Research On Accurate Recognition Method Of On-orbit Remote Sensing Ship Target Based On Artificial Intelligence

China has a vast territory,with a sea area of about one-third of its territory and a total coastline of 32,000 kilometers.Due to various reasons of history and reality,China often has frictions and disputes with other countries due to disputes over sea areas or the attribution of seabed resources.Therefore,China has always improved China's maritime military strength.The identification of maritime remote sensing ship targets has also become a research topic in China as a sea surface monitoring technology.Nowadays,with the improvement of computer hardware,the computing power of computers has also increased,and artificial intelligence technology has been widely used in various fields.Especially in the field of target recognition and detection,artificial intelligence-based target detection methods have become one of the most efficient detection methods.This paper mainly studies the met hod of accurate identification of remote sensing ships based on artificial intelligence,and has completed the following research work:(1)Identification method classification effectiveness analysisThis paper compares the traditional recognition methods and the ability of neural networks to extract features from high spatial resolution remote sensing images.By self-produced data sets from Google Earth,the experiment compares two methods and analyzes the indicators by means of average accuracy and false alarm rate.As a result,it is concluded that the ship target detection method based on artificial intelligence(Faster R-CNN)is an effective high-resolution remote sensing ship target,both in the higher accuracy and recall rate.recognition methods.small ship targets as one of the research priorities.Due to the large size of remote sensing images,small targets are almost invisible on large remote sensing images or low-resolution features,and the features are almost disappeared.In complex environments,they are more susceptible to interference from geographical environments such as land buildings and ports.In this paper,an improved method based on Faster R-CNN neural network is proposed.The FPN feature pyramid network is introduced to enhance the receptive field and enhance the recognition ability of small targets.Through the same data set test,the data based on the Faster R-CNN improved method and the original Faster R-CNN method are compared.The accuracy and recognition rate of this method are improved by 1.6% and 3%,respectively,reaching 99.4%.And 90.6%.(3)Research on ship identification under occlusion of clouds,fog,smoke,etc.The identification of ships under the cover of clouds and dense fog is a hot research direction of ship identification.The difficulty is that clouds and dense fog will obscure part of the ship's characteristics,and the features of the clouds will also be targeted as targets while extracting the target features.After learning and training,the accuracy and recall rate of the final output will be reduced,and the rate of false alarms and false alarms will increase.This paper introduces Soft-NMS and increases cascading to improve the ship identification accuracy and recall rate of this method under complex weather conditions.After testing,the accuracy and recall rate of the proposed method reached 0.8062 and 0.8876,which is better than the CVPR 2018 Cascade R-CNN method(one of the best target detection methods in the Faster R-CNN framework).The average accuracy rate is 3% higher and the recall rate is 4% higher.