Search

Nowadays, remote sensing images have been widely used in civil and military fields. But, because of the limitations of the current imaging sensors and complex atmospheric conditions, the resolution of remote sensing images is often low. In this paper, a superresolution reconstruction algorithm based on the deep convolution neural network to improve the resolution of the remote sensing image is proposed. First, this algorithm learned a series of features of the mapping between high and low resolution images in the training phase. This mapping is expressed as a kind of deep convolutional neural network; the trained network is a series of parameter optimization for super-resolution reconstruction of remote sensing image. Experimental results show that the superresolution algorithm proposed in this paper can keep the details subjectively and improve the evaluation index objectively. [ABSTRACT FROM AUTHOR]

TLD (Tracking-Learning-Detection) algorithm can be good for a long time to track the target in rotation, occlusion, illumination and other circumstances. However, in the case of uneven illumination, occlusion, tracking target fuzzy and so on, the problem of false tracking or tracking failure often occurs. Aiming at the shortcomings of TLD tracking algorithm, this paper will take TLD as the basic framework of target tracking and improve the detection module. When the tracking target has better texture feature, the SLBP (Semantic Local Binary Pattern) classifier is used to replace the nearest neighbor classifier in the detection module, which converts the image into SLBP texture feature vector to classify the samples using. In this paper, TLD-SLBP, MEEM, SCM, Struck and TLD are compared by using CVPR2013Benchmark test platform. The experiment results show that the TLD-SLBP algorithm has a higher success rate than other algorithms. [ABSTRACT FROM AUTHOR]