Ship detection methods and systems based on multidimensional features of the scene

摘要

The disclosure provides a ship detection method and system based on multidimensional scene features. The method includes: constructing a ship image sample database, and extracting all the edges of each frame of image to act as a fourth dimension of the image; extracting a coastline to make a sea surface area be a ship area; constructing a Faster RCNN-like convolutional network to act as a deep learning network, and inputting sample data into the deep learning network; constructing an RPN network, using a sliding window to generate region proposal boxes of different sizes in the ship area, combining the region proposal boxes with the deep learning network, and training a model according to an actual position of a ship; and performing ship detection on a part of the detected image between the coastline on the basis of the trained model. By extracting the coastline, the disclosure avoids interference by buildings on land and performs region proposal boxes on the ship area only, thus increasing a accuracy and a speed of the region proposal boxes; in addition, edge features are added to target detection to act as the fourth dimension of the image, which increases a detection precision and a detection speed.