Ship detection deep neural network algorithm based on an image

Abstract

The invention discloses a ship detection deep neural network algorithm based on an image, and the algorithm comprises the steps: 1, carrying out the up-sampling of the image, and enabling the length and width of the image to be twice of the original length and width; 2, dividing the image into S * S grids, providing B boundary predictions for each grid, and giving six parameters including a ship position, a confidence coefficient and a classification probability for each boundary prediction; 3, extracting the features of each grid through a cascaded hole convolutional neural network, achievingthe multi-resolution ship boundary prediction through feature fusion, and determining the position of a ship; and 4, designing a loss function, and balancing the prediction frame with the ship body part and the prediction frame without the ship body part by setting different scaling factors. Based on the research of the convolutional neural network in the field of computer vision, classificationand regression are fused in one deep neural network for multi-target real-time detection on the basis of feature learning, and the method has very high accuracy and rapidity.

Description

technical field [0001] The invention belongs to the field of ship detection, in particular to an image-based deep neural network algorithm (SD-DCNN) for ship detection. Background technique [0002] Image-based automatic detection of ships is one of the basic issues for automatic and safe driving of ships, and it is also an important issue in the field of computer vision. Especially small ships, because they are not equipped with AIS (Automatic Identification System) and other equipment, they cannot actively report their own position and other information, and are not easy to be automatically sensed by other ships, which may easily cause ship collisions. The traditional ship detection method uses manual features, the accuracy and robustness of the algorithm are not enough, and there will be false detection and missed detection. The existing methods based on deep learning can realize the automatic detection of ships. However, it is easy to miss detection for small ships, whi...

AI Technical Summary

Problems solved by technology

Especially small ships, because they are not equipped with AIS (Automatic Identification System) and other equipment, they cannot actively report their own position and other information, and are not easy to be automatically sensed by other ships, which may easily cause ship collisions

The traditional ship detection method uses manual features, the accuracy and robustness of the algorithm are not enough, and there will be false detection and missed detection

The existing methods based on deep learning can realize the automatic detection of ships. However, it is easy to miss detection for small ships, which affects driving safety.

Images