| In recent years,marine resources are the key concern of the state.Ships are only the most important means of transportation and carriers on the sea,and also the key targets in military activities.The supervision of the sea needs to obtain the information of marine targets quickly and efficiently.Image target detection and recognition has become an important means of sea monitoring,so the detection and recognition of ships are available in both civil and military fields It has a broad application prospect.Based on the above application scenarios,this paper studies the detection of aerial photographing of marine vessels,and designs and implements a detection and recognition system of marine vessels.Due to the influence of weather and light,and the weakening of ship image characteristics,the traditional ship target detection can not meet the needs of ship intelligent detection.In this paper,we choose the target detection method based on deep learning to detect and identify the ships on the sea.By investigating the ship target detection algorithm based on deep learning,we use the yolov3 algorithm based on the regression idea.In view of the characteristics of aerial ship images to improve the detection accuracy of the yorov3 network model,this paper improves the algorithm from two aspects: using dense module connection instead of the original features of yorov3 to extract the residual module in the mesoscale 2 and scale 3 prediction output layer of the network;for multi-scale detection,the 3-scale detection of yorov3 is upgraded to 4-scale detection.In this way,the improved yolov3 algorithm has better detection accuracy,also has some improvement in the detection of small targets,and can meet the real-time detection.In order to further improve the detection accuracy of small target ships and prevent the situation of missing detection.In this paper,another algorithm based on depth learning is used to detect small target ships in aerial images.In this paper,a feature fusion based SSD algorithm is proposed,which is based on SSD.The lightweight vgg16 network is selected as the backbone network of feature extraction,and then the feature fusion module is designed,so as to get the rich features of structure level.In this paper,an improved SSD method is proposed,which combines the low-level features with the high-level features.Low level features contain more small target information but less semantic information.High level features contain more semantic information but less small target information,so it is difficult to detect small target information in low level features and high level features.After pooling the features of the lower layer,and then fusing them with the features of the higher layer in series,the features of the lower layer will be added to the features of the higher layer,so as to get rich features to improve the accuracy of small target detection.The experimental results show that the feature at conv4＿3 + conv7 is the feature to be fused.Finally,according to the research results of this paper,combined with the actual needs.This system is composed of PC terminal vessel detection and identification software based on QT and Web terminal detection information management system based on Django framework.It designs and builds database to manage user information,detection information,report information file and model file.After functional testing,the system can meet the performance requirements and has high application value. |
