Ship Detection In Satellite Remote Sensing Images Based On Deep Learning

With the rapid development of space technology,satellite remote sensing imaging has become the main technical means and data source for space-to-surface observation.With space platforms such as satellites,we can use high-resolution optical remote sensing images to complete ocean monitoring according to different mission requirements.At the same time,the continuous improvement of the imaging resolution of optical remote sensing images also brings a wealth of texture detail data and complex useless information and other interference,resulting in a slight lack of ship detection under the traditional target detection method.Therefore,this paper introduces the deep learning theory into the ship target detection,and proposes a remote sensing image detection method based on deep learning.It also improves and optimizes the small target detection task of the ship in the remote sensing image.The main contents of the thesis are as follows:Firstly,this thesis focuses on the traditional ship target detection algorithms such as candidate region redundancy,poor feature robustness,low detection accuracy,and poor ability to resist complex background interference.This paper introduces the deep learning theory into the ship of remote sensing image.In the ship target detection task,the Pytorch deep learning framework is used to construct the single-step detection model YOLO based on the regression method,and the optical remote sensing image data set for the target detection model training and testing is established.Then,in order to solve the problem that the background information of satellite optical remote sensing image is complex and the image is susceptible to electrical interference,an image preprocessing strategy is proposed.The method of filtering noise reduction,target enhancement and cloud processing is used to improve the quality of remote sensing image.Get better detection results.Through experimental tests,the detection model has a high detection accuracy of the ship target and reaches the expected detection level.Finally,for the small target missed detection of YOLO model,this paper proposes to construct a new optimization scheme of feature fusion target detection layer capable of outputting 4 times feature map to extract more small target feature information,thus obtaining target for remote sensing ship.The detection model with stronger detection performance,at the same time,uses the TX2 calculation terminal to optimize the hardware of the ship target detectionsystem,and realizes the detection model to improve the detection speed of the ship.