CN118115896A - Unmanned aerial vehicle detection method and system based on improvement YOLOv3 - Google Patents
Unmanned aerial vehicle detection method and system based on improvement YOLOv3 Download PDFInfo
- Publication number
- CN118115896A CN118115896A CN202410035199.5A CN202410035199A CN118115896A CN 118115896 A CN118115896 A CN 118115896A CN 202410035199 A CN202410035199 A CN 202410035199A CN 118115896 A CN118115896 A CN 118115896A
- Authority
- CN
- China
- Prior art keywords
- unmanned aerial
- aerial vehicle
- model
- yolov
- detection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 102
- 230000006872 improvement Effects 0.000 title claims abstract description 48
- 238000012549 training Methods 0.000 claims abstract description 62
- 238000012360 testing method Methods 0.000 claims abstract description 42
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 23
- 238000000034 method Methods 0.000 claims abstract description 21
- 230000000694 effects Effects 0.000 claims abstract description 8
- 238000007781 pre-processing Methods 0.000 claims abstract description 5
- 230000007246 mechanism Effects 0.000 claims description 30
- 230000006870 function Effects 0.000 claims description 25
- 230000008859 change Effects 0.000 claims description 11
- 230000008569 process Effects 0.000 claims description 11
- 206010000117 Abnormal behaviour Diseases 0.000 claims description 9
- 230000001133 acceleration Effects 0.000 claims description 9
- 238000004590 computer program Methods 0.000 claims description 8
- 238000010276 construction Methods 0.000 claims description 7
- 238000005516 engineering process Methods 0.000 claims description 7
- 238000012544 monitoring process Methods 0.000 claims description 7
- 238000005070 sampling Methods 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 230000002776 aggregation Effects 0.000 claims description 6
- 238000004220 aggregation Methods 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 6
- 238000002360 preparation method Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 5
- 238000003860 storage Methods 0.000 claims description 5
- 238000005520 cutting process Methods 0.000 claims description 4
- 230000004927 fusion Effects 0.000 claims description 4
- 238000012805 post-processing Methods 0.000 claims description 4
- 230000001629 suppression Effects 0.000 claims description 4
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 230000015572 biosynthetic process Effects 0.000 claims description 3
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000002372 labelling Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000003786 synthesis reaction Methods 0.000 claims description 3
- 230000003014 reinforcing effect Effects 0.000 claims 1
- 238000011176 pooling Methods 0.000 description 8
- 238000013527 convolutional neural network Methods 0.000 description 6
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 4
- 238000000605 extraction Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000005316 response function Methods 0.000 description 3
- 230000003213 activating effect Effects 0.000 description 2
- 238000003491 array Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000011068 loading method Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000011049 filling Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 230000036961 partial effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Image Analysis (AREA)
Abstract
The invention belongs to the technical field of computer vision, and discloses a method for acquiring an aerial unmanned aerial vehicle image, wherein an unmanned aerial vehicle data set is obtained after preprocessing; constructing an improved YOLOv unmanned aerial vehicle detection network model; dividing the unmanned aerial vehicle data set into an independent and non-repeated training set and a test set, inputting the training set into the unmanned aerial vehicle detection network model of the improvement YOLOv for training, and obtaining a trained unmanned aerial vehicle detection model; and detecting the unmanned aerial vehicle by using a model with the best training effect. The invention can detect the obstacle of the unmanned aerial vehicle in front under various complex backgrounds, has average detection precision to the unmanned aerial vehicle superior to that of the existing algorithm, can effectively improve the detection accuracy of the unmanned aerial vehicle to the obstacle, and reduces the collision risk of the unmanned aerial vehicle.
Description
Technical Field
The invention belongs to the technical field of computer vision, and particularly relates to an unmanned aerial vehicle detection method based on improvement YOLOv.
Background
Along with rapid development of science and technology, unmanned aerial vehicles are widely applied to various industries due to the characteristics of wide visual field, high flexibility and the like. In the unmanned aerial vehicle flight process, because unmanned aerial vehicle flight speed is fast, and the image of shooing has various complicated backgrounds, consequently the target is difficult to discern for traditional detection algorithm often receives the influence of environment under complicated scene, is difficult to extract the characteristic of target, thereby seriously appears misdetection and omission phenomenon. However, with the great breakthrough of deep learning in the field of target detection, the target is identified through the neural network, and compared with the traditional target detection method, the detection precision and speed are greatly improved, so that the vigorous development of unmanned aerial vehicle on target detection is promoted. The target detection based on deep learning is mainly divided into two types, one type is an algorithm based on an R-CNN system (R-CNN, fastR-CNN, faster R-CNN, and the like), and the two types are two-stage, and target candidate frames, namely target positions, need to be generated first, and then classification and regression are carried out on the candidate frames. The other type is a one-stage algorithm such as YOLO and SSD, which only uses one convolutional neural network CNN to directly predict the types and positions of different targets, and the YOLO (You OnlyLook Once) algorithm is an object recognition and positioning algorithm based on a deep neural network, and has the biggest characteristics of high running speed and can be used for a real-time system.
Disclosure of Invention
In view of the above existing problems, the present invention provides an unmanned aerial vehicle detection method based on improvement YOLOv, which can more effectively detect an unmanned aerial vehicle flying in the air.
In order to solve the above technical problems, an unmanned aerial vehicle detection method based on improvement YOLOv is provided, which comprises the following steps:
Preparing an unmanned aerial vehicle image data set, and splitting the unmanned aerial vehicle image data set into an independent and non-repeated training set and a test set; constructing a network model of the improvement YOLOv and constructing a three-dimensional reconstruction model; training the improved YOLOv network model constructed in the step 2 by using an unmanned aerial vehicle training set obtained in an unmanned aerial vehicle image dataset; and detecting and identifying the unmanned aerial vehicle test set obtained in the unmanned aerial vehicle image data set by using the trained improved YOLOv network model, and identifying and tracking and predicting abnormal behaviors of the unmanned aerial vehicle.
As a preferred embodiment of the unmanned aerial vehicle detection method based on the improvement YOLOv according to the present invention, the following is adopted: the unmanned aerial vehicle image dataset comprises unmanned aerial vehicle pictures obtained through network collection and manual synthesis, PASCALVOC dataset formats used for YOLOv are required to be processed, the dataset is established, and the unmanned aerial vehicle image dataset is divided into a training set and a testing set according to the proportion of 9:1.
As a preferred embodiment of the unmanned aerial vehicle detection method based on the improvement YOLOv according to the present invention, the following is adopted: the network model for constructing the improvement YOLOv comprises a Darknet network for extracting features of image information, a Darknet total of 53 layers of convolution layers are taken as a main network, the improvement part is in three output parts of YOLO, and CBAM attention mechanisms comprise a spatial attention mechanism M C and a channel attention mechanism M s:
Wherein F is an input feature map, sigma is a Sigmoid activation function, W 0、W1 is two weight parameters of MLP, and F 7×7 is a convolution kernel of 7×7 size.
As a preferred embodiment of the unmanned aerial vehicle detection method based on the improvement YOLOv according to the present invention, the following is adopted: the improvement YOLOv network model includes the improvement YOLOv having a base network of Darknet-CBAM: the first layer is a3×3 convolution layer, residual learning is performed through 5 residual modules respectively comprising 1,2, 8 and 4, darknet effective feature layers with three different scales are output, the effective feature layers are transmitted to an FPN network adopting a path aggregation strategy to perform feature fusion, after the feature layers output by the res4 module are subjected to convolution processing for 5 times, the features are reinforced by a CBAM attention mechanism, so that YOLO output y1 is obtained, and prediction with the scale of 52×52 is performed.
After convolution and up-sampling, the information is spliced and fused with tensor output by the second res8 module, after the same convolution treatment is carried out for 5 times, the characteristic is reinforced by using CBAM attention mechanisms, so that YOLO output y2 is obtained, 26 x 26 prediction is carried out, the fused information is fused with output of the first res8 module after convolution and up-sampling, so that YOLO output y3 is obtained, and prediction with the scale of 13 x 13 is carried out.
The path aggregation strategy includes Darknet outputting three different scale effective feature layers: 13×13×1024, 26×26×512, 52×52×256, and performing convolution processing on the feature layer for 5 times to obtain detection results, and performing upsampling on a part to combine with 26×26×512 features, thereby performing interpolation to obtain three detection results of the model.
As a preferred embodiment of the unmanned aerial vehicle detection method based on the improvement YOLOv according to the present invention, the following is adopted: the detection and identification comprises the steps of comparing anchor frames marked by test sets based on a basic network Darknet-CBAM in a training process, detecting the training process of a model, generating a plurality of candidate frames with different lengths and widths simultaneously, adopting a non-maximum suppression algorithm to sort the confidence coefficient of each candidate frame in a descending order, selecting the candidate frame with the highest confidence coefficient, learning the characteristic content through a convolution layer, outputting a model capable of identifying the unmanned aerial vehicle, verifying the test sets by using the output model to obtain a final detection result, detecting unmanned aerial vehicle pictures under different backgrounds, testing the detection effect, and obtaining average detection precision:
where TP is the number of correctly divided positive examples, FP is the number of incorrectly divided positive examples, and C represents the number of categories in the dataset.
As a preferred embodiment of the unmanned aerial vehicle detection method based on the improvement YOLOv according to the present invention, the following is adopted: the abnormal behavior recognition comprises the steps of collecting normal flight data of the unmanned aerial vehicle, including flight tracks, speeds, accelerations and gestures, carrying out normalization and noise removal pretreatment on the data, detecting and recognizing unmanned aerial vehicle images by using an improved YOLOv algorithm, inputting a large number of unmanned aerial vehicle images and corresponding labeling information in a training stage, enabling a model to accurately recognize unmanned aerial vehicles and states by continuously optimizing model parameters, inputting current flight data of the unmanned aerial vehicle into the trained model in a real-time monitoring stage, outputting the current state of the unmanned aerial vehicle by the model, and comparing output results with the normal flight data, wherein the abnormal behavior recognition comprises the following steps:
When the speed change rate of the unmanned aerial vehicle in the vertical direction exceeds vm/s 2, the unmanned aerial vehicle is judged to be suddenly accelerated, and when the speed change rate of the unmanned aerial vehicle in the vertical direction is lower than-vm/s 2, the unmanned aerial vehicle is judged to be suddenly decelerated.
The flight path is analyzed, when the turning radius of the unmanned aerial vehicle in the horizontal plane is not less than d meters and the turning angle of the unmanned aerial vehicle in the horizontal plane is not greater than a degrees, the unmanned aerial vehicle is in a normal turning, and when the turning radius of the unmanned aerial vehicle in the horizontal plane is less than d meters or the turning angle of the unmanned aerial vehicle in the horizontal plane is greater than a degrees, the unmanned aerial vehicle is in an abnormal turning.
And analyzing the smoothness degree of the flight track, when the fluctuation amplitude of the flight track of the unmanned aerial vehicle in the horizontal plane is not more than r meters, the unmanned aerial vehicle is in a normal flight state, and when the fluctuation amplitude of the flight track of the unmanned aerial vehicle in the horizontal plane is more than r meters, judging that the unmanned aerial vehicle has severe fluctuation and deviates from the normal track.
As a preferred embodiment of the unmanned aerial vehicle detection method based on the improvement YOLOv according to the present invention, the following is adopted: the tracking prediction comprises the steps of detecting and identifying an unmanned aerial vehicle test set obtained in an unmanned aerial vehicle image data set by using a trained improved YOLOv network model, collecting unmanned aerial vehicle motion data comprising speed, acceleration and angular speed by using the identified image data, preprocessing the image data comprising scaling, cutting and graying, wherein an original image is I, the scaling is theta, the cutting range is (x 0,y0,x1,y1), the graying image is I' =theta x 0:x1,y0:y1, and predicting a future flight path of the unmanned aerial vehicle by using a prediction model: p k+1＝N(xk,hk).
Wherein x k contains the position, speed and acceleration information of the unmanned plane at the current moment, h k is,For a prediction model, receiving a current flight state x k and historical flight data h k as inputs, outputting a flight path p k+1 in a future period of time, training the prediction model through a large amount of historical flight data, adjusting model parameters, minimizing errors between the prediction path and an actual path, and adjusting a weight minimization loss function of a neural network by using a gradient descent method:
Where p k+1,i is the model predicted future path, y k+1,i is the actual future path, N is the number of samples in the dataset, ω is the learning rate, Is the gradient of the loss function with respect to the weight w.
In the flight process of the unmanned aerial vehicle, the flight state of the unmanned aerial vehicle is monitored in real time, and early warning is sent out in advance according to a prediction result: when (p k+1-xk) is more than mu, the system can send out an early warning signal, automatically adjust the flight path of the unmanned aerial vehicle, automatically slow down or stop the flight if the unmanned aerial vehicle is close to an object and exceeds the speed limit, automatically raise the height of the unmanned aerial vehicle if the unmanned aerial vehicle is detected to be close to other flying objects during low-altitude flight, inform an operator to perform manual intervention, control the unmanned aerial vehicle to keep away from danger, and when (p k+1-xk) is less than mu, the change of the current state and the predicted state does not exceed a threshold value, and the monitoring system still needs to continuously monitor the flight state of the unmanned aerial vehicle to ensure the flight safety.
Another object of the present invention is to provide an unmanned aerial vehicle detection system based on the improvement YOLOv, which can realize accurate detection and identification of unmanned aerial vehicle images through the improved YOLOv3 algorithm, and can still maintain high precision, especially under complex environments or when the unmanned aerial vehicle has large form change; the running speed of the detection algorithm is ensured, so that the detection algorithm can process image data in real time, and the actual application requirements of unmanned aerial vehicle monitoring and the like are met; improving measures such as CBAM attention mechanisms are introduced, so that the robustness of the algorithm to adverse factors such as illumination change, shielding and background interference is improved; by improving the network structure and training strategy, the performance of the model in unmanned aerial vehicle detection tasks is optimized, including accuracy, recall rate, model complexity and the like.
The unmanned aerial vehicle detection system based on the improvement YOLOv is a preferable scheme of the unmanned aerial vehicle detection system based on the improvement YOLOv, and is characterized by comprising a data set preparation module, a improvement YOLOv network construction module, a training module, a detection and identification module and a performance evaluation module.
The dataset preparation module provides high-quality and diversified unmanned aerial vehicle image data by collecting network resources and artificial synthetic images, and is used for training and testing an improved YOLOv model, so that the richness and the representativeness of datasets are ensured.
The improved YOLOv network construction module constructs a network model which integrates the latest technology and the high-efficiency structure through integrating the Darknet network and the CBAM attention mechanism, is used for unmanned aerial vehicle detection, and improves the learning and recognition capability of the model on unmanned aerial vehicle characteristics.
The training module enables the model to learn the knowledge and skill required by unmanned aerial vehicle detection through training, and optimizes the model by utilizing a training set until a satisfactory detection effect is achieved.
The detection and recognition module is used for carrying out actual detection and recognition on the unmanned aerial vehicle test set by using a trained model, and the model output and post-processing technology is adopted: and (5) suppressing the confidence threshold and the non-maximum value, screening and outputting a final detection result.
The performance evaluation module verifies the performance of the model through the test set, evaluates and feeds back the performance of the whole detection system, ensures that the detection precision meets the actual application requirement, and provides an improved direction.
A computer device comprising a memory and a processor, said memory storing a computer program, characterized in that said processor, when executing said computer program, implements the steps of the method for drone detection based on improvement YOLOv.
A computer readable storage medium having stored thereon a computer program, characterized in that the computer program when executed by a processor implements the steps of the method for drone detection based on improvement YOLOv.
The invention has the beneficial effects that: the invention can detect the obstacle of the unmanned aerial vehicle in front under various complex backgrounds, has average detection precision to the unmanned aerial vehicle superior to that of the existing algorithm, can effectively improve the detection accuracy of the unmanned aerial vehicle to the obstacle, and reduces the collision risk of the unmanned aerial vehicle.
Drawings
For a clearer description of the technical solutions of embodiments of the present invention, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art, wherein:
fig. 1 is a general flow chart of an unmanned aerial vehicle detection method based on the improvement YOLOv according to an embodiment of the present invention.
Fig. 2 is a general frame diagram of a YOLOv modified network based on the unmanned aerial vehicle detection method of the modification YOLOv according to one embodiment of the present invention.
Fig. 3 is a CBAM module of an unmanned aerial vehicle detection method based on the improvement YOLOv according to an embodiment of the present invention.
Fig. 4 is a schematic diagram of an effect of unmanned aerial vehicle detection based on the unmanned aerial vehicle detection method of the improvement YOLOv according to an embodiment of the present invention.
Fig. 5 is a system function architecture diagram of an unmanned aerial vehicle detection system based on the improvement YOLOv according to an embodiment of the present invention.
Detailed Description
So that the manner in which the above recited objects, features and advantages of the present invention can be understood in detail, a more particular description of the invention, briefly summarized above, may be had by reference to the embodiments, some of which are illustrated in the appended drawings. All other embodiments, which can be made by one of ordinary skill in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways other than those described herein, and persons skilled in the art will readily appreciate that the present invention is not limited to the specific embodiments disclosed below.
Further, reference herein to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic can be included in at least one implementation of the invention. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments.
While the embodiments of the present invention have been illustrated and described in detail in the drawings, the cross-sectional view of the device structure is not to scale in the general sense for ease of illustration, and the drawings are merely exemplary and should not be construed as limiting the scope of the invention. In addition, the three-dimensional dimensions of length, width and depth should be included in actual fabrication.
Also in the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "upper, lower, inner and outer", etc. are based on the orientation or positional relationship shown in the drawings, are merely for convenience of describing the present invention and simplifying the description, and do not indicate or imply that the apparatus or elements referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as limiting the present invention. Furthermore, the terms "first, second, or third" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
The terms "mounted, connected, and coupled" should be construed broadly in this disclosure unless otherwise specifically indicated and defined, such as: can be fixed connection, detachable connection or integral connection; it may also be a mechanical connection, an electrical connection, or a direct connection, or may be indirectly connected through an intermediate medium, or may be a communication between two elements. The specific meaning of the above terms in the present invention will be understood in specific cases by those of ordinary skill in the art.
Example 1
Referring to fig. 1-3, a first embodiment of the present invention provides an unmanned aerial vehicle detection method based on improvement YOLOv, which includes:
s1: and preparing an unmanned aerial vehicle image data set, and splitting the unmanned aerial vehicle image data set into an independent and non-repeated training set and a test set.
Furthermore, the public data set of the unmanned aerial vehicle is fewer and the data is imperfect, and the data set needs to be reestablished and marked. The unmanned aerial vehicle data set picture mainly comes from network collection and synthesis. In order to improve the application capability of the unmanned aerial vehicle image processing method in the actual scene, collected data set pictures are all based on different backgrounds, and then the unmanned aerial vehicle image information is obtained by marking through a target detection marking tool. It is converted to VOC format by encoding.
It should be noted that the collected unmanned aerial vehicle image data is divided into independent and non-repeated training sets and test sets according to the proportion of 9:1 by adopting a random sampling mode.
S2: a network model of the improvement YOLOv is constructed and a three-dimensional reconstruction model is constructed.
Further, darknet networks for extracting features of image information, together 53 convolutions. FPN (Feature Pyramid Networks) is added with an attention mechanism layer CBAMLayer, and as the attention mechanism contains a learnable weight, the attention degree of the main channel characteristics can be enhanced, and the detection precision of a smaller target can be improved. The improvement to YOLOv is mainly comprised of the steps of:
Further, the backbone feature extraction network Darknet53 of YOLOv performs feature extraction on the input image in Darknet, and three feature layers are obtained for constructing the next network.
Further, YOLOv is a reinforced feature extraction network FPN, feature fusion is performed on three feature layers obtained from the trunk portion, and a attention mechanism CBAM is added into the FPN to enhance the attention degree to the main channel features.
Further, attention mechanism CBAM is a mechanism module that combines space and channels. Given an intermediate feature map, our module gets attention maps sequentially along two independent dimensions of channel and space, and then multiplies the attention maps by the input feature map for adaptive feature refinement.
The channel attention mechanism uses global average pooling and global maximum pooling to input feature graphs, and respectively inputs the feature graphs into the MLP, and outputs the result after the result is subjected to element-wiseadd and an activation function, as follows:
and a spatial attention mechanism, which applies global average pooling and global maximum pooling to the feature map along the channel direction, and outputs the feature map after activating the function, as follows:
Wherein F is an input feature map, sigma is a Sigmoid activation function, W 0、W1 is two weight parameters of MLP, and F 7×7 is a convolution kernel of 7×7 size.
It should be noted that, collecting multiple images of the unmanned aerial vehicle at different angles and positions and preprocessing the collected images, including denoising, contrast enhancement, color correction and the like, so as to improve the image quality; the image is geometrically corrected to eliminate lens distortion and image distortion.
Extracting feature points from the image by using a deep learning algorithm, wherein the feature points comprise edges, corner points and textures, and the feature points are transformed by scale-invariant features:
L(x,y,α)＝I(x,y)*G(x,y,α)
where G (x, y, α) is a gaussian function, α is a scale parameter, and x is a convolution operation.
Then, detecting key points, wherein the response function D (x, y, alpha) of the key points is as follows:
This response function is responsive to scale changes, with local maxima corresponding to potential keypoints, and pinpointing detected keypoints, including determining their position in the image and the direction of the keypoints: the location of the keypoints is obtained by solving the local maxima of the response function D (x, y, σ), while the direction of the keypoints is estimated by calculating the direction of the local gradients of the image.
Generating a descriptor for each key point, wherein the descriptor is a histogram of local gradient directions of an image area around the key point, the generation of the descriptor involves gradient calculation of the image area around the key point, and statistics of frequency of each gradient direction; matching of key points is achieved by comparing descriptors of key points in different images, matching is achieved by calculating the distance between the two descriptors, and the smaller the distance is, the higher the matching degree is.
Finding the corresponding characteristic points in the multiple images and then calibrating the cameras is a key step of three-dimensional reconstruction, because the accurate correspondence between the pixel coordinates in the images and the actual world coordinates is ensured, and the three-dimensional coordinates of the unmanned plane surface points are calculated by using the multi-view geometric principle and combining the camera calibration parameters and the characteristic point matching.
The three-dimensional reconstruction algorithm comprises the steps of structure motion, simultaneous positioning and map construction, the extracted characteristic points are optimized by using a deep learning model, the three-dimensional reconstruction precision is improved, and post-processing of the reconstructed three-dimensional model comprises surface smoothing, hole filling, texture mapping and the like so as to improve the appearance and practicability of the model.
S3: and training the improved YOLOv network model constructed in the step2 by using the unmanned aerial vehicle training set obtained in the unmanned aerial vehicle image dataset.
Still further, the underlying network of improvement YOLOv is Darknet, 53-CBAM: the first layer is a3×3 convolution layer, residual learning is performed through 5 residual modules respectively comprising 1,2, 8 and 4, darknet effective feature layers with three different scales are output, the effective feature layers are transmitted to an FPN network adopting a path aggregation strategy to perform feature fusion, after the feature layers output by the res4 module are subjected to convolution processing for 5 times, the features are reinforced by a CBAM attention mechanism, so that YOLO output y1 is obtained, and prediction with the scale of 52×52 is performed.
After convolution and up-sampling, the information is spliced and fused with tensor output by the second res8 module, after the same convolution treatment is carried out for 5 times, the characteristic is reinforced by using CBAM attention mechanisms, so that YOLO output y2 is obtained, 26 x 26 prediction is carried out, the fused information is fused with output of the first res8 module after convolution and up-sampling, so that YOLO output y3 is obtained, and prediction with the scale of 13 x 13 is carried out.
The path aggregation strategy includes Darknet outputting three different scale effective feature layers: 13×13×1024, 26×26×512, 52×52×256, and performing convolution processing on the feature layer for 5 times to obtain detection results, and performing upsampling on a part to combine with 26×26×512 features, thereby performing interpolation to obtain three detection results of the model.
Further, training the improved model, wherein the picture input size of the model is 416×416, the initial learning rate is set to be 0.001, the processed training data set is input into the model according to the set batch_size (set according to hardware conditions), forward propagation is performed, loss is calculated, and then the parameters in the network are updated according to the loss function in a backward propagation mode. After multiple iterations, when the network loss tends to be stable, model training is stopped, and parameters of the network model are stored. The loss function formula is as follows:
Wherein the method comprises the steps of Is the center coordinate of a rectangular frame of network prediction,/>Is the center coordinates of the marked rectangular box,/>Indicating whether the rectangular box is responsible for predicting a target object. Lambda coord is a coordination coefficient set for coordinating the inconsistent contribution of rectangular frames with different sizes to the error function,/>The width and height of the rectangular frame predicted for the network,Is to mark the width and height of a rectangular frame,/>To predict the probability score of a target object contained within a frame,/>Representing the true value, lambda noobj is a weight value,/>Representing the probability that the (i, j) th prediction box belongs to category C,/>And representing the true value of the category to which the mark frame belongs.
It should be further noted that, the unmanned aerial vehicle data set obtained in the first step is input into the built improved network in batches, 4 values of boundary positions of each prediction frame, confidence degrees of the frame positions and confidence degrees of unmanned aerial vehicle categories are output by the network once each iteration, a loss function is built according to the values predicted by the network and real tag values of the training set, loss is calculated, parameters of the improved network are updated by using an Adam optimization algorithm in a back propagation mode until the loss is not reduced, and at the moment, network parameters are saved as a model.
Adam updates rules and calculates the gradient of the t-th iteration:
S4: and detecting and identifying the unmanned aerial vehicle test set obtained in the unmanned aerial vehicle image data set by using the trained improved YOLOv network model, and identifying and tracking and predicting abnormal behaviors of the unmanned aerial vehicle.
Furthermore, based on the basic network Darknet-CBAM, the anchor frames marked by the test set are adopted for comparison in the training process, the training process of the model is detected, meanwhile, a plurality of candidate frames with different lengths and widths are generated, a non-maximum suppression algorithm is adopted for descending order of confidence coefficient of each candidate frame, the candidate frame with the highest confidence coefficient is selected from the candidate frames, the model capable of identifying the unmanned aerial vehicle is output through the learning characteristic content of the convolution layer, the test set is verified by the output model, the final detection result is obtained, the unmanned aerial vehicle images under different backgrounds are detected, the detection effect is tested, the unmanned aerial vehicle with the detection model under various backgrounds can be effectively detected, and collision between unmanned aerial vehicles can be effectively avoided when the unmanned aerial vehicle is applied to the unmanned aerial vehicle.
The trained model is used for detecting the test data set to obtain average detection precision, and the result shows that the detection precision of the improved model on the unmanned aerial vehicle is improved:
where TP is the number of correctly divided positive examples, FP is the number of incorrectly divided positive examples, and C represents the number of categories in the dataset.
It should be further noted that, the abnormal behavior recognition includes collecting normal flight data of the unmanned aerial vehicle including flight track, speed, acceleration and gesture, carrying out normalization and noise removal pretreatment on the data, detecting and recognizing the unmanned aerial vehicle image by using an improved YOLOv algorithm, inputting a large amount of unmanned aerial vehicle images and corresponding labeling information in a training stage, enabling the model to accurately recognize the unmanned aerial vehicle and the state by continuously optimizing model parameters, inputting the current flight data of the unmanned aerial vehicle into the trained model in a real-time monitoring stage, outputting the current state of the unmanned aerial vehicle by the model, comparing the output result with the normal flight data, and recognizing the abnormal behavior includes: when the speed change rate of the unmanned aerial vehicle in the vertical direction exceeds 5m/s 2, the unmanned aerial vehicle is judged to be suddenly accelerated, and when the speed change rate of the unmanned aerial vehicle in the vertical direction is lower than-5 m/s 2, the unmanned aerial vehicle is judged to be suddenly decelerated.
The flight path is analyzed, when the turning radius of the unmanned aerial vehicle in the horizontal plane is not less than 10 meters and the turning angle of the unmanned aerial vehicle in the horizontal plane is not more than 45 degrees, the unmanned aerial vehicle is in a normal turning, and when the turning radius of the unmanned aerial vehicle in the horizontal plane is less than 10 meters or the turning angle of the unmanned aerial vehicle in the horizontal plane is more than 45 degrees, the unmanned aerial vehicle is in an abnormal turning.
And analyzing the smoothness degree of the flight track, when the fluctuation amplitude of the flight track of the unmanned aerial vehicle in the horizontal plane is not more than 2 meters, the unmanned aerial vehicle is in a normal flight state, and when the fluctuation amplitude of the flight track of the unmanned aerial vehicle in the horizontal plane is more than 2 meters, judging that the unmanned aerial vehicle has severe fluctuation and deviates from the normal track.
It should be noted that, tracking prediction includes detecting and identifying a test set of the unmanned aerial vehicle obtained in an image dataset of the unmanned aerial vehicle by using a trained improved YOLOv network model, collecting motion data of the unmanned aerial vehicle including speed, acceleration and angular velocity by using the identified image data, preprocessing the image data including scaling, clipping and graying, the original image is I, the scaling is θ, the clipping range is (x 0,y0,x1,y1), the graying image is I' =θ x 0:x1,y0:y1, and predicting a future flight path of the unmanned aerial vehicle by using a prediction model: p k+1＝N(xk,hk).
Wherein x k contains the position, speed and acceleration information of the unmanned plane at the current moment, h k is,For a prediction model, receiving a current flight state x k and historical flight data h k as inputs, outputting a flight path p k+1 in a future period of time, training the prediction model through a large amount of historical flight data, adjusting model parameters, minimizing errors between the prediction path and an actual path, and adjusting a weight minimization loss function of a neural network by using a gradient descent method:
Where p k+1,i is the model predicted future path, y k+1,i is the actual future path, N is the number of samples in the dataset, ω is the learning rate, Is the gradient of the loss function with respect to the weight w.
In the flight process of the unmanned aerial vehicle, the flight state of the unmanned aerial vehicle is monitored in real time, and early warning is sent out in advance according to a prediction result: when (p k+1-xk) is more than mu, the system can send out an early warning signal, automatically adjust the flight path of the unmanned aerial vehicle, automatically slow down or stop the flight if the unmanned aerial vehicle is close to an object and exceeds the speed limit, automatically raise the height of the unmanned aerial vehicle if the unmanned aerial vehicle is detected to be close to other flying objects during low-altitude flight, inform an operator to perform manual intervention, control the unmanned aerial vehicle to keep away from danger, and when (p k+1-xk) is less than mu, the change of the current state and the predicted state does not exceed a threshold value, and the monitoring system still needs to continuously monitor the flight state of the unmanned aerial vehicle to ensure the flight safety.
Example 2
Referring to fig. 4, for one embodiment of the present invention, an unmanned aerial vehicle detection method based on the improvement YOLOv is provided, and in order to verify the beneficial effects of the present invention, scientific demonstration is performed through experiments.
First, a data set of the unmanned aerial vehicle in the VOC format is established.
(1) Image and tag data required for the object detection network are prepared.
The unmanned aerial vehicle image is obtained through network collection, the unmanned aerial vehicle image is labeled by LabelImg, a unmanned aerial vehicle data set is obtained, the unmanned aerial vehicle data set is divided into a training set and a testing set through a 9:1 ratio, wherein the training set comprises 450 images, the testing set comprises 50 images, the resolution ratio of the training set and the testing set is 416 multiplied by 416, and only one type of unmanned aerial vehicle exists in the images.
(2) The tag values of the data sets are all in xml format, and in order to be able to use the data sets in the network used in this patent, it is necessary to convert the tag data of the data sets into VOC format.
(3) A drone dataset folder VOCdevkit is created that stores the VOC format, comprising three subfiles under the folder, annotation, imageSets and JPEGImages respectively. The prepared training pictures are placed in JPEGImages folders and stored in a naming order starting at 0001.Jpg according to the VOC format. And marking the placed picture by using LabelImg tools, generating an xml file with the same name as the picture according to the category and the position information of the target in the picture, and placing the xml file in an animation folder. Creating subfiles in IMAGESETS folders, named Main, proportionally generating training sample sets and test sample sets for the existing unmanned aerial vehicle picture data, wherein the training sample sets are named train. Txt, the test sample sets are named test. Txt, absolute paths of pictures in JPEGImages are stored in the test sample sets, and the two. Txt files are placed in the Main folders.
Second, building a network and training
(1) Building an improved network
YOLOv3 an improved network with an attention mechanism module is built, and the structure of the improved network is shown in a figure II. The DBL module in FIG. two consists of a 3×3 convolution module, BN and LeakyReLU, wherein the resunit module consists of two DBL modules, and wherein add represents an element-wise operation on the two feature maps. CBAM attention mechanism is added to the YOLOv network at three output feature layers, and the structure is shown in figure three. Mainly consists of a spatial attention mechanism and a channel attention mechanism. Two attention mechanisms are combined, important feature extraction is focused, and useless features are ignored.
The channel attention mechanism uses global average pooling and global maximum pooling to input feature graphs, and respectively inputs the feature graphs into the MLP, and outputs the results after the results are subjected to element-wiseadd and an activation function, as follows:
and a spatial attention mechanism, which applies global average pooling and global maximum pooling to the feature map along the channel direction, and outputs the feature map after activating the function, as follows:
here σ is a Sigmoid activation function, the formula of which is as follows:
the function of Sigmoid maps a real number to the interval of (0, 1) for two classifications.
(2) Initializing trainable parameters of a network
Training is performed on the PASCALVOC dataset using YOLOv networks, and the trained network parameters are saved as a model, referred to as a pre-training model. And reading parameters of the same network layer in the pre-training model as those in the improved network, and loading pre-training weights into the built improved network.
(3) Begin training and save model
Inputting the unmanned aerial vehicle data set obtained in the first step into a built improved network in batches, setting batch_size to 8, setting training times to 100, outputting 4 values of boundary positions of each prediction frame, confidence degree of frame positions and confidence degree of unmanned aerial vehicle types by the network once per iteration, constructing a loss function according to the values predicted by the network and real tag values of the training set, calculating loss, and using an Adam optimization algorithm to perform back propagation to update parameters of the improved network until the loss is not reduced, and saving network parameters as a model at the moment.
Adam updates rules and calculates the gradient of the t-th iteration:
Fourth step, detecting and identifying unmanned aerial vehicle
And loading a final model, and inputting all images of the test set into the model for testing. When the NMS threshold is set to 0.3, the IoU threshold is set to 0.5, and the category confidence threshold is set to 0.5, the average detection precision mAP of the network on a test set is 96.92%. The partial effect of the test is shown in figure four. In an actual scene, the unmanned aerial vehicle camera can be called to identify and detect the unmanned aerial vehicle flying in front, so that a corresponding obstacle avoidance strategy is made, and collision is prevented. The final detection result shows that the detection precision of the improved algorithm is 96.92% and is superior to that of the existing algorithm.
It should be noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made thereto without departing from the spirit and scope of the technical solution of the present invention, which is intended to be covered by the scope of the claims of the present invention.
Example 3
A third embodiment of the present invention, which is different from the first two embodiments, is:
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a usb disk, a removable hard disk, a Read-only memory (ROM), a random access memory (RAM, randomAccessMemory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
Logic and/or steps represented in the flowcharts or otherwise described herein, e.g., a ordered listing of executable instructions for implementing logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic device) having one or more wires, a portable computer diskette (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer readable medium may even be paper or other suitable medium on which the program is printed, as the program may be electronically captured, via, for instance, optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory.
It is to be understood that portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above-described embodiments, the various steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, may be implemented using any one or combination of the following techniques, as is well known in the art: discrete logic circuits having logic gates for implementing logic functions on data signals, application specific integrated circuits having suitable combinational logic gates, programmable Gate Arrays (PGAs), field Programmable Gate Arrays (FPGAs), and the like.
Example 4
Referring to fig. 5, a fourth embodiment of the present invention provides an unmanned aerial vehicle detection system based on the improvement YOLOv, which includes a data set preparation module, a improvement YOLOv3 network construction module, a training module, a detection and identification module, and a performance evaluation module.
The dataset preparation module provides high-quality and diversified unmanned aerial vehicle image data by collecting network resources and artificial synthetic images, and is used for training and testing an improved YOLOv model, so that the richness and representativeness of the dataset are ensured.
The improved YOLOv network construction module constructs a network model which integrates the latest technology and the high-efficiency structure through integrating the Darknet network and the CBAM attention mechanism, is used for unmanned aerial vehicle detection, and improves the learning and recognition capability of the model on unmanned aerial vehicle characteristics.
The training module learns the knowledge and skill required by unmanned aerial vehicle detection through training, optimizes the model by utilizing a training set until a satisfactory detection effect is achieved.
The detection and recognition module is used for carrying out actual detection and recognition on the unmanned aerial vehicle test set by using a trained model, and the model output and post-processing technology is adopted: and (5) suppressing the confidence threshold and the non-maximum value, screening and outputting a final detection result.
The performance evaluation module verifies the performance of the model through the test set, evaluates and feeds back the performance of the whole detection system, ensures that the detection precision meets the actual application requirement, and provides an improvement direction.
It should be noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made thereto without departing from the spirit and scope of the technical solution of the present invention, which is intended to be covered by the scope of the claims of the present invention.
Claims (10)
1. An unmanned aerial vehicle detection method based on improvement YOLOv is characterized in that: comprising the steps of (a) a step of,
Preparing an unmanned aerial vehicle image data set, and splitting the unmanned aerial vehicle image data set into an independent and non-repeated training set and a test set;
constructing a network model of the improvement YOLOv and constructing a three-dimensional reconstruction model;
Training the improved YOLOv network model constructed in the step 2 by using an unmanned aerial vehicle training set obtained in an unmanned aerial vehicle image dataset;
And detecting and identifying the unmanned aerial vehicle test set obtained in the unmanned aerial vehicle image data set by using the trained improved YOLOv network model, and identifying and tracking and predicting abnormal behaviors of the unmanned aerial vehicle.
2. The unmanned aerial vehicle detection method based on the improvement YOLOv as set forth in claim 1, wherein: the unmanned aerial vehicle image dataset comprises unmanned aerial vehicle pictures obtained through network collection and manual synthesis, PASCALVOC dataset formats used for YOLOv are required to be processed, the dataset is established, and the unmanned aerial vehicle image dataset is divided into a training set and a testing set according to the proportion of 9:1.
3. The unmanned aerial vehicle detection method based on the improvement YOLOv as set forth in claim 2, wherein: the network model for constructing the improvement YOLOv comprises a Darknet network for extracting features of image information, a Darknet total of 53 layers of convolution layers are taken as a main network, the improvement part is in three output parts of YOLO, and CBAM attention mechanisms comprise a spatial attention mechanism M C and a channel attention mechanism M s:
Wherein F is an input feature map, sigma is a Sigmoid activation function, W 0、W1 is two weight parameters of MLP, and F 7×7 is a convolution kernel of 7×7 size.
4. A method of unmanned aerial vehicle detection based on improvement YOLOv as claimed in claim 3, wherein: the improvement YOLOv network model includes the improvement YOLOv having a base network of Darknet-CBAM: the first layer is a 3×3 convolution layer, residual learning is carried out through 5 residual modules respectively comprising 1, 2, 8 and 4, darknet effective feature layers with three different scales are output, the three effective feature layers are transmitted to an FPN network adopting a path aggregation strategy for feature fusion, after the feature layers output by the res4 module are subjected to convolution processing for 5 times, the features are reinforced by a CBAM attention mechanism, so that YOLO output y1 is obtained, and prediction with the scale of 52×52 is carried out;
after convolution and up-sampling, splicing and fusing the information with tensor output by a second res8 module, after the same convolution treatment for 5 times, reinforcing the characteristics by using CBAM attention mechanisms to obtain YOLO output y2, predicting 26×26, fusing the fused information with output of the first res8 module after convolution and up-sampling to obtain YOLO output y3, and predicting the scale of 13×13;
The path aggregation strategy includes Darknet outputting three different scale effective feature layers: 13×13×1024, 26×26×512, 52×52×256, and performing convolution processing on the feature layer for 5 times to obtain detection results, and performing upsampling on a part to combine with 26×26×512 features, thereby performing interpolation to obtain three detection results of the model.
5. The unmanned aerial vehicle detection method based on the improvement YOLOv as set forth in claim 4, wherein: the detection and identification comprises the steps of comparing anchor frames marked by test sets based on a basic network Darknet-CBAM in a training process, detecting the training process of a model, generating a plurality of candidate frames with different lengths and widths simultaneously, adopting a non-maximum suppression algorithm to sort the confidence coefficient of each candidate frame in a descending order, selecting the candidate frame with the highest confidence coefficient, learning the characteristic content through a convolution layer, outputting a model capable of identifying the unmanned aerial vehicle, verifying the test sets by using the output model to obtain a final detection result, detecting unmanned aerial vehicle pictures under different backgrounds, testing the detection effect, and obtaining average detection precision:
where TP is the number of correctly divided positive examples, FP is the number of incorrectly divided positive examples, and C represents the number of categories in the dataset.
6. The unmanned aerial vehicle detection method based on the improvement YOLOv as set forth in claim 5, wherein: the abnormal behavior recognition comprises the steps of collecting normal flight data of the unmanned aerial vehicle, including flight tracks, speeds, accelerations and gestures, carrying out normalization and noise removal pretreatment on the data, detecting and recognizing unmanned aerial vehicle images by using an improved YOLOv algorithm, inputting a large number of unmanned aerial vehicle images and corresponding labeling information in a training stage, enabling a model to accurately recognize unmanned aerial vehicles and states by continuously optimizing model parameters, inputting current flight data of the unmanned aerial vehicle into the trained model in a real-time monitoring stage, outputting the current state of the unmanned aerial vehicle by the model, and comparing output results with the normal flight data, wherein the abnormal behavior recognition comprises the following steps:
When the speed change rate of the unmanned aerial vehicle in the vertical direction exceeds v m/s 2, judging that the unmanned aerial vehicle is suddenly accelerated, and when the speed change rate of the unmanned aerial vehicle in the vertical direction is lower than-v m/s 2, judging that the unmanned aerial vehicle is suddenly decelerated;
Analyzing a flight track, wherein when the turning radius of the unmanned aerial vehicle in the horizontal plane is not smaller than d meters and the turning angle of the unmanned aerial vehicle in the horizontal plane is not larger than a degrees, the unmanned aerial vehicle is in a normal turning, and when the turning radius of the unmanned aerial vehicle in the horizontal plane is smaller than d meters or the turning angle of the unmanned aerial vehicle in the horizontal plane is larger than a degrees, the unmanned aerial vehicle is in an abnormal turning;
And analyzing the smoothness degree of the flight track, when the fluctuation amplitude of the flight track of the unmanned aerial vehicle in the horizontal plane is not more than r meters, the unmanned aerial vehicle is in a normal flight state, and when the fluctuation amplitude of the flight track of the unmanned aerial vehicle in the horizontal plane is more than r meters, judging that the unmanned aerial vehicle has severe fluctuation and deviates from the normal track.
7. The unmanned aerial vehicle detection method based on the improvement YOLOv as set forth in claim 6, wherein: the tracking prediction comprises the steps of detecting and identifying an unmanned aerial vehicle test set obtained in an unmanned aerial vehicle image data set by using a trained improved YOLOv network model, collecting unmanned aerial vehicle motion data comprising speed, acceleration and angular speed by using the identified image data, preprocessing the image data comprising scaling, cutting and graying, wherein an original image is I, the scaling is theta, the cutting range is (x 0,y0,x1,y1), the graying image is I' =theta x 0:x1,y0:y1, and predicting a future flight path of the unmanned aerial vehicle by using a prediction model: p k+1＝N(xk,hk);
Wherein x k contains the position, speed and acceleration information of the unmanned plane at the current moment, h k is, For a prediction model, receiving a current flight state x k and historical flight data h k as inputs, outputting a flight path p k+1 in a future period of time, training the prediction model through a large amount of historical flight data, adjusting model parameters, minimizing errors between the prediction path and an actual path, and adjusting a weight minimization loss function of a neural network by using a gradient descent method:
Where p k+1,i is the model predicted future path, y k+1,i is the actual future path, N is the number of samples in the dataset, ω is the learning rate, Is the gradient of the loss function with respect to the weight w;
In the flight process of the unmanned aerial vehicle, the flight state of the unmanned aerial vehicle is monitored in real time, and early warning is sent out in advance according to a prediction result: when (p k+1-xk) is more than mu, the system can send out an early warning signal, automatically adjust the flight path of the unmanned aerial vehicle, automatically slow down or stop the flight if the unmanned aerial vehicle is close to an object and exceeds the speed limit, automatically raise the height of the unmanned aerial vehicle if the unmanned aerial vehicle is detected to be close to other flying objects during low-altitude flight, inform an operator to perform manual intervention, control the unmanned aerial vehicle to keep away from danger, and when (p k+1-xk) is less than mu, the change of the current state and the predicted state does not exceed a threshold value, and the monitoring system still needs to continuously monitor the flight state of the unmanned aerial vehicle to ensure the flight safety.
8. A system employing a modified YOLOv-based drone detection method as claimed in any one of claims 1 to 7, wherein: the system comprises a data set preparation module, an improvement YOLOv network construction module, a training module, a detection and identification module and a performance evaluation module;
The data set preparation module is used for providing high-quality and diversified unmanned aerial vehicle image data by collecting network resources and artificial synthetic images, and is used for training and testing an improved YOLOv model so as to ensure the richness and the representativeness of the data set;
the improved YOLOv network construction module constructs a network model which integrates the latest technology and the high-efficiency structure through integrating the Darknet network and the CBAM attention mechanism, is used for unmanned aerial vehicle detection, and improves the learning and recognition capability of the model on unmanned aerial vehicle characteristics;
The training module enables the model to learn the knowledge and skill required by unmanned aerial vehicle detection through training, and optimizes the model by utilizing a training set until a satisfactory detection effect is achieved;
The detection and recognition module is used for carrying out actual detection and recognition on the unmanned aerial vehicle test set by using a trained model, and the model output and post-processing technology is adopted: confidence threshold and non-maximum suppression, screening and outputting a final detection result;
The performance evaluation module verifies the performance of the model through the test set, evaluates and feeds back the performance of the whole detection system, ensures that the detection precision meets the actual application requirement, and provides an improved direction.
9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any one of claims 1 to 8 when the computer program is executed.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410035199.5A CN118115896A (en) | 2024-01-10 | 2024-01-10 | Unmanned aerial vehicle detection method and system based on improvement YOLOv3 |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410035199.5A CN118115896A (en) | 2024-01-10 | 2024-01-10 | Unmanned aerial vehicle detection method and system based on improvement YOLOv3 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118115896A true CN118115896A (en) | 2024-05-31 |
Family
ID=91218804
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410035199.5A Pending CN118115896A (en) | 2024-01-10 | 2024-01-10 | Unmanned aerial vehicle detection method and system based on improvement YOLOv3 |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118115896A (en) |
-
2024
- 2024-01-10 CN CN202410035199.5A patent/CN118115896A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110472627B (en) | End-to-end SAR image recognition method, device and storage medium | |
| CN112464883B (en) | Automatic detection and identification method and system for ship target in natural scene | |
| CN113269073B (en) | Ship multi-target tracking method based on YOLO V5 algorithm | |
| US20130251246A1 (en) | Method and a device for training a pose classifier and an object classifier, a method and a device for object detection | |
| CN110084299B (en) | Target detection method and device based on multi-head fusion attention | |
| CN105654067A (en) | Vehicle detection method and device | |
| CN112836639A (en) | Pedestrian multi-target tracking video identification method based on improved YOLOv3 model | |
| CN111797769B (en) | Small-target-sensitive vehicle detection system | |
| CN111368634B (en) | Human head detection method, system and storage medium based on neural network | |
| CN113487610B (en) | Herpes image recognition method and device, computer equipment and storage medium | |
| CN108985375B (en) | Multi-feature fusion tracking method considering particle weight spatial distribution | |
| CN115049821A (en) | Three-dimensional environment target detection method based on multi-sensor fusion | |
| CN110659601A (en) | Depth full convolution network remote sensing image dense vehicle detection method based on central point | |
| CN117274768A (en) | Training method of target detection network, target detection method and related device | |
| Fan et al. | A novel sonar target detection and classification algorithm | |
| CN116310837A (en) | SAR ship target rotation detection method and system | |
| CN114283326A (en) | Underwater target re-identification method combining local perception and high-order feature reconstruction | |
| CN111428567B (en) | Pedestrian tracking system and method based on affine multitask regression | |
| CN114972492A (en) | Position and pose determination method and device based on aerial view and computer storage medium | |
| CN116994049A (en) | Full-automatic flat knitting machine and method thereof | |
| US20230245466A1 (en) | Vehicle Lidar System and Object Classification Method Therewith | |
| CN116343143A (en) | Target detection method, storage medium, road side equipment and automatic driving system | |
| CN115937520A (en) | Point cloud moving target segmentation method based on semantic information guidance | |
| CN116453109A (en) | 3D target detection method, device, equipment and storage medium | |
| CN111144422A (en) | Positioning identification method and system for aircraft component |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication |