Ship Detection Algorithm Research Of Optical Remote Sensing Image

Due to the fact that the optical remote sensing image are large and easily affected by many factors,such as weather,light,sea condition,imaging sensor parameters and other factors,traditional algorithm of ship detection often has high false alarm rates,poor robustness and no real-time performance and it is not suitable for application in engineering.In order to improve the detection rate of the ship target and quickly extract the ship target in real time,a ship target detection algorithm for optical remote sensing image is proposed in this thesis which mainly consists of the sea-land segmentation and detection of the ship target.The sea-land segmentation is always the most important part of ship detection in optical remote sensing image,most sea-land segmentation algorithms relying on the prior information or the difference of grayscale features may result in poor segmentation effect and wrong segmentations,which make it difficult for subsequent processing.To solve the problem,an improved algorithm is proposed,which combines ROEWA(Ratio of Exponentially Weighted Averages)operator with wavelet transform operator.First,the edge of the original image is detected by ROEWA operator to get the edge intensity.Second,this thesis uses the time-frequency characteristics and multi-scale analysis of the wavelet transform to extract the direction information of the edge.Finally,the result of segmentation is attained after regional growth.On the basis of the sea-land segmentation for remote sensing images,the research on ship target detection algorithm is carried out.After analyzing varieties of classical visual saliency models,this thesis adopts visual saliency model based on spectral residual and Otsu target threshold segmentation to obtain the possible positions of the target.Then the direction information of the ship target is calculated by using of sobel gradient operator.The top-hat transform is used to enhance the details of the ship’s target information.Finally,the SVM target recognition algorithm is used to output the real ship target.The experimental results show that the average running time of the ship target detection algorithm for optical remote sensing image proposed in this thesis is about 1.35 seconds,the detection rate of the algorithm is about 92.21% and the false alarm rate is about 11.07% while the image size is 8192*8192.Compared with the traditional method of ship detection,the algorithm proposed in this thesis can effectively reduce the false alarm and the missed detection,it has good robustness and high detection rate.