2: Identify valid light detection data, and obtain all light detection position points and corresponding confidences based on the valid light detection data and its corresponding drone data, and obtain all magnetic detection position points and corresponding confidences based on the magnetic detection data and its corresponding drone data;

[0009] S3: combining each optical detection position point and each magnetic detection position point in a group one by one to calculate the distance of each combination, and then finding all combinations whose distances are less than a first distance threshold;

[0010] S4: Filter out the combination of magnetic detection position points with the shortest distance to each optical detection position point from all the combinations whose distance is less than the first distance threshold, and take the optical detection position point and the magnetic detection position point in each combination as a matching point pair;

[0011] S5: Get a target point and its confidence based on each matching point pair;

[0012] S6: Find the target point whose confidence level is not lower than the first preset threshold and store it in the target point set;

[0013] S7: storing the remaining target points, as well as the optical detection position points and the magnetic detection position points that do not form matching point pairs and whose confidence is not less than a second preset threshold, into a secondary detection point set;

[0014] S8: Establish a grid with the secondary detection point as the center, detect the grid again, and find the target point based on the re-detection result and store it in the target point set.

[0015] The beneficial effects of the above technical solution are as follows: the optical detection data and magnetic detection data collected by the drone for the first time are identified to eliminate invalid data and obtain the optical detection position points and confidence levels, the magnetic detection position points and confidence levels, and then the corresponding target points are obtained based on the combination matching of all distance combinations of the optical detection position points and the magnetic detection position points that are less than the first distance threshold. Then, a batch of hazardous target points are first found through the first preset threshold, and then the remaining target points, as well as the optical detection position points and magnetic detection position points that are not matched to the target points and whose confidence levels are not lower than the second preset threshold, are subjected to a secondary detection with a narrowed range to find the remaining hazardous target points, which not only improves the efficiency and reliability of detection, but also greatly simplifies the processing process.

[0016] Based on a further improvement of the above method, the light detection data includes an optical image;

[0017] The identifying and finding out the valid light detection data, and obtaining the light detection position point and confidence level based on the valid light detection data and the drone data, specifically includes:

[0018] After preprocessing, the optical image is input into the light detection recognition model for target recognition. If the target is recognized, the light detection data is used as valid light detection data, and an image with a target detection frame is output. The corresponding light detection position point is obtained based on the target detection frame and the drone data, and the confidence of the light detection position point is set to 1.

[0019] The beneficial effects of the above further improvement scheme are: the image recognition collected by light detection can directly determine whether there is evidence related to hazardous objects; for valid light detection data, that is, the confidence level of evidence of the existence of hazardous objects is set to 1, otherwise it is processed without confidence, which reduces the complexity of subsequent processing based on light detection data.

[0020] Based on the further improvement of the above method, the drone data includes light detection viewing angle, light detection image resolution, drone height, drone coordinates, and drone head direction; the corresponding light detection position point is obtained based on the target detection frame and drone data, specifically including:

[0021] The actual length and width corresponding to the light detection image are calculated based on the light detection viewing angle and the height of the drone;

[0022] Taking the upper left corner vertex of the light detection image as the origin, the normalized position of the pixel at the center point of the target detection frame in the light detection image relative to the center point of the light detection image is calculated based on the height of the drone;

[0023] Based on the normalized position, the direction of the drone head and the drone coordinates, the light detection position point is calculated.

[0024] The beneficial effect of the above further improvement scheme is that the light detection image position point can be obtained by collecting the light detection image, the height of the drone, the coordinates of the drone, and the direction of the drone head through the drone.

[0025] Based on the further improvement of the above method, the actual length and width of the target detection frame are calculated based on the light detection viewing angle and the height of the drone by the following formulas, which are specifically expressed as follows:

[0026] Where Widthl is the actual width of the light detection image, droneHeight is the relative height of the drone, Θ is the light detection viewing angle of the drone,

[0027] Wherein, Height1 is the actual length corresponding to the light detection image, deviceHeight is the length of the light detection image, and deviceWidth is the width of the light detection image;

[0028] The top left corner vertex of the light detection image is the origin. Based on the height of the drone, the normalized position of the pixel at the center point of the target detection frame in the light detection image relative to the center point of the light detection image is calculated by the following formula. The formula is specifically expressed as:

[0029] Wherein, tmpX is the normalized position horizontal coordinate, and x is the horizontal coordinate of the center point of the target detection frame in the light detection image pixel;

[0030] Wherein, tmpY is the normalized position ordinate, and y is the pixel ordinate of the center point of the target detection frame in the light detection image.

[0031] The beneficial effect of the above further improvement scheme is: the above formula can be used to convert the actual length and width corresponding to the light detection image target frame based on the light detection image target frame, the light detection viewing angle, and the drone height, and the normalized position of the center point of the light detection image can be obtained based on the coordinate system constructed with the upper left corner fixed point of the light detection image as the origin.

[0032] Based on the further improvement of the above method, the corresponding light detection position point is calculated by the following formula based on the normalized position, the direction of the drone head and the drone coordinates. The formula is specifically expressed as:

[0033] xLng = droneLng + (ttnpX × cos(droneHeading) - tmpY × sin(droneHeading)) × M_TO_LAT, where xLng is the longitude of the corresponding light detection position point, droneLng is the corresponding drone longitude, droneHeading is the direction of the drone head, and M_TO_LAT is the longitude and latitude resolution;

[0034] yLat=droneLat+(tmpY×cos(droneHeading)+tmpX×sin(droneHeading))×M_TO_LAT, where yLat is the latitude of the corresponding light detection position point, and droneLat is the corresponding drone latitude.

[0035] The beneficial effect of the above further improved solution is that the corresponding light detection position point can be calculated by the above formula based on the normalized position, the drone head direction and the drone coordinates.

[0036] Based on the further improvement of the above method, the magnetic detection data includes two-dimensional magnetic field value data; before the UAV detects the target area, the horizontal center error and depth error of the magnetic detection are first obtained by simulating target detection; the magnetic detection position point and confidence are obtained based on the magnetic detection data through the following steps, and the steps specifically include:

[0037] Based on the two-dimensional magnetic field value data, the corresponding coordinates of the magnetic field maximum and minimum values, as well as the aeromagnetic dynamic noise, the repeat line accuracy, and the total accuracy of the aeromagnetic survey are obtained;

[0038] The midpoint of the line connecting the maximum and minimum coordinates of the magnetic field is taken as the magnetic detection position point;

[0039] The magnetic detection position point confidence is adjusted based on the initial value of the magnetic detection position point confidence based on the aeromagnetic dynamic noise, repeat line accuracy, total aeromagnetic accuracy, magnetic detection horizontal center error and depth error and obtained.

[0040] The beneficial effect of the above further improved solution is that: through the above steps, the magnetic detection position point and confidence level can be obtained based on the magnetic detection data.

[0041] Based on a further improvement of the above method, a combination of magnetic detection position points having the shortest distance to each optical detection position point is selected from all combinations whose distance is less than the first distance threshold, and the optical detection position point and the magnetic detection position point in each combination are taken as a matching point pair, specifically including:

[0042] S81, arranging all the combinations whose distances are less than the first distance threshold in order of distance from small to large;

[0043] The optical detection position point and the magnetic detection position point which are closest to each other in the distance queue are regarded as a matching point pair;

[0044] S82: Eliminate all combinations of optical detection position points or magnetic detection position points in the matching point pair from the distance queue, update the distance queue, and return to repeat S81 to S82 until the distance queue is empty.

[0045] The beneficial effect of the above-mentioned further improvement scheme is: to find the combination of light detection position points and magnetic detection position points that are less than the first distance threshold is to first locate the possible matching combination of light detection position points and magnetic detection position points; then find the closest unique matching point pair of light detection position points and magnetic detection position points by sorting, with the purpose of determining the unique matching relationship between the light detection position points and the magnetic detection position points, that is, the light detection position points and the magnetic detection position points uniquely match the corresponding target points, which can be regarded as the target points of the hazardous objects.

[0046] Based on the further improvement of the above method, the target point is obtained by the following formula, which is specifically expressed as:

[0047]

[0048] In the formula,

[0049] LonT represents the longitude of the target point, LatT represents the latitude of the target point, lonMag is the longitude of the magnetic detection position point in the matching point pair, latMag is the latitude of the magnetic detection position point in the matching point pair, ConMag is the confidence of the magnetic detection position point in the matching point pair, lonPic is the longitude of the optical detection position point in the matching point pair, latPic is the latitude of the optical detection position point in the matching point pair, and ConPic is the confidence of the optical detection position point in the matching point pair;

[0050] The confidence of the target point is obtained by adding the confidence of the optical detection position point and the confidence of the magnetic detection position point in the matching point pair.

[0051] The beneficial effect of the above further improved solution is that the longitude and latitude of the corresponding position points of the matched optical detection position points and magnetic detection position points can be obtained through the above formula, so as to realize the positioning of the actual target point of the hazardous object.

[0052] Based on the further improvement of the above method, the grids are established with the target point as the center, specifically including:

[0053] Obtaining an estimated damage circle of the hazardous target based on a theoretical damage circle of the hazardous target and a target depth;

[0054] A square which is concentric with the estimated damage circle and has four sides tangent to the estimated damage circle is used as the grid.

[0055] The beneficial effect of the above further improvement scheme is that when performing secondary detection, the secondary detection range can be reduced as much as possible while ensuring safety, thereby improving the detection efficiency.

[0056] Based on the further improvement of the above method, the grid is established with the secondary detection point as the center, the grid is detected again, and the target point is found based on the re-detection result and stored in the target point set, which specifically includes:

[0057] Repeating S1-S5 for the grid, storing all target points obtained by re-detection and matching and light detection position points that do not form matching point pairs into a target point set;

[0058] Determine whether the confidence of the magnetic detection position point that does not form a matching point pair is lower than a second preset threshold, and if so, exclude the magnetic detection position point that does not form a matching point pair,

[0059] Otherwise, the magnetic detection position points that do not form matching point pairs are stored in the target point set.

[0060] The beneficial effect of the above-mentioned further improvement scheme is: based on the confidence of the light detection position point and the confidence of the magnetic detection position point obtained by secondary detection of the grid, the target points that can be matched and the light detection position points that match the target points are stored in the target point set, thereby avoiding detection omissions, and then excluding the magnetic detection position points that are below the second preset threshold and are not matched to the target points, thereby reducing the false alarm rate as much as possible and making the recognition result more reliable.

[0061] In the present invention, the above-mentioned technical solutions can also be combined with each other to achieve more preferred combination solutions. Other features and advantages of the present invention will be described in the subsequent description, and some advantages can become obvious from the description, or can be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained through the contents particularly pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The accompanying drawings are only used for the purpose of illustrating specific embodiments and are not to be considered as limiting the present invention. In the entire drawings, the same reference symbols represent the same components;

[0063] Figure 1 This is a flow chart of a method according to an embodiment of the present invention;

[0064] Figure 2 A schematic diagram of the correspondence between a light detection image and a real area according to an embodiment of the present invention;

[0065] Figure 3 A schematic diagram of determining magnetic detection position points according to an embodiment of the present invention;

[0066] Figure 4 Schematic diagram of the secondary detection area according to an embodiment of the present invention. DETAILED DESCRIPTION

[0067] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0068] A specific embodiment of the present invention discloses a method for identifying hazards by integrating multi-source information collected by drones, such as Figure 1 shown.

[0069] The method specifically comprises:

[0070] S1: The drone detects the target area and collects light detection data, magnetic detection data, and drone data corresponding to the light detection data and magnetic detection data respectively;

[0071] S2: Identify valid light detection data, and obtain all light detection position points and corresponding confidences based on the valid light detection data and its corresponding drone data, and obtain all magnetic detection position points and corresponding confidences based on the magnetic detection data and its corresponding drone data;

[0072] S3: combining each optical detection position point and each magnetic detection position point in a group one by one to calculate the distance of each combination, and then finding all combinations whose distances are less than a first distance threshold;

[0073] S4: Filter out the combination of magnetic detection position points with the shortest distance to each optical detection position point from all the combinations whose distance is less than the first distance threshold, and take the optical detection position point and the magnetic detection position point in each combination as a matching point pair;

[0074] S5: Get a target point and its confidence based on each matching point pair;

[0075] S6: Find the target point whose confidence level is not lower than the first preset threshold and store it in the target point set;

[0076] S7: storing the remaining target points, as well as the optical detection position points and the magnetic detection position points that do not form matching point pairs and whose confidence is not less than a second preset threshold, into a secondary detection point set;

[0077] S8: Establish a grid with the secondary detection point as the center, detect the grid again, and find the target point based on the re-detection result and store it in the target point set.

[0078] Generally, the hazards that need to be eliminated include the following:

[0079] 1. Located on the surface and not completely covered, it can be identified and located by light detection;

[0080] 2. Located on the surface and covered, optical detection and identification cannot confirm. If the hazardous object is a ferromagnetic object, magnetic detection and identification are combined with optical detection and identification to locate it;

[0081] 3. If the hazardous object is buried underground, optical detection and identification cannot locate it. Magnetic detection and identification are used to locate the hazardous object with ferromagnetic properties.

[0082] 4. If a hazardous object is placed underground, optical detection can identify the pits or holes caused by its placement and integrate it with magnetic detection to locate it.

[0083] The hazard detection is generally used to eliminate remaining hazards after training, or buried unknown hazards; among them, the approximate location and material properties of the hazards to be eliminated after training are known in advance; the location and material properties of the buried unknown hazards are unknown, and they are buried underground, making them difficult to identify by collecting image features, so they mostly rely on magnetic detection for identification.

[0084] The use of drones to detect target areas where hazards may exist is to install light detection equipment and magnetic detection equipment on the drone, and plan the route for the drone to detect based on the longitude and latitude of the target area to be detected. The route is determined based on the detection coverage of the light detection equipment and magnetic detection equipment installed on the drone. When the drone navigates in the target area according to the planned route, it can achieve full coverage detection of the target area and collect light detection data, magnetic detection data, and the drone's own data.

[0085] In step S1, the light detection data includes an optical image; the drone data corresponding to the light detection data includes the light detection viewing angle, the light detection image resolution, the drone height, the drone coordinates, and the drone head direction. It can be understood that the drone data refers to the light detection viewing angle, the light detection image resolution, the drone height, the drone coordinates, and the drone head direction when the optical image is collected, that is, each time the collection is performed, each optical image obtained will correspond to a drone data.

[0086] The magnetic detection data includes two-dimensional magnetic field value data; the drone data corresponding to the magnetic detection data includes drone coordinates. It can be understood that the drone data refers to the drone coordinates when the two-dimensional magnetic field value data is collected, that is, each two-dimensional magnetic field value data obtained each time the collection is performed will correspond to a drone coordinate.

[0087] Specifically, in step S2, the light detection data is identified through the following steps to find valid light detection data, and the corresponding light detection position point and confidence are found based on the valid light detection data and drone data. The steps specifically include:

[0088] After preprocessing, the optical image is input into the light detection recognition model for target recognition. If the target is recognized, the light detection data is used as valid light detection data, and an image with a target detection frame is output. The corresponding light detection position point is obtained based on the target detection frame and the drone data, and the confidence of the light detection position point is set to 1.

[0089] Specifically, the light detection data collected by the drone includes optical images; optionally, the equipment carried by the drone for optical collection can collect images and videos under various spectra based on the need for hazardous image detection, and the images and videos are processed into images to facilitate optical image recognition. The optical images collected by the drone are input into the light detection recognition model to identify the target. Preferably, the light detection recognition model adopts the YOLOv5 target detection algorithm, and can recognize based on the optical images collected by the drone by learning the appearance features of the hazardous objects, the pits or holes caused by the placement of hazardous objects, and other features, and output an image with a target detection frame.

[0090] The drone data includes light detection viewing angle, light detection image resolution, drone altitude, drone coordinates, and drone head direction.

[0091] Furthermore, obtaining the corresponding light detection position point based on the target detection frame and the drone data specifically includes:

[0092] The actual length and width corresponding to the light detection image are calculated based on the light detection viewing angle and the height of the drone;

[0093] Taking the upper left corner vertex of the light detection image as the origin, the normalized position of the pixel at the center point of the target detection frame in the light detection image relative to the center point of the light detection image is calculated based on the height of the drone;

[0094] Based on the normalized position, the direction of the drone head and the drone coordinates, the light detection position point is calculated.

[0095] Specifically, Figure 2 As shown, based on the light detection viewing angle and the height of the drone, the actual length and width of the target detection frame are calculated by the following formulas, which are specifically expressed as follows:

[0096] Where Widthl is the actual width of the light detection image, droneHeight is the relative height of the drone, Θ is the light detection viewing angle of the drone,

[0097] In the formula, Height1 is the field length corresponding to the light detection image, deviceHeight is the length of the light detection image, and deviceWidth is the width of the light detection image; the formula reflects the corresponding conversion relationship between the image with the target detection frame and the corresponding field size.

[0098] Furthermore, taking the upper left corner vertex of the light detection image as the origin and based on the height of the drone, the normalized position of the pixel at the center point of the target detection frame in the light detection image relative to the center point of the light detection image is calculated by the following formula, which is specifically expressed as:

[0099] Wherein, tmpX is the normalized horizontal coordinate of the position, and x is the horizontal coordinate of the center point of the target detection frame in the pixel of the light detection image;

[0100] Wherein, tmpY is the normalized position ordinate, and y is the pixel ordinate of the center point of the target detection frame in the light detection image. The normalized position of the pixel at the center point of the target detection frame in the light detection image relative to the center point of the light detection image is consistent with the normalized position of the field area corresponding to the light detection position point and the light detection image. Therefore, further, the corresponding light detection position point is calculated based on the normalized position, the drone head direction and the drone coordinates by the following formula, which is specifically expressed as:

[0101] xLng = droneLng + (tmpX × cos(droneHeading) - tmpY × sin(droneHeading)) × M_TO_LAT, where xLng is the longitude of the corresponding light detection position, droneLng is the corresponding drone longitude, droneHeading is the direction of the drone head, and M_TO_LAT is the longitude and latitude resolution;

[0102] yLat=droneLat+(tmpY×cos(droneHeading)+tmpX×sin(droneHeading))×MTO_LAT, where yLat is the latitude of the corresponding light detection position point, and droneLat is the corresponding drone latitude.

[0103] Generally, the nose direction of a drone refers to the horizontal angle between the flight direction of the drone and the direction of the drone's nose.

[0104] Before the UAV detects the target area, the horizontal center error and depth error of the magnetic detection are first obtained by simulating the target detection; in step S2, the magnetic detection position point and confidence are obtained based on the magnetic detection data through the following steps, and the steps specifically include:

[0105] Based on the two-dimensional magnetic field value data, the corresponding coordinates of the magnetic field maximum and minimum values, as well as the aeromagnetic dynamic noise, the repeat line accuracy, and the total accuracy of the aeromagnetic survey are obtained;

[0106] The midpoint of the line connecting the maximum and minimum coordinates of the magnetic field is taken as the magnetic detection position point;

[0107] The magnetic detection position point confidence is adjusted based on the initial value of the magnetic detection position point confidence based on the aeromagnetic dynamic noise, repeat line accuracy, total aeromagnetic accuracy, magnetic detection horizontal center error and depth error and obtained.

[0108] Specifically, through the magnetic detection equipment set on the drone, the drone continuously samples the geomagnetic field data at a set frequency while flying along the planned route. Since magnetic detection is greatly affected by the environment and terrain, in order to further ensure the accuracy of the magnetic detection results and reduce errors, before the drone formally detects, it is necessary to set up a simulated target in the target area for simulated detection. The specific detection method is basically the same as the formal detection method of the drone. The drone performs repeated simulated detection on the simulated target in the target area to collect magnetic detection data. Since the horizontal center point (i.e., the position point) and depth of the simulated detection target are known, the magnetic detection position point and depth are located based on the magnetic detection data, and the horizontal center error and depth error of the magnetic detection position point are calculated respectively, and recorded as deviation values, which are used to remove the deviation and calculate the confidence of the magnetic detection position point after obtaining the magnetic detection data of the formal detection.

[0109] Specifically, the magnetic detection depth error is calculated by the following formula, which is specifically expressed as:

[0110] Where PosDeltaV is the magnetic detection depth error, DepthS is the actual depth of the simulated target, and DepthS′ is the simulated target depth obtained by magnetic detection;

[0111] The horizontal center error of magnetic detection is calculated by the following formula, which is specifically expressed as:

[0112] Wherein, PosDeltaH is the horizontal center error of magnetic detection, LenS is the length of the simulated target, and distanceCal is the distance between the horizontal center point of magnetic detection and the actual horizontal center point of the simulated target. The distance between the horizontal center point of magnetic detection and the actual horizontal center point of the simulated target is calculated by the following formula, which is specifically expressed as follows: Where EARTH_RADIUS is the radius of the earth, deltaLat is the difference between the latitude of the horizontal center point of magnetic detection and the actual horizontal center point of the simulated target after conversion to a circular arc, deltaLon is the difference between the longitude of the horizontal center point of magnetic detection and the actual horizontal center point of the simulated target after conversion to a circular arc, LatS is the latitude of the actual horizontal center point of the simulated target, and LatS' is the latitude of the horizontal center point of magnetic detection.

[0113] The UAV removes the geomagnetic field data sampling points of take-off, landing, and turning, and then obtains the two-dimensional magnetic field value data through magnetic compensation and Kriging interpolation processing. The two-dimensional magnetic field value data are corrected based on the horizontal center error of the magnetic detection to ensure the reliability of the data. The purpose of the magnetic detection is to detect geomagnetic anomalies caused by harmful objects containing metal. Usually, each geomagnetic anomaly detected may contain harmful magnetic bodies. When the size of the harmful object is greatly different from the burial depth, the harmful magnetic body is equivalent to a sphere, and the coordinates of the center of the sphere are (0, 0, R). Then the two-dimensional magnetic field value data of the vertical magnetic anomaly profile of the sphere in the two-dimensional coordinate system (time domain is the horizontal axis, x is the vertical axis) is expressed as follows Figure 3 The curve shown in FIG. 1 is a graph showing a two-dimensional magnetic field value. The two-dimensional magnetic field value data expression is specifically expressed as:

[0114] In the formula, H az is the two-dimensional magnetic field data, M S is the projection of the total magnetization intensity vector M on the XZ plane in the X, Y, Z coordinate system, and x is H az The corresponding x-axis coordinate value, i S It is M s The angle with the X-axis direction, R is the burial depth of the hazardous material, μ 0 is a specific magnetic scalar potential value or initial magnetic scalar potential condition;

[0115] make Then we can get the maximum value coordinates and minimum value coordinates, and calculate the distance d between the maximum value coordinates and the minimum value coordinates by the following formula: m , the formula is specifically expressed as:

[0116] In the formula, x max is the maximum coordinate, x min are the minimum coordinates;

[0117] Based on the above formula, the formula for calculating the burial depth R of hazardous materials is further transformed, which is specifically expressed as:

[0118] R = f(i s )·d m ;

[0119] in,

[0120]

[0121] in,

[0122] Set the extreme point coordinate x max Substitute H az The expression can be obtained:

[0123]

[0124] After transformation, the expression of Ms is obtained, which is specifically expressed as:

[0125]

[0126] remember:

[0127]

[0128] Then K, f and φ are all i s The function values ​​of angles and their relationship to each other are shown in the following table:

[0129] 0° 15° 30° 45° 60° 75° 90° K 1.00 0.53 0.29 0.15 0.10 0.04 0.02 f 1.00 0.98 0.92 0.83 0.72 0.61 0.50 φ 0.43 0.56 0.70 0.82 0.92 0.98 1.00

[0130] Based on the two-dimensional magnetic field value data, d m And the extreme value ratio K, then look up the table based on the above table to get the corresponding i s value, and then obtain f(i s ) and φ(i s ) is substituted into the two-dimensional magnetic field value data expression to calculate the hazardous buried depth R. The hazardous buried depth is further corrected based on the magnetic detection depth error to obtain the corrected hazardous buried depth, so as to improve the reliability of the data.

[0131] The midpoint of the line connecting the maximum and minimum coordinates of the magnetic field is used as the magnetic detection position point; specifically, the coordinates returned by the drone obtaining the maximum and minimum values ​​of the magnetic field can correspond to the maximum and minimum values ​​of the magnetic field, and the midpoint coordinates of the line connecting the maximum and minimum values ​​of the magnetic field are used as the coordinates of the magnetic detection position point.

[0132] The aeromagnetic dynamic noise, repeat line accuracy and aeromagnetic accuracy are quality indicators of flight magnetic detection.

[0133] Specifically, when the drone is in flight, the magnetic field value is sampled at a maximum interval of 0.5 seconds to obtain a magnetic field curve, the magnetic field value of the sampling point on the anomaly with horizontal gradient>600nT / km is removed, and the aeromagnetic dynamic noise is calculated by the following formula, which is specifically expressed as:

[0134] Where n is the number of sampling points, B i is the fourth-order difference value of the magnetic field data Ti at the i-th sampling point, which is specifically expressed as the following formula:

[0135] B i =T i-2 -4T i-1 +6T i -4T i+1 +6T i+2 ,;

[0136] It is the sum of the fourth-order difference values ​​of the magnetic field data Ti at n sampling points, which can be specifically expressed as:

[0137]

[0138] The repeat line accuracy is the error obtained by calculating the root mean square of the magnetic detection data collected by the drone through selecting no less than 1% of the original planned route for re-flight and the magnetic detection data collected during the first flight, and is expressed as:

[0139] In the formula, This is the first flight to collect magnetic detection data. Fly again to collect magnetic detection data.

[0140] The total accuracy of the aerial magnetic survey is the sum of the measurement and positioning errors of the magnetic detection equipment installed on the UAV and the errors such as insufficient or inaccurate correction. The mean square error of the magnetic field difference at the intersection of the control line and the survey line is calculated by the following formula, which is specifically expressed as:

[0141] Where σ is the total accuracy of aeromagnetic survey, δ i is the magnetic field difference at the intersection of the i-th control line and the survey line, and m is the number of intersections between the control line and the survey line.

[0142] Specifically, for each magnetic detection position point, its initial confidence value is set to 0.5. Based on the aeromagnetic dynamic noise, repeat line accuracy, and total aeromagnetic accuracy corresponding to each magnetic detection position point, the confidence of each magnetic detection position point is adjusted according to the following principles based on the initial confidence value of the magnetic detection position point, and the principles specifically include:

[0143] When the corresponding aeromagnetic dynamic noise is ≤0.08nT, the confidence level of the magnetic detection position point is +0.2;

[0144] When 0.08nT<corresponding aeromagnetic dynamic noise≤0.14nT, the confidence level of the magnetic detection position point is +0.15;

[0145] When 0.15nT<corresponding aeromagnetic dynamic noise≤0.2nT, the confidence level of the magnetic detection position point is +0.1;

[0146] When 0.2nT<corresponding aeromagnetic dynamic noise≤1nT, the confidence of the magnetic detection position point remains unchanged;

[0147] When the corresponding aeromagnetic dynamic noise is greater than 1nT, the confidence level of the magnetic detection position point is -1;

[0148] When the geological structure of the magnetic detection environment is simple and the corresponding repeat line accuracy is less than 1nT, or the geological structure of the magnetic detection environment is complex and the corresponding repeat line accuracy is less than 3nT, the confidence of the detection position point is increased by 0.1 or 0.2;

[0149] When the geological structure of the magnetic detection environment is simple and the corresponding repeat line accuracy is ≥1nT, the geological structure of the magnetic detection environment is complex and the corresponding repeat line accuracy is ≥3nT, or the confidence of the detection position point is reduced by 0.1 or 0.2;

[0150] When the corresponding total accuracy of the aeromagnetic survey is less than 2nT, the confidence level of the magnetic detection position point is +0.1;

[0151] When the corresponding total accuracy of aeromagnetic survey is ≥2nT, the confidence level of the magnetic detection position point is -0.1;

[0152] When the horizontal center error of magnetic detection is ≤5% and the depth error is ≤10%, the confidence level of the magnetic detection position point is +0.1;

[0153] When the horizontal center error of the magnetic detection is greater than 5% and the depth error is greater than 10%, the confidence level of the magnetic detection position point is +0.1.

[0154] Through the confidence calculation of magnetic detection location points, not only the impact of environmental factors on magnetic detection data is fully considered and reduced as much as possible, but also the quantitative evaluation of the reliability of magnetic detection data is simplified, making the magnetic detection location points closer to the actual situation.

[0155] After the UAV completes the initial detection based on the planned route and obtains the optical detection position points and confidences, as well as the magnetic detection position points and confidences, it is necessary to further integrate and evaluate the optical detection position points and confidences, and the magnetic detection position points and confidences, and verify each other in order to find the matching correspondence between the optical detection position points and the magnetic detection position points, so as to accurately locate the target point of the hazard.

[0156] In step S3, in order to obtain the matching correspondence between the optical detection position points and the magnetic detection position points, each optical detection position point and each magnetic detection position point are combined into a group to calculate the distance of each combination, and then all combinations whose distance is less than the first distance threshold are found.

[0157] For the horizontal coordinate distances of the light detection position points and the magnetic detection position points obtained by detecting the same hazardous object, even if the hazardous object is drilled into the ground, it is highly likely that they are relatively close. Therefore, first, the light detection position points and the magnetic detection position points whose mutual distances are less than the first distance threshold are selected. Preferably, the first distance threshold is 1 meter.

[0158] Specifically, in step S4, a combination of magnetic detection position points with the shortest distance to each optical detection position point is selected from all combinations whose distance is less than the first distance threshold, and the optical detection position point and the magnetic detection position point in each combination are used as a matching point pair, specifically including:

[0159] S81, arranging all the combinations whose distances are less than the first distance threshold in order of distance from small to large;

[0160] The optical detection position point and the magnetic detection position point which are closest to each other in the distance queue are regarded as a matching point pair;

[0161] S82: Eliminate all combinations of optical detection position points or magnetic detection position points in the matching point pair from the distance queue, update the distance queue, and return to repeat S81 to S82 until the distance queue is empty.

[0162] Specifically, the found combination of light detection position points and magnetic detection position points that is less than the first distance threshold may exist in the case of one point to multiple points or multiple points to multiple points, which is inconsistent with the actual situation. It is necessary to further sort the combination distances from small to large, starting from the minimum distance of the sorting, and find out the unique corresponding combination of light detection position points and magnetic detection position points one by one, and exclude other combinations. Specifically, the light detection position points and magnetic detection position points corresponding to the minimum distance of the sorting are taken out from the distance queue as the unique corresponding combination of the two and saved separately, and other combinations including the light detection position points or magnetic detection position points corresponding to the minimum distance are eliminated from the distance queue, and then the remaining light detection position points and magnetic detection position points in the distance queue are repeatedly executed to find out the minimum distance combination, and eliminate other combinations of light detection position points and magnetic detection position points corresponding to the minimum distance from the distance queue, until the distance queue is clear.

[0163] Next, in step S5, the combination of the optical detection position point and the magnetic detection position point found needs to be fused and processed into the target point coordinates corresponding to the same hazardous object. Then, the target point is obtained by the following formula, which is specifically expressed as:

[0164]

[0165] In the formula,

[0166] LonT represents the longitude of the target point, LatT represents the latitude of the target point, lonMag is the longitude of the magnetic detection position point in the matching point pair, latMag is the latitude of the magnetic detection position point in the matching point pair, ConMag is the confidence of the magnetic detection position point in the matching point pair, lonPic is the longitude of the optical detection position point in the matching point pair, latPic is the latitude of the optical detection position point in the matching point pair, and ConPic is the confidence of the optical detection position point in the matching point pair;

[0167] The confidence of the target point is obtained by adding the confidence of the optical detection position point and the confidence of the magnetic detection position point in the matching point pair.

[0168] Then, in step S6, target points whose confidence is not less than a first preset threshold are further found from the target points and stored in the target point set. The first preset threshold is preferably 1.2. Through the dual verification of distance and confidence, a batch of hazardous target points can be found first. The hazardous target points meet the requirements of being close enough to the optical detection position points and the magnetic detection position points, and the optical detection image can identify the high confidence verified by the fusion of image features and magnetic detection data, and can basically locate the hazardous target points.

[0169] Next, in step S7, the remaining target points, as well as the optical detection position points and magnetic detection position points that are not matched to the target points and whose confidence is not less than the second preset threshold, are stored in the secondary detection point set.

[0170] Specifically, for the target point whose confidence is lower than the first preset threshold, since the confidence of the optical detection position point is 1, the confidence of the magnetic detection position point may not be high due to environmental factors, the hazardous object is buried too deeply, etc., and an initial value lower than the confidence of the magnetic detection position point is set, and a secondary detection of the target point is required for further confirmation; in addition, in the combination of optical detection position points and magnetic detection position points whose distance is greater than the first distance threshold, although it is impossible to determine whether the optical detection position point and the magnetic detection position point belong to the same hazardous object due to the long distance, a single optical detection position point or a magnetic detection position point has a confidence higher than the second preset threshold, and the second preset threshold is preferably 0.6, which reflects the great possibility of the existence of hazardous objects and also needs to be further confirmed by secondary detection.

[0171] Furthermore, in step S8, a grid is established with the secondary detection point as the center, the grid is detected again, and the target point is found based on the re-detection result and stored in the target point set.

[0172] Compared with the first detection, the secondary detection points and corresponding areas are relatively smaller. In order to further improve the detection efficiency and accuracy, the drone can be set to lower the flight altitude and speed, and a grid can be established based on the secondary detection points to further narrow the search range.

[0173] Specifically, Figure 4 As shown, the grids are established with the target point as the center, specifically including:

[0174] Obtaining an estimated damage circle of the hazardous target based on a theoretical damage circle of the hazardous target and a target depth;

[0175] A square which is concentric with the estimated damage circle and has four sides tangent to the estimated damage circle is used as the grid.

[0176] For eliminating the remaining hazards after training, the hazard type is basically known, so the theoretical range of the hazard target can be known through the hazard type. The theoretical range of the hazard target can be regarded as a circle, and the actual damage range is related to the depth of the hazard target. Generally, the deeper the burial, the smaller the damage range. Therefore, based on the hazard target theoretical damage circle and the target depth, the damage circle of the hazard target can be estimated, and then based on the multiples of the radius of the hazard target damage circle, preferably 10 times, it is expanded outward to obtain the estimated damage circle of the hazard target; then based on the estimated damage circle of the hazard target, the circumscribed quadrilateral of the circle is obtained as the grid corresponding to the secondary detection point.

[0177] Furthermore, the step of establishing a grid with the secondary detection point as the center, detecting the grid again, and finding the target point based on the re-detection result and storing it in the target point set specifically includes:

[0178] All target points obtained by re-detection and matching and light detection position points that do not form matching point pairs are stored in the target point set;

[0179] Determine whether the confidence of the magnetic detection position point that does not form a matching point pair is lower than a second preset threshold, and if so, exclude the magnetic detection position point that does not form a matching point pair,

[0180] Otherwise, the magnetic detection position points that do not form matching point pairs are stored in the target point set.

[0181] Specifically, the drone detects each secondary detection point grid, reacquires and identifies the light detection position point and confidence, the magnetic detection position point and confidence, and based on the first distance threshold, finds the unique match between the light detection position point and the magnetic detection position point and obtains the corresponding target point and the target point confidence.

[0182] During implementation, in order to obtain a more reliable detection effect, in addition to reducing the altitude and speed, the drone can also plan a route around the grid, change the light detection perspective, and collect optical images and magnetic field data of each grid from different directions. Among them, for multi-directional optical images collected from the same grid, as long as a light detection position point is identified from any one of them, it is considered to meet the characteristics of the existence of a hazardous object, and the confidence of the light detection position point is set to 1; if the multi-directional optical images collected from the same grid identify multiple light detection position points, it can be further identified whether it is the same hazardous object or there are multiple hazardous objects, and the light detection position points identified as the same hazardous object are merged into the same position point, and multiple hazardous light detection position points are marked and saved separately.

[0183] The target points obtained by the secondary detection and the optical detection position points that are not matched to the target points are all included in the target point set, that is, as long as the optical detection identifies the characteristics of the hazardous object and obtains the position points, they are regarded as the existence of hazardous objects for the purpose of avoiding omissions; for the magnetic detection position points that are not matched to the target points, although they are buried under the surface and the optical detection recognition cannot find the optical image features that match the existence of hazardous objects, if there are hazardous objects, generally, even if there is environmental interference, the difference between the magnetic detection position points and the confidences obtained by the first detection and the secondary detection will not be very large. Therefore, it is judged whether the confidence of the magnetic detection position points that are not matched to the target points is lower than the second preset threshold. If so, the magnetic detection position points of the unmatched target points are excluded. Otherwise, the magnetic detection position points of the unmatched target points are stored in the target point set. This processing method can not only effectively exclude the position points with too low confidence in the secondary magnetic detection and avoid retaining pseudo target points, but also further verify and confirm the magnetic detection position points with high confidence in the first detection, so that the detection results are more reliable.

[0184] The present embodiment discloses a method for identifying hazardous objects by integrating multi-source information collected by a drone. The method identifies optical detection data and magnetic detection data collected by the drone for the first time, eliminates invalid data, and obtains optical detection position points and confidence levels, magnetic detection position points and confidence levels. Then, corresponding target points are obtained based on combination matching of all distance combinations of optical detection position points and magnetic detection position points that are less than a first distance threshold. Then, a batch of hazardous object target points are first found through a first preset threshold, and then a secondary detection of the remaining target points with a narrowed range is performed to find the remaining hazardous object target points. Compared with the prior art, the method not only improves the efficiency and reliability of detection, but also greatly simplifies the processing process.

[0185] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0186] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A hazard identification method integrating multi-source information collected by drones, It is characterized in that The method specifically comprises: S1: The drone detects the target area and collects light detection data, magnetic detection data, and drone data corresponding to the light detection data and magnetic detection data respectively; S2: Identify valid light detection data, and obtain all light detection position points and corresponding confidences based on the valid light detection data and its corresponding drone data, and obtain all magnetic detection position points and corresponding confidences based on the magnetic detection data and its corresponding drone data; S3: combining each optical detection position point and each magnetic detection position point in a group one by one to calculate the distance of each combination, and then finding all combinations whose distance is less than a first distance threshold; S4: Filter out the combination of magnetic detection position points with the shortest distance to each optical detection position point from all the combinations whose distance is less than the first distance threshold, and take the optical detection position point and the magnetic detection position point in each combination as a matching point pair; S5: Get a target point and its confidence based on each matching point pair; S6: Find the target point whose confidence level is not lower than the first preset threshold and store it in the target point set; S7: storing the remaining target points, as well as the optical detection position points and the magnetic detection position points that do not form matching point pairs and whose confidence is not less than a second preset threshold, into a secondary detection point set; S8: Establish a grid with the secondary detection point as the center, detect the grid again, and find the target point based on the re-detection result and store it in the target point set.

2. According to claim 1, a method for identifying hazards by integrating multi-source information collected by drones, It is characterized in that The light detection data includes an optical image; The identifying and finding out the valid light detection data, and obtaining the light detection position point and confidence level based on the valid light detection data and the drone data, specifically includes: After preprocessing, the optical image is input into the light detection recognition model for target recognition. If the target is recognized, the light detection data is used as valid light detection data, and an image with a target detection frame is output. The corresponding light detection position point is obtained based on the target detection frame and the drone data, and the confidence of the light detection position point is set to 1.

3. According to claim 2, a method for identifying hazards by integrating multi-source information collected by drones, It is characterized in that The drone data corresponding to the light detection data includes the light detection viewing angle, the light detection image resolution, the drone height, the drone coordinates, and the drone head direction; the corresponding light detection position point is obtained based on the target detection frame and the drone data, specifically including: The actual length and width corresponding to the light detection image are calculated based on the light detection viewing angle and the height of the drone; Taking the upper left corner vertex of the light detection image as the origin, the normalized position of the pixel at the center point of the target detection frame in the light detection image relative to the center point of the light detection image is calculated based on the height of the drone; Based on the normalized position, the direction of the drone head and the drone coordinates, the light detection position point is calculated.

4. According to claim 3, a method for identifying hazards by integrating multi-source information collected by drones, It is characterized in that Based on the light detection angle of view and the height of the drone, the actual length and width of the target detection frame are calculated by the following formulas, which are specifically expressed as follows: Where Width1 is the actual width of the light detection image, droneHeight is the relative height of the drone, Θ is the light detection viewing angle of the drone, Wherein, Height1 is the actual length corresponding to the light detection image, deviceHeight is the length of the light detection image, and deviceWidth is the width of the light detection image; The upper left corner vertex of the light detection image is taken as the origin, based on the height of the drone, the normalized position of the pixel at the center point of the target detection frame in the light detection image relative to the center point of the light detection image is calculated by the following formula, and the formula is specifically expressed as: Wherein, tmpX is the normalized horizontal coordinate of the position, and x is the horizontal coordinate of the center point of the target detection frame in the pixel of the light detection image; Wherein, tmpY is the normalized position ordinate, and y is the pixel ordinate of the center point of the target detection frame in the light detection image.

5. According to claim 4, a method for identifying hazards by integrating multi-source information collected by drones, It is characterized in that Based on the normalized position, the direction of the drone head and the drone coordinates, the corresponding light detection position point is calculated by the following formula, which is specifically expressed as: xLng = droneLng + (tmpX × cos(droneHeading) - tmpY × sin(droneHeading)) × M_TO_LAT, where xLng is the longitude of the corresponding light detection position, droneLng is the corresponding drone longitude, droneHeading is the direction of the drone head, and M_TO_LAT is the longitude and latitude resolution; yLat=droneLat+(tmpY×cos(droneHeading))+tmpX×sin(droneHeading))×M_TO_LAT, where yLat is the latitude of the corresponding light detection position point, and droneLat is the corresponding drone latitude.

6. According to claim 1, a method for identifying hazards by integrating multi-source information collected by drones, It is characterized in that The magnetic detection data includes two-dimensional magnetic field value data; before the UAV detects the target area, the horizontal center error and depth error of the magnetic detection are obtained by simulating target detection; The magnetic detection position point and confidence level are obtained based on the magnetic detection data through the following steps, the steps specifically comprising: Based on the two-dimensional magnetic field value data, the corresponding coordinates of the magnetic field maximum and minimum values, as well as the aeromagnetic dynamic noise, the repeat line accuracy, and the total accuracy of the aeromagnetic survey are obtained; The midpoint of the line connecting the maximum and minimum coordinates of the magnetic field is taken as the magnetic detection position point; The magnetic detection position point confidence is adjusted based on the initial value of the magnetic detection position point confidence based on the aeromagnetic dynamic noise, repeat line accuracy, total aeromagnetic accuracy, magnetic detection horizontal center error and depth error and obtained.

7. According to claim 1, a method for identifying hazards by integrating multi-source information collected by drones, It is characterized in that Selecting a combination of magnetic detection position points with the shortest distance to each optical detection position point from all combinations whose distance is less than the first distance threshold, and taking the optical detection position point and the magnetic detection position point in each combination as a matching point pair, specifically includes: S81, arranging all the combinations whose distances are less than the first distance threshold in order of distance from small to large; The optical detection position point and the magnetic detection position point which are closest to each other in the distance queue are regarded as a matching point pair; S82: Eliminate all combinations of optical detection position points or magnetic detection position points in the matching point pair from the distance queue, update the distance queue, and return to repeat S81 to S82 until the distance queue is empty.

8. According to claim 1, a method for identifying hazards by integrating multi-source information collected by drones, It is characterized in that The target point is obtained by the following formula, which is specifically expressed as: In the formula, LonT represents the longitude of the target point, LatT represents the latitude of the target point, LonMag is the longitude of the magnetic detection position point in the matching point pair, latMag is the latitude of the magnetic detection position point in the matching point pair, ConMag is the confidence of the magnetic detection position point in the matching point pair, lonPic is the longitude of the optical detection position point in the matching point pair, latPic is the latitude of the optical detection position point in the matching point pair, and ConPic is the confidence of the optical detection position point in the matching point pair; The confidence of the target point is obtained by adding the confidence of the optical detection position point and the confidence of the magnetic detection position point in the matching point pair.

9. The method for identifying hazards by integrating multi-source information collected by drones according to claim 8, It is characterized in that The grids are established with the target point as the center, specifically including: Obtaining an estimated damage circle of the hazardous target based on a theoretical damage circle of the hazardous target and a target depth; A square which is concentric with the estimated damage circle and has four sides tangent to the estimated damage circle is used as the grid.

10. The method for identifying hazards by integrating multi-source information collected by drones according to claim 1, It is characterized in that The step of establishing a grid with the secondary detection point as the center, detecting the grid again, and finding the target point based on the detection result again and storing it in the target point set specifically includes: Execute S1-S5 on the grid, and store all target points obtained by re-detection and matching and light detection position points that do not form matching point pairs into a target point set; Determine whether the confidence of the magnetic detection position points that do not form a matching point pair is lower than a second preset threshold; if so, exclude the magnetic detection position points that do not form a matching point pair; otherwise, store the magnetic detection position points that do not form a matching point pair in the target point set.

Images