Research On UAV Image Recognition Algorithm Based On Deep Learning

As an important research direction in the field of computer vision,UAV image recognition technology has important research significance and application value in both military and civilian fields.The UAV image recognition technology is mainly to analyze and process the pictures taken by the monitoring system to accurately identify the UAV target.This paper mainly studies the design and implementation of UAV recognition algorithm based on deep convolutional neural network model to complete the task of UAV classification and recognition.Since there is currently no public drone data set,for specific research needs,this article collected 6 common drone image data sets,including 3,500 images of different background environments,different flight conditions,and different scales.Secondly,according to different scale data sets,two methods of UAV image classification and recognition are designed.A low-altitude UAV recognition method based on the CBAM-AlexNet network is designed and implemented,which can complete the classification and recognition of three types of UAVs.Through the design of multiple sets of comparative experiments,the optimal UAV target classification and recognition model is designed according to the experimental results,and then the convolutional attention module is introduced to enhance or suppress the feature map elements,and the batch normalization layer is introduced to accelerate the model convergence and improve Generalization.It has been verified by experiments that compared with other neural network models,it has the advantages of high recognition rate and fast convergence speed.The fourth chapter of this paper designs and implements the UAV recognition method based on SE-InceptionV2 network.The preprocessed UAV images are first input into the model for training,using the feature extraction capabilities of the InceptionV2 model,combined with the SE-Net module to enhance useful feature channels,weaken useless channels,enhance the feature learning capabilities of the model,and introduce batch normalization layer accelerates the convergence of the model,and completes the classification and recognition of 6 common UAV models in the fully connected layer.Experiments have verified that compared with other models,the method in this paper has improved recognition accuracy and execution efficiency to varying degrees,which can effectively solve the problem of low-altitude UAV classification and recognition.