Unmanned aerial vehicle end height adaptive target detection method and system
By calculating the target distance on the drone and adjusting the number of detection network layers, the problem of reducing the drone target detection accuracy is solved, and efficient detection and computing resource savings are achieved when altitude changes are achieved.
Patent Information
- Application Number
- CN202510298683.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-13
- Publication Date
- 2025-07-08
AI Technical Summary
When the existing drone target detection algorithm changes in flight altitude, the target detection accuracy decreases and the computing resource consumption increases. Multi-scale strategies fail when the altitude changes greatly, resulting in increased algorithm complexity and difficulty in promoting and applying.
By calculating the target distance based on the drone height, camera loading height and pitch angle, adjusting the network layer number and structure of the target detection algorithm, deleting the oversampling layer, and adjusting the level of the detection head using height information to enhance detection capabilities and reduce computing resource consumption.
It improves the accuracy of drone target detection, reduces computing resource consumption, and achieves efficient target detection when height changes.
Smart Images
Figure CN120279244A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the fields of aerial detection and surveillance of unmanned aerial vehicles and target detection technology in computer vision; specifically, it relates to a method and system for height-adaptive target detection at the unmanned aerial vehicle end. Background Art
[0002] With the support of artificial intelligence technology, the field of computer vision has made great progress. The average precision (mAP) of general target detection algorithms on public datasets has reached 70%+ or even higher. After the appearance of the RCNN algorithm, the accuracy of general target detection algorithms on the Pascal-VOC2007 dataset (http: / / host.robots.ox.ac.uk / pascal / VOC / ) was directly increased to 58.5%. After RCNN, general target detection algorithms have emerged continuously. In 2018, the highest precision rate on the Pascal-VOC2007 public dataset was RefineDet, reaching 83.8%. The highest precision rate on the Pascal-VOC2012 (http: / / host.robots.ox.ac.uk / pascal / VOC / ) dataset reached 83.5%. The highest accuracy on the COCO public dataset also reached 62.9%, and this accuracy was further increased to 69.7% by TridentNet in 2019. The best target detection algorithm on the Pascal-VOC2007 public dataset is Cascade Eff-B7NAS-FPN, with an accuracy of 89.3%. As of February 2024, for the precision rate of current various target detection algorithms on the publicly available COCO test-dev dataset, Co-DETR proposed in 2023 achieved the highest correct rate of 66% box mAP on this dataset. If mAP@0.5 is used as the standard, the EVA algorithm proposed in 2023 can even reach 81.9%.
[0003] The improvement of the performance of general target detection algorithms has reached a bottleneck period. It is very difficult to increase the average precision by even 0.1%, while the algorithm complexity increases exponentially. In the process of algorithm application, this method of improving the average precision has extremely low cost performance and no practical significance, resulting in difficulties in promoting the application of the algorithm. Target detection algorithms based on deep neural networks often face the problem that when the flight height of the unmanned aerial vehicle changes, the distance from the optoelectronic sensor to the target changes, resulting in scale changes of the target, and further leading to a decrease in the accuracy of target detection. Although most current target detection algorithms adopt multi-scale strategies to cope with the problem of reduced target detection accuracy caused by changes in the flight height of unmanned aerial vehicles within a certain range, when the height change is relatively large, the multi-scale strategy will fail and consume computing resources. Summary of the Invention
[0004] To solve the above technical problems, the present invention provides a method for height - adaptive target detection on an unmanned aerial vehicle (UAV), including:
[0005] Determine a first distance between the target object and the camera based on the UAV height, the height of the UAV - mounted camera, and the pitch angle;
[0006] Use the UAV - mounted camera to collect an initial image and an image after height adjustment of the target object;
[0007] Determine the number of network layers required for the target detection algorithm of the UAV - mounted camera corresponding to the target object based on the initial image and the image after height adjustment;
[0008] Process the network structure based on the number of network layers required for the target detection algorithm;
[0009] Perform UAV target detection using the processed network structure.
[0010] Preferably, the determining the first distance between the target object and the camera based on the UAV height, the height of the UAV - mounted camera, and the pitch angle includes:
[0011] Obtain the flight height of the UAV using the UAV - mounted pose sensor;
[0012] Obtain the pitch angle of the camera using the UAV - mounted camera sensor;
[0013] Calculate the first distance between the target object and the camera using the distance formula based on the installation height difference between the camera and the UAV, the flight height of the UAV, and the pitch angle.
[0014] Preferably, before using the UAV - mounted camera to collect the initial image of the target object, it includes: initially calibrating the UAV and the camera focal length.
[0015] Preferably, the determining the number of network layers in the target detection algorithm of the UAV - mounted camera corresponding to the target object based on the initial image and the image after height adjustment includes:
[0016] Calculate the first maximum side - length pixels of the target object based on the initial image and the initial calibration, and measure the first maximum side - length of the target object in the initial image;
[0017] Measure the second maximum side - length of the target object for the image after height adjustment, and calculate the second distance between the target object and the camera using the distance formula;
[0018] Calculate the second maximum side length image corresponding to the height adjustment based on the first maximum side length pixel, the first maximum side length, the second maximum side length, the first distance, and the second distance;
[0019] Determine the number of network layers in the drone-mounted camera target detection algorithm corresponding to the target object based on the second maximum side length image.
[0020] Preferably, the processing of the network structure based on the number of network layers includes:
[0021] Let the number of network layers be n;
[0022] When the number of network layers n is less than the number of sampling layers set by the target detection algorithm of the deep neural network, delete the detection head and the corresponding network layer corresponding to the nth and deeper sampling layers in the target detection algorithm, otherwise do not perform network structure adjustment.
[0023] Preferably, the drone target detection using the processed network structure includes:
[0024] When network structure adjustment is performed:
[0025] Upsample the feature map corresponding to the shallowest detection head in the current detection network;
[0026] Fuse the upsampled feature map with the feature map of the network layer that has not been fused with the upsampled layer feature map in the shallower layer, and add a new detection head to detect the fused feature map.
[0027] Based on the same inventive concept, the present application also provides a drone-side height adaptive target detection system, including:
[0028] A first distance calculation module, configured to determine a first distance between the target object and the camera based on the drone height, the drone-mounted camera height, and the pitch angle;
[0029] An image acquisition module, configured to use the drone-mounted camera to acquire an initial image and a height-adjusted image of the target object;
[0030] A network layer number calculation module, configured to determine the number of network layers in the drone-mounted camera target detection algorithm corresponding to the target object based on the initial image and the height-adjusted image;
[0031] A network structure processing module, configured to process the network structure based on the number of network layers;
[0032] A detection module, configured to perform drone target detection using the processed network structure.
[0033] Preferably, the network structure processing module is specifically configured to: set the number of network layers as n; when the number of network layers n is less than the number of sampling layers set by the target detection algorithm of the deep neural network, delete the detection head and the corresponding network layer corresponding to the nth and deeper sampling layers in the target detection algorithm, otherwise do not perform network structure adjustment.
[0034] The detection module is specifically configured to: when network structure adjustment is performed: upsample the feature map corresponding to the shallowest detection head in the current detection network; fuse the upsampled feature map with the feature map of the network layer that is shallower and has not been fused with the upsampled layer feature map, and add a new detection head to detect the fused feature map.
[0035] Based on the same inventive concept, the present application also provides an electronic device, including: at least one processor and a memory; the memory and the processor are connected by a bus;
[0036] The memory is used to store one or more programs;
[0037] When the one or more programs are executed by the at least one processor, an unmanned aerial vehicle (UAV)-side height adaptive target detection method provided by the present application is implemented.
[0038] Based on the same inventive concept, the present application also provides a readable storage medium, on which an execution program is stored, and when the execution program is executed, an unmanned aerial vehicle (UAV)-side height adaptive target detection method provided by the present application is implemented.
[0039] Compared with the closest prior art, the beneficial effects of the present invention are:
[0040] 1. The present invention provides an unmanned aerial vehicle (UAV)-side height adaptive target detection method and system, including: determining a first distance between a target object and the camera based on the UAV height, the height of the UAV-mounted camera, and the pitch angle; using the UAV-mounted camera to collect an initial image of the target object and an image after height adjustment; determining the number of network layers required by the target detection algorithm of the UAV-mounted camera corresponding to the target object based on the initial image and the image after height adjustment; processing the network structure based on the required number of network layers; the present application introduces UAV height information into the target detection network, estimates the scale of the target in the image using the UAV height, deletes over-sampled network layers and detection heads using the scale information of the target in the image and the number of samplings of the deep neural network, and only retains the detection heads at the target scale, enabling the limited computing power on the UAV to be concentrated on the detection heads consistent with the target scale, and reducing the amount of calculation to a certain extent while maintaining the detection accuracy.
[0041] 2. While using the height d information of the target to delete the over-sampled network layers and detection heads, the present invention adds detection heads for the target to shallow network layers, enhancing the detection ability for the target. Description of the Drawings
[0042] Figure 1 is a flowchart of a method for height-adaptive target detection on a drone terminal according to the present invention;
[0043] Figure 2 is the overall flowchart of the present invention;
[0044] Figure 3 is a schematic diagram of the modification of the network structure according to the present invention;
[0045] Figure 4 is the network structure diagram before modification in an embodiment of the present invention;
[0046] Figure 5 is the network structure diagram after modification in the first embodiment of the present invention;
[0047] Figure 6 is the network structure diagram after modification in the second embodiment of the present invention;
[0048] Figure 7 is the structure diagram of a height-adaptive target detection system on a drone terminal according to the present invention;
[0049] Figure 8 is the structure diagram of an electronic device according to the present invention;
[0050] Among them, Input is the input layer, stride is the step size, Block is the block, Maxpool is the max pooling, Conv is the convolutional layer, batchnorm is the normalization layer, leaky is the leaky activation layer, linear is the linear activation layer, yolo-head is the detection head, route is the routing layer, and Upsamle is the upsampling. Detailed Embodiments
[0051] A method for height - adaptive target detection on an unmanned aerial vehicle (UAV) side provided by the present invention calculates the distance between a ground or water surface target and a camera by using the height information of the UAV itself and the attitude angle information of the camera obtained by a pose sensor on the UAV. Combining the actual size of the target and the internal parameters of the camera, the number of pixels occupied by the target in the image collected by the camera can be calculated. According to the number of pixels of the target, the number of pixels occupied by the image on each feature map after inputting the target detection depth neural network can be estimated. Thus, the feature map with the number of corresponding target pixels greater than a specific value can be input into the detection head of the detection model, and anchor boxes conforming to the number of target pixels are set at this level. By doing so, a single - scale target detection network can adaptively change the level where the detection head is located according to the target height, enabling it to have the ability to detect multi - scale targets, not only improving the detection accuracy but also reducing the consumption of computing resources on the UAV. The technical solution of the present invention will be further described below with reference to the accompanying drawings, but the present invention is not limited to these embodiments.
[0052] Embodiment 1:
[0053] This application provides a method for height - adaptive target detection on a UAV side, as Figure 1 shown, including:
[0054] S1. Determine the first distance between the target object and the camera based on the UAV height, the height of the UAV - mounted camera, and the pitch angle;
[0055] S2. Use the UAV - mounted camera to collect the initial image and the image after height adjustment of the target object;
[0056] S3. Determine the number of network layers required for the UAV - mounted camera target detection algorithm corresponding to the target object based on the initial image and the image after height adjustment;
[0057] S4. Process the network structure based on the number of network layers required for the target detection algorithm;
[0058] S5. Use the processed network structure for UAV target detection.
[0059] The specific process is as Figure 2 shown:
[0060] 1. Use the UAV - mounted pose sensor IMU to obtain the height H of the aircraft, use the UAV - mounted camera sensor to obtain the pitch angle θ of the camera, and the installation height difference between the camera and the UAV is h. Then, the distance between the target and the camera can be obtained
[0061] D = (H + h) / sinθ……………………………………………(1)
[0062] In the formula, D is the distance between the target and the camera;
[0063] 2. Simply calibrate the drone and the camera focal length. First, select a target and capture an image, and measure the maximum side length \(l_0\) pixels of the target in the image, and measure the actual maximum side length \(L\) of the target. o , and from formula (1), the distance \(D_0\) between the target and the camera can be obtained.
[0064]
[0065] Where: \(f\) is the camera focal length;
[0066] 3. When a new image is captured, the maximum side length of the target is \(L_1\), and from formula (1), the distance \(D_1\) between the target and the camera can be obtained.
[0067]
[0068] 4. From steps 2 and 3, the maximum side length pixel number \(l_1\) of the target in the new image can be obtained.
[0069]
[0070] 5. Define the parameter \(n\), and let \(2\) n \(< l_1\leq2\) n+1 , from which the value of \(n\) can be obtained (\(n\) is an integer). \(n\) is just a parameter that can determine the deepest layer number of the network (the network required for the target detection algorithm), and the network layers greater than this number can be deleted.
[0071] 6. The object detection algorithm based on the deep neural network has \(N\) (generally \(N\leq5\)) sampling layers (samplelayer), that is, the smallest feature map is \(\frac{1}{2^N}\) of the original image. We have obtained the value of \(n\) from step 5. If \(n\geq N\), the network structure does not need to be modified. If \(n < N\), the detection network needs to be deeply trimmed, and step 7 is executed.
[0072] 7. Delete the detection head and the corresponding network layer corresponding to the \(n\)th and deeper sampling layers among the \(N\) sampling layers.
[0073] 8. Upsample the feature map corresponding to the detection head of the shallowest layer (the highest feature map resolution) in the current detection network, and then fuse it with the feature map of the network layer that has not been fused with the upsampled layer feature map in the shallower layer, and add a new detection head to detect the fused feature map.
[0074] The schematic diagram of the network structure modification of the present invention is as Figure 3 shown.
[0075] Use the processed network structure to continuously track and detect the target object and the target with a size similar to that of the target object. When the flight altitude of the drone changes, readjust the network structure according to the above method.
[0076] Among them, objects with similar target sizes can be targets with the same or similar models or styles;
[0077] If the new target has a large size deviation from the original target, it can be targets of different types, such as cars, drones, etc.; or they can both be cars but are respectively sedans or coaches;
[0078] Of course, in order to improve the detection accuracy, targets of the same model can be considered as similar targets, such as SUV cars belonging to the same category. And a mini car and an SUV car are considered targets with a large deviation, etc. This application does not limit whether the targets are similar and can be set according to the actual needs of tracking detection.
[0079] Embodiment 2:
[0080] Taking the DJI M300 quadcopter drone as an example, an exemplary description is given of a method for height-adaptive target detection on the drone side provided by the present invention:
[0081] 1. Use the DJI M300 quadcopter drone equipped with the DJI Zenmuse H20 camera to detect ground / water vehicles. The length L of the ground target vehicle can be measured, and the distance D between the camera and the target vehicle can be obtained from step 1 of the technical solution.
[0082] 2. According to the parameters of the drone and the carried camera, f can be obtained according to step 2 of the technical solution.
[0083] 3. With the camera focal length unchanged, obtain an image of a new target vehicle.
[0084] 4. According to formula (4), the number of pixels l1 of the maximum side length of the target vehicle in the new image can be obtained.
[0085] 5. When l1 = 30, n = 4 can be obtained.
[0086] 6. The target detection algorithm deployed on the drone is yolov4-tiny-3l, and the network structure of this target detection algorithm is as Figure 4 shown. This network has 5 sampling layers, that is, N = 5. Since n ≤ N, directly execute step 7 of the technical solution.
[0087] 7. Prune the network of the target detection algorithm, and prune the network layers and detection heads with more than n (that is, 5 times) samplings. The network structure of the target detection network pruned according to this scheme is as Figure 5 shown. The sampling layer is the layer with a stride of 2, and the network layer with more than 4 samplings (that is, more than 4 layers with stride = 2) is Figure 5 The part in the blue box is the network structure that needs to be pruned.
[0088] 8. Upsample on the network layer corresponding to the current deepest detection head, fuse it with the feature map output by block1, and add a new detection head and network structure. The specific operation is to add a routing layer after the last convolution layer, normalization layer, and leaky activation layer module, add an upsampling layer after the module composed of the convolution layer, normalization layer, and leaky activation layer, and then input it into the routing layer together with the output of block1 for an addition operation, and then pass through the module composed of the convolution layer, normalization layer, and leaky activation layer, and finally enter the detection head (yolo-head) after passing through the module composed of the convolution layer, normalization layer, and linear activation layer. As Figure 5 shown, the network structure in the red box is the newly added one.
[0089] Example 3:
[0090] Steps 1 to 4 are the same as in Example 2.
[0091] 5. When l = 12, n = 3 can be obtained.
[0092] 6. The object detection algorithm deployed on the UAV is yolov4-tiny-3l. The network structure of this object detection algorithm is as Figure 4 shown. This network has 5 sampling layers, that is, N = 5. Since n ≤ N, step 7 of the technical solution is executed.
[0093] 7. Prune the network of the object detection algorithm, and prune the feature adjustment layers and detection heads with more than n (i.e., 3 times) samplings. The network structure of the object detection network pruned according to this solution is as Figure 6 shown. The network structure to be pruned is in the blue box.
[0094] 8. Upsample on the network layer corresponding to the current deepest detection head, fuse it with the feature map output by block1, and add a new detection head. As Figure 6 shown, the network structure in the red box is the newly added one.
[0095] Example 4:
[0096] To implement the above-mentioned UAV-side height-adaptive object detection method, the present application also provides a UAV-side height-adaptive object detection system, as Figure 7 shown, including:
[0097] The first distance calculation module is used to determine the first distance between the target object and the camera based on the UAV height, the height of the UAV-mounted camera, and the pitch angle;
[0098] The image acquisition module is used to collect the initial image and the height-adjusted image of the target object by using the UAV-mounted camera;
[0099] A network layer number calculation module, configured to determine the network layer number in the target detection algorithm of the airborne camera corresponding to the target object based on the initial image and the image after height adjustment;
[0100] A network structure processing module, configured to process the network structure based on the network layer number;
[0101] A detection module, configured to perform unmanned aerial vehicle (UAV) target detection by using the processed network structure.
[0102] Further, the network structure processing module is specifically configured to: set the network layer number as n; when the network layer number n is less than the number of sampling layers set in the target detection algorithm of the deep neural network, delete the detection head and the corresponding network layer corresponding to the nth and deeper sampling layers in the target detection algorithm, otherwise do not perform network structure adjustment.
[0103] The detection module is specifically configured to: when network structure adjustment is performed: perform upsampling on the feature map corresponding to the shallowest detection head in the current detection network; fuse the upsampling with the feature map of the network layer that has not been fused with the feature map of the upsampling layer in a shallower layer, and add a new detection head to detect the fused feature map.
[0104] The specific functions implemented by each functional module in this embodiment are as described in the above embodiments, and will not be elaborated here.
[0105] Embodiment 5:
[0106] As Figure 8 shown, the present invention further provides an electronic device, which may be a computer device, a single-chip microcomputer device, a smart mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, the processor, and the transceiver component are connected through a bus; the memory may be used to store an execution program, and an exemplary execution program may include instructions; the processor is configured to execute the instructions stored in the memory. The memory may also be used to store data, and the data may be called and / or modified when the instructions are executed.
[0107] The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the storage medium to implement the corresponding method flow or corresponding function, so as to implement the steps of a method for height adaptive target detection on a drone end in the above embodiments.
[0108] Embodiment 6
[0109] Based on the same inventive concept, the present invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory). The electronic device readable storage medium is a memory device in the electronic device, and is used to store programs and data. It can be understood that the storage medium here can include both the built-in storage medium in the electronic device, and of course can also include the extended storage medium supported by the electronic device. The storage medium provides a storage space, and this storage space stores the operating system of the terminal. And, one or more instructions suitable for being loaded and executed by the processor are also stored in this storage space. These instructions can be one or more execution programs (including program codes). It should be noted that the storage medium here can be a high-speed RAM memory, or a non-volatile memory, such as at least one disk memory. By loading and executing one or more instructions stored in the storage medium by the processor, the steps of a method for height adaptive target detection on a drone end in the above embodiments can be implemented.
Claims
1. A method for height - adaptive target detection on an unmanned aerial vehicle side, characterized in that, Comprising: Determining a first distance between the target object and the camera based on the height of the unmanned aerial vehicle (UAV), the height of the camera carried by the UAV, and the pitch angle; Collecting an initial image of the target object and an image after height adjustment by using the camera carried by the UAV; Determining the number of network layers required for the target detection algorithm of the camera carried by the UAV corresponding to the target object based on the initial image and the image after height adjustment; Processing the network structure based on the number of network layers required for the target detection algorithm; Performing UAV target detection by using the processed network structure.
2. The method according to claim 1, characterized in that, The determining a first distance between the target object and the camera based on the height of the UAV, the height of the camera carried by the UAV, and the pitch angle comprises: Obtaining the flight height of the UAV by using the pose sensor carried by the UAV; Obtaining the pitch angle of the camera by using the camera sensor carried by the UAV; Calculating the first distance between the target object and the camera by using the distance formula based on the installation height difference between the camera and the UAV, the flight height of the UAV, and the pitch angle.
3. The method according to claim 1, characterized in that Before collecting the initial image of the target object by using the camera carried by the UAV, initial calibration is performed on the UAV and the camera focal length.
4. The method according to claim 3, characterized in that, The determining the number of network layers in the target detection algorithm of the camera carried by the UAV corresponding to the target object based on the initial image and the image after height adjustment comprises: Calculating the first maximum side length pixel of the target object based on the initial image and the initial calibration, and measuring the first maximum side length of the target object in the initial image; Measuring the second maximum side length of the target object for the image after height adjustment, and calculating the second distance between the target object and the camera by using the distance formula; Calculating the second maximum side length image corresponding to the height adjustment based on the first maximum side length pixel, the first maximum side length, the second maximum side length, the first distance, and the second distance; Determining the number of network layers in the target detection algorithm of the camera carried by the UAV corresponding to the target object based on the second maximum side length image.
5. The method according to claim 1, characterized in that The processing the network structure based on the number of network layers comprises: Setting the number of network layers as n; When the number of network layers n is less than the number of sampling layers set by the target detection algorithm of the deep neural network, deleting the detection head corresponding to the nth and deeper sampling layers in the target detection algorithm and the corresponding network layers, otherwise not performing network structure adjustment.
6. The method according to claim 5, wherein The performing UAV target detection by using the processed network structure comprises: When network structure adjustment is performed: Upsampling the feature map corresponding to the shallowest detection head in the current detection network; Fusing the upsampled feature map with the feature map of the network layer that has not been fused with the upsampled layer feature map in the shallower layer, and adding a new detection head to detect the fused feature map.
7. An altitude self-adaptive target detection system on the UAV side, characterized in that, Comprising: A first distance calculation module, configured to determine a first distance between the target object and the camera based on the height of the UAV, the height of the camera carried by the UAV, and the pitch angle; An image acquisition module, configured to collect an initial image of the target object and an image after height adjustment by using the camera carried by the UAV; A network layer number calculation module, configured to determine the number of network layers in the target detection algorithm of the camera carried by the UAV corresponding to the target object based on the initial image and the image after height adjustment; A network structure processing module, configured to process a network structure based on the number of network layers; A detection module, configured to perform UAV target detection by using the processed network structure.
8. The system according to claim 7, wherein the network structure processing module is specifically configured to: set the number of network layers as n; when the number of network layers n is less than the number of sampling layers set by the target detection algorithm of the deep neural network, delete the detection head and the corresponding network layer corresponding to the nth and deeper sampling layers in the target detection algorithm, otherwise no network structure adjustment is performed; the detection module is specifically configured to: when network structure adjustment is performed: perform upsampling on the feature map corresponding to the shallowest detection head in the current detection network; fuse the upsampling with the feature map of the network layer that has not been fused with the upsampled layer feature map in a shallower layer, and add a new detection head to detect the fused feature map.
9. An electronic device, characterized in that, Comprising: at least one processor and a memory; the memory and the processor are connected through a bus; the memory is configured to store one or more programs; when the one or more programs are executed by the at least one processor, implement a UAV-side height adaptive target detection method according to any one of claims 1 to 6.
10. A readable storage medium, characterized in that, There is an execution program stored thereon, and when the execution program is executed, implement a UAV-side height adaptive target detection method according to any one of claims 1 to 6.