Vehicle Detection From Remote Sensing Images Based On Deep Learning

With the rapid development of China's economy,the number of vehicles is increasing,which leads to the increasingly serious traffic congestion.A wide range of vehicle detection can obtain traffic information in time,which is very helpful for monitoring,managing,and scheduling road traffic.Compared with ground sensors,it is more convenient and economical to use satellite remote sensing and aerial remote sensing to obtain city traffic information.In recent years,deep learning has developed rapidly and has been well applied in the fields of speech recognition and face recognition.Based on the analysis of deep learning techniques and their characteristics,we studies vehicle detection from remote sensing images based on deep learning to improve the detection performance of the current system.Firstly,the Faster R-CNN is introduced to detect the vehicle target,and the characteristics and disadvantages of the Faster R-CNN in remote sensing image vehicle target detection are analyzed.On account of the characteristics of small vehicle target and similar vehicle structure in remote sensing image,we proposes a vehicle target detection algorithm:detection based on SOSCN.Compared with Faster R-CNN,this method has higher detection efficiency and accuracy.Finally,we design the small object recognition network SORCN,adds the batch normalize layer in the network,and adopts the focal loss training network to improve the performance of the detection method.In addition,we adds road segmentation in the segmentation network,this can effectively reduce the scope of detection.Compared with Faster R-CNN,the performance of the proposed detection algorithm is improved by 55.96%.The main work of this paper is as follows:1)Introducing the Faster R-CNN network and expounding how to effectively train the Faster R-CNN network for use in remote sensing image vehicle target detection.2)Aiming at the small scale of vehicle targets in remote sensing images,a network SOSCN dedicated to segmenting small targets is proposed,and remote sensing image vehicle target detection is realized based on SOSCN network.3)Proposing a small object recognition network SORCN.The batch normalization and focal loss functions are applied to the SORCN-based detection method,and add road segmentation to the detection method.