Reliability Analysis of Regional Features in Visible Light Images of Maritime Ship Targets

Given the current state of visible light monitoring and recognition of ship images, this paper proposes a method for analyzing the regional features of maritime ship targets based on visible light image data. By performing preprocessing operations such as grayscaling, binarization, and edge extraction on ship images, better input is provided for the extraction of Hu invariant moments and Angular Radial Transform (ART) descriptors in this experiment. The paper selects the two most discriminative feature values of Hu invariant moments, retaining their sensitivity to image brightness and contrast, to form a new feature vector, NEWHu. NEWHu reduces the dimensionality of the feature vector, simplifying the computation and storage process while reliably distinguishing different types of ships. The paper proposes linearly weighted fusion of NEWHu and ART features into a feature vector, NHA, for the classification of maritime ship targets. NHA captures multiple aspects of the target's shape, texture, and spatial distribution, allowing for more accurate and reliable classification of different types of ships. This comprehensive feature vector enhances the recognition accuracy and improves the distinction between different ship types compared to Hu invariant moments, NEWHu, and ART. Therefore, this method offers greater practicality and reliability in naval military and monitoring applications.