Near Harbor Ships Detection And Recognition In High Resolution SAR Images

As a proactive microwave radar,Synthetic Aperture Radar(SAR)has strong penetrability and can surmount the adverse climatic conditions such as sea gale,fog and so on.It is the primary choice for marine ship target monitoring.Due to frequent activities,large quantities and high density of ships in harbor,it is of great significance to use SAR to monitor the port area in real time by using high resolution SAR image.Based on the above background,this paper studies the ship target detection and recognition technology in high-resolution SAR images in near-harbor area.Because of its superior performance,the target detection technology based on deep learning develops rapidly.The sea-land segmentation algorithm has always been the pre-processing part of the traditional ship target detection algorithm.In order to solve the difficulties of landing ship detection,the academic community has been pursuing the accuracy of sea-land segmentation,but this standard is not applicable to the field of deep learning.This paper analyzes the significance of sea-land segmentation in deep learning detection algorithm,and proposes a fast sea-land segmentation algorithm based on 2d-Otsu algorithm and improved K-means algorithm.The algorithm proposed in this paper can quickly obtain the results of sea-land segmentation which is suitable for the pre-processing requirements of deep learning framework.Based on the characteristics of SAR image ship targets,this paper analyzes the applicability of horizontal and tilt detection box,and the applicability of VGG＿16 and Res Net＿101 feature extraction networks.Through analysis and experiment,it is verified that the tilt detection box and VGG＿16 network are more suitable for SAR image.In order to improve the performance,the paper proposes a contextual merging-multi-layer rotational region-based CNN framework.By adding context information and merging multi-layer feature map,this algorithm effectively improves the detection performance.There are still many false alarms in the detection results of panoramic images.This paper analyzes various features of ships,and proposes a target identification method based on multi-feature level discrimination.The proposed method effectively removes false alarms and reduces the false alarm rate.Finally,the paper explores the recognition of military and civilian ships under high-resolution images,and proposes a recognition index based on elliptical similarity.Based on the main research contents of the thesis,the SAR image ship target detection software with sea-land segmentation,target detection and false alarm identification is designed.