Reserch On Ship Detection Technology For High Resolution Remote Sensing Images

In recent years,with the rapid development of remote sensing observation technology,optical remote sensing imaging and UAV aerial imaging technology have become increasingly mature,providing an extremely rich data source for marine target detection and recognition.Ship detection technology in remote sensing images is Both the military and civilian sectors have important economic and military values.Because the traditional ship detection technology is susceptible to cloud and clutter interference,the detection accuracy is low,and the new deep learning technology requires large computational power to process high-resolution remote sensing images.In order to solve the problems of slower target segmentation and ambiguous boundary separation in images with large complexity,a ship target segmentation algorithm combining saliency with improved Grabcut algorithm and an improved multi-scale feature pyramid ship detection frame based on regional recommendations are proposed.The research work in this paper can effectively integrate low-level location information and high-level semantic information,and provide more detailed and detailed feature information for target detection;Before processing,a series of remote sensing image preprocessing methods were proposed for the characteristics of remote sensing images and the characteristics of ship targets.The related algorithms were compared experimentally.The results show that the comprehensive performance of the model is significantly improved and improved compared with other detection models.The specific research contents are as follows:(1)Firstly,a series of remote sensing image preprocessing methods are proposed for the characteristics of remote sensing images and the characteristics of ship targets.An adaptive median filtering with Gaussian weighting is proposed for background noise;MASK and Split-Merge homogenization method;for the elimination of thin cloud effect,a dark channel a priori remote sensing image to eliminate cloud interference algorithm is proposed;the darkening ocean remote sensing image is enhanced in detail.(2)Then,a ship target segmentation algorithm based on saliency detection fusion improved Grabcut algorithm and SLIC superpixel segmentation is proposed.It can be used in the pre-processing image and segmentation stage of remote sensing image target detection and identification research.The segmented sample set will be used as the training network input for ship target detection,classification and bow direction prediction.(3)Finally,a new ship detection framework is proposed,which is a multi-scale feature pyramid ship detection model based on regional recommendations.The model includes a multi-layered feature pyramid network,improved RPN network,and improved Non-maximum suppression and multi-scale RPS-ROI alignment,which can effectivelyintegrate low-level position information and high-level semantic information,and provide more detailed and detailed feature information for target detection.(4)The experimental results show that the proposed algorithm has good robustness and adaptability for ship detection.Through experimental comparison,the overall performance is improved by at least 3.1%,and the time is increased by more than 15%.The algorithm performs well for samples with insignificant target features and multi-scale and high-density samples.