Research On Ship Target Detection Algorithm In Wide-range Remote Sensing Image

The rapid development in recent years,remote sensing for earth observation technology,all kinds of high resolution optical imaging satellites constantly emerging,now available on the remote sensing image resolution is better compared to before,can be more easily in the image directly capture the appearance characteristics of all kinds of feature information,in the field of military defense and civil economic construction has important application.As an important carrier of water transportation,ships mainly operate on the sea surface of wide rivers.Due to the lack of ground equipment and little information about other ground objects,ships mainly operate on water and are easy to be captured by aerial detectors.Therefore,they are a common target object to be detected by earth observation technology.Target detection in remote sensing images is of great significance to maritime rescue,military application and ship management.The detection of remote sensing ship target has become an important branch of remote sensing earth observation technology.At present,the main problem in this field is that the acquisition of remote sensing images is restricted by many aspects,such as the shooting time,weather,illumination,and various influencing factors on the sea,which hinder the detection of targets.At the same time,with the improvement of imaging resolution,the amount of data to be processed by the algorithm also increases significantly,which brings higher requirements on the speed of the algorithm.How to overcome all kinds of influences and maximize the rapid and accurate completion of detection tasks from a large amount of data has become the main research direction in this field.Aiming at the existing data resources,this paper carries out ship detection for large and wide remote sensing images,and speeds up the detection speed of the whole algorithm while ensuring the detection accuracy.The main content of this paper can be divided into the following two parts:1.Research on candidate region extraction algorithm based on significance model of spectral residualIn order to reduce the detection field quickly and reduce the computational burden of the whole algorithm,a candidate region extraction algorithm is designed in this paper.The saliency map is calculated on the basis of spectral residual saliency algorithm,and the calculation method of scale selection and segmentation threshold is improved in combination with the characteristics of algorithm and remote sensing image,which can effectively overcome the interference caused by cloud,waves and other background and minimize target omission.Finally,the noise and other obvious false alarms are filtered by a few features.2.Research on detection algorithm of small and medium targets in wide remote sensing image and light ship target in complex backgroundThis paper proposes a ship target detection method based on large area remote sensing images.The whole algorithm is divided into candidate region judgment and target fine detection.Because the wide image cannot be directly used as the detection network input,the image needs to be clipped to the size that can be input.In terms of candidate region location,a region extraction method based on image saliency based on the first part of the study is adopted to judge whether there is a suspected target in the clipped sub-image to ensure subsequent detection.In terms of detection network,in order to ensure the speed of detection,y OLOV4-TINY lightweight target detection model is selected as the main body,and the detection accuracy of the algorithm is reduced due to the large number of ships and small targets in the data.Attention mechanism and data enhancement module are added in the network,and adaptive prior box selection method is adopted to deal with the large number of small targets in the image and the changes in size and shape brought by data enhancement.The detection effect of the model for such samples is enhanced.Compared with the original yolov4-tiny network,the improved algorithm enhances the detection ability of the target under the condition of ensuring the detection speed.While ensuring the accuracy of detection,the recall rate is greatly improved and the missed detection is reduced.At the same time,compared with the pure target detection method based on deep learning,the overall algorithm not only ensures the accuracy,but also reduces the amount of calculation and improves the detection efficiency,which can be used as a reference in practical engineering.