Research On Target Hierarchical Recognition Of Remote Sensing Images Based On Support Tensor Machine

Target detection and recognition on images is one of the main application of space remote sensing technology.In recent years,with the resolution increasing of remote sensing images,more and more information of targets is dug out from images,which supports target description.Therefore,it attract much attention on target details research in remote sensing application area.However,the increasing of resolution also brings out lots of problems while handling remote sensing images.New challenges occurs in traditional methods,for example,if using the usual method to describing features in images with vector when we deal with the high-resolution images of high data size,it may cause a huge operand and a low efficiency.On the other side,it cannot satisfy the demand of a hierarchical recognition from targets to their details.In order to solve these problems,the paper presents a target hierarchical recognition method of remote sensing images based on support tensor machine model.We describe images with tensor,and identify the focused target hierarchically which detect the partial details ulteriorly so as to describe and confirm targets better.Firstly,the paper introduced the exaction of local spatial features,and then sets up a model of feature tensor expression.Remote sensing images can be regard as tensors,so it is straighter to show the spatial coordinate and spectral information of images by tensor.Therefore,we combine the image and the exacted local spatial features as a feature tensor,which can be used to train and test the support tensor model in the following study.So as to remain the spatial features of tensors,the study utilized the scale invariant feature transform and speed up robust features to build up the feature tensor model.What's more,their merits,demerits and applicable condition are analyzed.After that,the learning algorithm and process of support tensor machine classification model has been elaborated.With the application of tensor model,the spatial structure information of images themselves are utilized adequately,while reduce the curse of dimensionality in vector models as well.In this research,the support tensor machine has been trained with gradient descent algorithm.After an iterative course among feature tensors of training samples and their categories,the optimal separating hyper plane are obtained by which we distinguish targets and backgrounds.As the support tensor machine model has been got,different classifiers are trained by different training samples.Experiments on planes and ships as well as some of their details are tested in the paper,where the accuracy and reliability of support tensor machines are confirmed in the application of target hierarchical recognition.Results shows it is available to identify targets hierarchically with support tensor machine models.The method has practical significance and application value.