Aircraft tractor automatic identification method based on YOLO

Abstract

The invention discloses an aircraft tractor automatic identification method based on YOLO. The core of the method is the network structure and algorithm of the YOLO model. The YOLO model can realize real-time target detection, divides an input image into a plurality of grids, and predicts a plurality of bounding boxes and corresponding category probabilities on each grid. By screening and optimizing the bounding boxes, a YOLO model and an optimization algorithm are trained, and the position and the category of the aircraft tractor can be determined. FPGA acceleration processing is adopted, so that the image processing speed and the target recognition speed are increased. The method aims at achieving automatic identification and detection of the aircraft tractor so as to improve the efficiency and safety of aviation site operation.

Description

Technical Field

[0001] This patent relates to aircraft tractor automatic identification technology, and in particular to an aircraft tractor automatic identification method based on YOLO (You Only LookOnce). The method combines computer vision and deep learning technology to achieve accurate and rapid automatic identification of aircraft tractors by training a neural network model. The invention has broad application prospects in the aviation field. Background Art

[0002] In the aviation field, aircraft tractors play an important role in pushing, towing and steering aircraft. However, traditional aircraft tractor identification methods require a lot of manpower and time, are inefficient and prone to errors. Therefore, developing an automatic aircraft tractor identification method based on YOLO has important research and practical value. Summary of the invention

[0003] The purpose of the present invention is to provide an automatic identification method of aircraft tractors based on YOLO to solve the problems in the prior art. The method comprises the following steps:

[0004] Data collection: Take a large number of images of tractors at the airport from multiple directions, collect and annotate images of aircraft tractors and related information as training data sets. In addition to shooting from multiple directions, you can also consider collecting images of tractors in different environments to increase the diversity of the data set and the robustness of the model.

[0005] Neural network design: Design and build a YOLO neural network model to detect and identify aircraft tractors. When designing the network, consider using a deep convolutional neural network (DCNN) structure to better capture the characteristics and contextual information of the tractor. Residual connections, attention mechanisms, and other methods can be used to enhance the network's expressiveness and ability to recognize objects of different scales.

[0006] Data preprocessing: Preprocess the input image data, including image enhancement, resizing, and other operations, to meet the input requirements of the neural network. Preprocessing operations can include image brightness adjustment, contrast enhancement, histogram equalization, etc. to enhance the recognizability of the image. In addition, data enhancement methods such as translation, rotation, and scaling can be used to expand the training data set and improve the generalization ability of the model.

[0007] Model training: Use the preprocessed training data set to train and optimize the YOLO model to improve the recognition accuracy and robustness of the model. You can use optimization algorithms such as stochastic gradient descent (SGD) and set appropriate hyperparameters such as learning rate, batch size, regularization, etc. to speed up the convergence of the model and improve generalization ability. During the training process, you can monitor the performance indicators of the model, such as loss function, accuracy, etc., and adjust and improve them as needed.

[0008] Automatic identification application: Apply the trained model to real-time images or video streams to achieve automatic identification and positioning of aircraft tractors. In the application stage, real-time target detection algorithms such as Kalman filter and non-maximum suppression can be used in combination with the output of the YOLO model to track and locate aircraft tractors. In addition, deep learning model compression and acceleration algorithms can be introduced to improve real-time performance and adapt to the needs of different computing platforms.

[0009] Output results: Generate corresponding outputs based on the recognition results, such as marking the location of the aircraft tractor, displaying the confidence level of recognition, and other information. In addition to basic outputs, you can also consider using technologies such as image segmentation and motion estimation to analyze and understand the environment around the tractor, further extract scene information, and assist in decision-making. At the same time, the output results can be integrated with other systems such as the airport management system to achieve intelligent scheduling and management of tractors. Advantages

[0010] The advantages of the present invention are:

[0011] High accuracy: The YOLO-based method can achieve accurate identification of aircraft tugs, reducing the possibility of misidentification.

[0012] Efficiency: Automated recognition methods can greatly improve recognition speed and save labor costs and time.

[0013] Real-time: This method can be applied to real-time images and video streams to achieve rapid identification and positioning of aircraft tractors.

[0014] Robustness: The trained model maintains high stability and reliability under various lighting, angles, and occlusion conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 :Flowchart of automatic identification of aircraft tractors

[0016] Figure 2 : Yolo network structure diagram DETAILED DESCRIPTION

[0017] A specific method for implementing the present invention is provided below:

[0018] Collect image data containing aircraft tractors and have them annotated by professionals. You can install cameras at airports and other aviation locations to collect image data regularly. At the same time, the collected image data is handed over to professionals for annotation, marking the location of the aircraft tractor in the image and other relevant information, providing an accurate training data set for subsequent model training.

[0019] Design and train a neural network model based on YOLO, using classic network structure and optimization algorithms. When designing the network model, you can refer to classic YOLO versions such as YOLOv3 and YOLOv4. At the same time, you can use common optimization algorithms such as stochastic gradient descent (SGD) and Adam to train the network model. During the training process, you can adjust parameters and improve the model according to the performance of the network model on the validation data set to improve the accuracy and robustness of the model.

[0020] Preprocess the image data, including image enhancement, resizing, etc. Preprocessing operations can include brightness adjustment, contrast enhancement, histogram equalization, etc. to improve the quality and recognizability of the image. In addition, in order to meet the input requirements of the network model, the image needs to be resized to adapt to the input size of the network model.

[0021] Model training is performed on large-scale datasets, parameters are adjusted, and network structure is optimized to improve the accuracy and robustness of the model. The collected annotated dataset can be used in combination with preprocessed image data for model training. A batch and iterative training strategy can be adopted to gradually optimize the performance of the network model. During the training process, training parameters such as learning rate, weight decay, etc., as well as network structure such as number of channels, convolution kernel size, etc. can be adjusted to further improve the accuracy and robustness of the model.

[0022] Apply the trained model to real-time images or video streams to automatically identify and locate the position and attribute information of aircraft tractors. Embed the trained YOLO model into the real-time image or video processing system to achieve real-time recognition and positioning of aircraft tractors. The model processes the image or video frame and outputs the position coordinates, bounding box information, and confidence level of the tractor to accurately identify the aircraft tractor.

[0023] Generate corresponding outputs based on the recognition results, such as marking the location of the tractor, displaying the confidence of the recognition, and other information. Based on the results of the model output, the location of the tractor can be marked in the image or video, which is convenient for users to observe and judge intuitively. At the same time, the confidence of the recognition can be displayed in the form of numerical values ​​or colors to help users understand the credibility of the recognition results.

Claims

1. A YOLO-based automatic identification method for aircraft tractors, characterized in that: Step S1, collecting and labeling aircraft tractor image samples to establish a data set for training; Step S2, using the training data set to train the YOLO model, and deploying the trained model to the aircraft tractor to achieve accurate positioning and identification of the aircraft tractor; Step S3, using a laser radar and a camera to obtain three-dimensional spatial information and image features of the surrounding environment; Step S4, based on the YOLO model and the acquired data, the obtained image is processed and calculated in real time to obtain the corresponding accuracy, and the position and posture of the aircraft tractor are identified and tracked; Step S5, using machine learning and pattern recognition technology to accurately calculate and determine the distance, angle and relative position between the aircraft and the tractor; Step S6, based on the calibration data and the algorithm model, accurately locate and track the position of the aircraft tug in a real-time environment; Step S7, combining the autonomous control algorithm, adjusting the wheel speed, steering and braking of the unmanned tractor to achieve driving and stopping according to the predetermined path and stop point; Step S8, realize automatic docking and connection through the communication interface with the aircraft to complete operations such as cargo transportation, refueling, and maintenance.

2. According to the YOLO-based aircraft tractor automatic identification method according to claim 1, the YOLO model adopts an optimization algorithm. The multi-scale training, that is, the object is very small or located at the edge of the image, which may be detected incorrectly or cannot be detected. The performance of the YOLO algorithm is improved by using images of multiple scales during training. In the YOLO algorithm, an attention mechanism can be used to enhance the model's detection ability for small objects and dense objects. The attention mechanism is used to weight the information of different feature maps so that the model pays more attention to certain specific areas. Multiple detectors are used to gradually screen and correct the detection results.

3. According to the YOLO-based automatic identification method for aircraft tractors according to claim 1 or 2, the laser radar and camera respectively obtain information about the surrounding environment through laser beam and image acquisition.

4. The YOLO-based aircraft tractor automatic identification method according to claim 1, wherein the autonomous control algorithm adjusts the driving state of the unmanned tractor to achieve accurate docking based on pre-calculated path planning and real-time position tracking data.

5. The YOLO-based automatic identification method for an aircraft and a tractor according to claim 1, wherein machine learning and pattern recognition techniques are used to accurately calculate and determine the distance, angle, and relative position between the aircraft and the tractor.

6. According to the YOLO-based aircraft tractor automatic identification method of claim 1, the communication interface is used to realize the automatic docking and connection between the unmanned tractor and the aircraft, and the required data is packaged for transmission according to the communication format prepared in advance, with the data frame header as && and the frame tail as &.

7. The image training according to claim 1 increases the diversity of the training data by randomly transforming the training data, thereby improving the generalization ability and robustness of the model. In the YOLO algorithm, data augmentation is used to improve the performance of the model.

8. The YOLO-based automatic identification method for aircraft tractors according to claim 1. The scope of protection shall be based on the YOLO-based automatic identification method for aircraft tractors according to claim 1, and shall include all technical elements and steps required to implement the method. In addition, this patent also includes the optimized YOLO model according to claim 2, the method for obtaining environmental information through lidar and camera according to claim 3, the autonomous control algorithm based on pre-calculated path planning and real-time position tracking according to claim 4, the machine learning and pattern recognition technology for calculating and judging the positional relationship between the aircraft and the tractor according to claims 5 and 6, and the communication interface and related methods for realizing automatic docking and connection between the unmanned tractor and the aircraft according to claim 7.

Images