Deep neural networks-based vehicle detection in satellite images

Vehicle detection in satellite images is an Challenging task, but meaningful at the same time. This paper propose a vehicle detection method in satellite images using Deep Convolutional Neural Network(DNN). DNN is a model of deep learning and it has a high learning capacity when dealing with images. DNN consist of several convolution layers and pooling layers, the last layer is full connection with output(this can be considered as neural network). DNN can automatically learn rich features from trainning dataset, and has achieved excellent performance in many applications such as image classification and object recognition. To benefit from this method, we propose a vehicle detection framework. Firstly we use a graph-based superpixel segmentation to extract a set of image patches, which can help us locate vehicle effectively. And then we train a DNN network to classify these pathes into vehicle and non-vehicle. Experimental results indicate that the proposed method has a good performance, with high detection rates and very few false alarms for all test road segment.