Research On Ship Target Detection And Recognition Methods Based On Visible Remote Sensing Images With Different Resolutions

Ship is an important military and civil target,and it is strategic and valuable to study ship detection and recognition for remote sensing images.Aiming to meet the demands of large-scale maritime ship target monitoring based on visible remote sensing images(VRSI)of different resolutions,this article first investigates on ship target detection methods based on VRSI of medium and low resolution in a large area,and then studies the detection and recognition methods based on VRSI of high resolution.The above researches can lay the foundation of collaborative detection and recognition of ship target in a large scale based on VRSI of different resolutions.The main research contents of this article are summarized as follows:(1)Aiming at the problems of the weak characteristics of ship target and the interference of complex background in the ship detection under complex seas in VRSI of medium and low resolution,a saliency detection algorithm that fuses visual saliency features of spatial and frequency domains is proposed.First,the images are divided into blocks,and the features of spatial domain are obtained by measuring the difference of the feature covariance matrices between the image blocks.Then,the features of frequency domain are extracted by the Phase Spectrum Quaternion Fourier Transform algorithm(PQFT).Finally,the cellular automata algorithm is introduced to complete the ship detection by fusing visual saliency features of spatial and frequency domains.Experimental results show that the proposed method can suppress the complex background while highlighting the ship target.(2)In the ship target detection and recognition processes in VRSI of high resolution,the conventional methods based on artificial features and classifiers fail in ship classification and recognition,and the deep learning methods are inappropriate for processing VRSI of a large and wide range.To solve the above problems,an algorithm for ship detection and recognition under various detection backgrounds based on a coarse-to-fine strategy is proposed.Under the port background,a ship detection method based on edge line feature location and channel feature aggregation is proposed;In the sea background,a ship detection method based on visual saliency and channel feature aggregation is proposed;Based on the above detection results,the Yolov3 algorithm is improved by rotating rectangular box regression,which manages to fine-tune the detection positioning results and achieve classification recognition.Experimental results show that the proposed methods can detect and recognize ship targets under various backgrounds in VRSI of different widths and ranges.(3)Software design of ship target detection and recognition.Based on the aforementioned ship target detection and recognition methods,a software for ship target detection and recognition based on VRSI is designed.The software includes the modules of system management,image preprocessing,ship detection in medium and low resolution sea area,ship detection in high resolution sea area,ship detection in high resolution port area and high resolution ship recognition.