Ship target accurate detection and segmentation method

Abstract

The invention belongs to the field of machine vision and image processing, and relates to a ship target accurate detection and segmentation method. The method comprises the following steps: (S1) training a deep convolutional neural network model; And (S2) obtaining a to-be-detected and segmented image, inputting the to-be-detected and segmented image into the deep convolutional neural network model in the step (S1), and outputting a ship target result. The method can effectively realize accurate detection and segmentation of the ship target, and has high detection accuracy for dense targets, side-by-side targets and near-shore targets. According to the method, a rotating frame prediction method is adopted, so that a candidate region and a true value frame have a relatively high cross-to-parallel ratio, confidence coefficients, positions and segmentation masks of targets are output in parallel by setting three independent loss layers, network training is carried out through targeted training data amplification, and the robustness of a model is improved.

Description

Technical field [0001] The invention belongs to the field of machine vision and image processing, relates to the design and training of a deep learning model, and realizes a method for accurate detection and segmentation of ship targets in a complex background. Background technique [0002] As a water transport carrier and an important military target, ship targets are of great practical significance for precise detection and segmentation. For example, ship search and rescue, entry and exit ship monitoring, illegal dumping of pollutants from ships, maritime traffic management, all have extremely high requirements for the accurate detection of ship targets. Convolutional neural networks and deep learning technologies continue to improve, especially in complex In the field of target detection and recognition under the background, a large amount of literature and technical experience have been accumulated. Thanks to its powerful feature extraction and learning capabilities, deep co...

AI Technical Summary

Problems solved by technology

[0004] (1) Ground objects such as trestles, wharves, loading and unloading platforms, docking piers, containers, and work sheds are very similar to ship objects in the top view image, which is very easy to cause false detection and interfere with the detection results

[0005] (2) For some dense targets: such as multiple ships side by side or connected end to end, it is easy to be mistakenly detected as the same target in high-resolution images

Images