| In recent years,the air crashes and shipwrecks occurred frequently.As a result,they have brought great loss to our life and property.Timely and effective maritime search and rescue is the most effective way to reduce the impact of maritime accidents.However the marine search for the accident site is the premise of maritime rescue.Cur-rently marine search mainly relies on the human search over large area.It is particularly difficult in the deep sea or in bad sea conditions.As a result the success rate is relatively low.Remote sensing technology has macroscopic,real-time and flexible,plays an im-portant role in marine search.By integrating macroscopic and microcosmic characters,and complementing spectral information and spatial information,hyperspectral and high-resolution,remote sensing technology can effectively complete the maritime search target detection.However there are still many shortcomings in the research of marine search with the help of remote sensing,for example simplex data source,incom-plete typical search objectives.It is far away from meeting the demand of rapid re-sponse and high precision in marine search.This paper uses multi-source remote sensing images as data sources,proposes rap-id and effective detection methods towards different typical marine search targets.Firstly,using high resolution images to detect oil slick targets,this paper puts forward a GBVS saliency model(bottom-up)and a spectral similarity saliency model(top-down),positioning large format images of complex scenes into simple and small images.After that,the oil slick is extracted by genetic neural network algorithm.The run time of the whole ROI detection module meets the emergent need in marine search.And the seg-mentation algorithm works well in oil extraction.Secondly,there are many existing limitations in endmember determination,including the low efficiency and precision.In the research of oil slick detection using hyperspectral images,this paper proposes a data field-spectral preprocessing module to complete the endmember determination process and then uses FCLS model to invert the abundance.It is demonstrated that the proposed DSPP has advantages of precision,noise immunity and efficiency.Thirdly,in order to detect an offshore ship in distress,this paper uses high resolution images to calculate NDWI and make morphological operating,to separate sea and land.Then the accurate detection of ship targets is achieved by extracting the region of interest and segmenting the image using multi feature information of remote sensing images.The results show that MRF texture based model does better than the shape-based active contour evolution by level set.At last,using hyperspectral images to detect the artificial debris target,this paper proposes a DWT-SVDD anomaly detection algorithm and compares it to RX,WKRX,DWT-SVDD is more sensitive in both pixel and even the subpixel of 1/9 than the other algorithms. |
