Rasearch On The Algorithm Of Aerial Vehicle Detection Based On Improved Faster R-CNN

In recent years,unmanned aerial vehicle aerial technology has gradually become an important part of intelligent transportation systems.However,in actual application scenarios,the background of aerial vehicles is complex and the scales vary.These factors limit the improvement of the performance of aerial vehicle detection models.In response to the above problems,this thesis analyzes the current deficiencies of aerial vehicle detection algorithms and gives the research plan of this thesis.The specific research contents are as follows.1)Aiming at the multi-scale problem of vehicles from the aerial perspective,an aerial vehicle detection model based on the receptive field area proposal network is proposed.Firstly,the defect that the current feature pyramid fusion will introduce additional noise information is analyzed,and an improved feature fusion method is designed.Then points out that the size of the sliding window of the region proposal network in the Faster R-CNN algorithm is too single and does not consider the correlation between the size of the anchor frame and the size of the convolution kernel,and designs the receptive field region proposal network,thus alleviate the problem of mismatch between the convolution kernel and the object scale.The experimental results on a public aerial vehicle data set show that the proposed aerial vehicle detection model improves the detection accuracy of aerial vehicle and has certain universality.2)In order to further improve the detection accuracy and speed of aerial vehicle detection algorithm based on deep learning,an aerial vehicle detection model based on more effective anchor frame generation network is proposed.Firstly,aiming at the problem that the aerial vehicle detection model based on receptive field region proposal network contains a large number of invalid anchor frames,an anchor frame location prediction branch is added to the receptive field region proposal network to predict the distribution of anchor frames,which reduces the calculation cost.Then,the necessity of balancing the multi task loss function is analyzed,the location loss function is improved,and the balanced multi task loss function is proposed.It increases the contribution of the inline sample to the loss function gradient,and improves the balance between the classification loss and the location loss function in the training process.The experimental results on the public aerial vehicle data set show that the proposed aerial vehicle detection model optimizes the training process and further improves the efficiency of aerial vehicle detection.