Beacon light quality intelligent recognition method based on deep learning

Abstract

The invention provides a beacon light quality intelligent recognition method based on deep learning. The method comprises the steps: collecting video data, and carrying out the preprocessing; constructing a lamp flashing network, and performing training; the light flashing network takes the brightness channel image of the preprocessed video as an input, uses a recurrent neural network to extract brightness features of the beacon light, and then uses the recurrent neural network to extract a light flashing period time sequence formed by the brightness features; constructing a color network andcarrying out training; the color network takes an RGB image of the video as an input, and color features of the beacon light are extracted by using a convolutional neural network; and fusing the recognition results of the light flashing network and the color network to obtain the light quality classification of the beacon light color plus the light flashing period. According to the invention, thelight flashing network and the color network are mainly used to train and identify the light flashing period and the color separately, and the multi-label classification effect of the beacon light quality is achieved. And an attention mechanism is added, so that the feature map can more easily pay attention to the lamp quality information to be observed during training, and the accuracy of the model is improved.

Description

technical field [0001] The present invention relates to the technical field of identification of the quality of navigation lights, in particular, to an intelligent identification method for the quality of navigation lights based on deep learning. Background technique [0002] A navigation mark is a facility or system that helps guide ships to navigate, locates and marks obstructions and expresses warnings, and provides safety information for various water activities. It is usually installed in or near navigable waters to mark the positions of waterways, anchorages, shoals and other obstacles. Aids to aids include visual aids to aids, acoustic aids to aids and radio aids to aids to three categories, among which visual aids to aids are the most convenient and most important aids to use. In order to enable drivers to quickly identify water areas through direct observation, the visual aids to navigation have distinctive body colors and shapes for daytime observation and identif...

AI Technical Summary

Problems solved by technology

The present invention mainly utilizes the light flash network and the color network to separate the light flash period and color for training and identification, achieves the multi-label classification effect of the light quality of the navigation mark, and can even identify the light quality types that do not exist in the training video data set, and solves the problem of data under-set problem

Images