Ship target detection method based on joint training of deep learning features and visual features

Abstract

The invention provides a ship target detection method for joint training of deep learning features and visual features, comprising the following steps: sample data collection, CNN feature extraction, traditional invariant moment feature and LOMO feature extraction, feature dimensionality reduction, and feature fusion network FCNN construction , and finally use the sample data to train the network, and use the test data to test the model. Compared with the prior art, the visual feature extraction process of the present invention comprehensively considers the characteristics of ship shape, color and texture, making the detection process interpretable, and can standardize the CNN back propagation process to learn other features than traditional features. This method is fast, efficient, and highly accurate. It still has good detection results for complex scenes such as clouds, cloudy days, and rain, and has high robustness. Features that are complementary to traditional features can be extracted, and the speed is extremely fast, which can achieve the effect of real-time monitoring.

Description

Technical field[0001]The present invention belongs to the field of vessel detection computer visual field, and specifically, the present invention relates to a depth learning characteristic and a visual characteristic joint training vessel target detection method.Background technique[0002]my country has a broad coastline, sea area, and rich marine resources. As the economy is growing, the number of sea vessels is increasing, and vessel has an urgent practical demand. The target detection of vessel is to use computer vision and image processing technology to detect the target of interest in the image, further extracting a large amount of useful information, has a wide range of application prospects in military and civilian fields. For example, in the civilian field, by obtaining information such as the location, size, driving direction, driving speed, etc., can monitor the specific sea, the bay port, and monitor the transportation, illegal fishing, illegal smuggling, illegal dump oil...

AI Technical Summary

Problems solved by technology

However, the black box-style features are poorly understood, and ships of different sizes have different degrees of feature retention after convolution, which will also lead to inconsistencies in the detection results of different ships.

Images