Vehicle Extraction In UAV Image Based On Deep Learning

Traffic is closely related to human life.However,with the development of economy,traffic problems such as road congestion and frequent accidents caused by the rapid increase in the number of automobiles have become increasingly serious.Many countries have gradually established intelligent transportation systems to resolve this contradiction.The vehicle detection module is an important part of the intelligent traffic system.Accurate and efficient detection methods are of great significance for real-time analysis of current traffic conditions.Traditional vehicle target detection methods need to design recognition rules based on the characteristics of vehicles in different images.The adaptability and detection effects are not ideal.In recent years,the rapid development of deep learning has made remarkable achievements in the field of computer vision,and it has been even more brilliant in image recognition.In this paper,the deep learning method is used for the detection of vehicle targets in drone images,and the following work is mainly performed:1)Using the UAV image with higher resolution and easier access as the data source,the characteristics of the vehicle in the image are analyzed.Based on this,the detection process of the target in the drone image is studied.Using this image,a training sample data set containing the vehicle class and location information was created,and the data set was used to train and detect the deep learning detection model.2)Take the object-oriented method as an example to introduce the process and features of the traditional method for vehicle extraction: Firstly,according to the spectral and spatial distribution characteristics of the vehicle,the features in the image are segmented and extracted,and combined with the visual interpretation method.Bright and dark vehicles in the image.3)Carry out deep learning model for vehicle detection: Use RPN to generate a series of candidate regions in the sample image,and then input the generated candidate frame and sample image to Fast R-CNN for classification to obtain object category and location information.After inputting the sample data into the model,the input image is first convolved to get the global convolution feature of the image,and then the RPN further calculates the image features in the form of a sliding window on the feature map,and generates a series of inclusions in the center of the window.In the target candidate frame,Fast R-CNN is used to classify the features in the candidate frame,and the final output contains the target candidate frame and confidence level.In order to reduce the training difficulty of the model and achieve end-to-end optimization,Faster R-CNN enhances the generalization by combining the features of partial convolution layers in RPN and Fast R-CNN so that the candidate region generation process and the object classification process are unified.Ability,and achieved good test results on the sample set.By verifying the two methods on the data set,the experimental results show that the Faster R-CNN network can effectively extract the vehicle targets in the image.Different from the traditional vehicle classification algorithm,the deep learning method avoids manually designing the target features and makes the model have better generalization ability and practicality.