Hyperspectral Image Target Detection Based On Visual Attention

With the development of hyperspectral imaging technology, the amount of data hyperspectral images is dramatically increasing. Target detection of hyperspectral images brought a series of problems, such as high-dimensional data how to fast, high-precision detective target and other issues. When hyperspectral image target detecting, if all the data using the same method, handling efficiency will be very low. At this time we can use an intelligent approach that is the human visual system of visual attention mechanism. When detecting target it can quickly select few first priority areas, then analyze extracted target areas. So the approach can have a high processing efficiency and accuracy, meet plenty of efficient and real-time applications.In this paper, considering the shortcomings of the traditional hyperspectral image target detection, such as inefficiency and low accuracy. So this paper focuses on the use of human visual attention mechanism for hyperspectral image target detection method, including three key technologies, details as follows:The first is calculating the bottom-up hyperspectral target detection visual attention model. This paper considers the hyperspectral image with spectra of unity, so we proposed to extract the primary visual features of hyperspectral image, comprising extract the edge strength features, the spectral average radiation intensity features, the spectral maximum radiation intensity features and the spectral intensity distribution features.These image features contain spectral information and image information, and according to these features the paper proposes a kind of method about calculating high-dimensional feature about saliency map.The second is calculating the top-down hyperspectral target detection visual attention model. This paper analyzes the information of the specific target spectral lines, advanced the visual features proposed calculation method for hyperspectral images, including the extraction of hyperspectral image about the difference between the peaks and valleys features of specific goals, peaks and valleys ratio features. These advanced visual features can represent the spectral information in hyperspectral image features for detecting the specific target.The third is the fusion of these two visual attention models. Bayesian model is developed and refined, the primary visual features about bottom-up model and the advanced visual features about top-down with a combination. After calculating integrated saliency map, eventually detected hyperspectral image target destination.Experimental results show that, from the subjective and objective points of view, compared a variety of algorithms, integrated model of visual attention in this paper not only has a good target detection result, complete, accurate and effective hyperspectral image target detected, but also has strong anti-jamming ability.