Ship classification based on convolutional neural networks

Abstract

Ship classification in optical images has been challenged by the complexity of various ships, different imaging conditions, and limited labelled images. Traditional methods focus on extracting handcrafted features for classification, but often fails to design well-performed features for complex images. Here, the authors propose a ship classification approach with CNN. It is capable of learning discriminative features itself by supervised learning and achieving good classification performance. They build two small datasets of optical ship images for training and validation, and conduct several experiments. The experimental results indicate that their approach is effective for ship classification.

Keywords

• learning (artificial intelligence)
• ships
• image classification
• feature extraction
• convolutional neural nets
• limited labelled images
• different imaging conditions
• optical images
• convolutional neural networks
• optical ship images
• achieving good classification performance
• supervised learning
• ship classification approach
• complex images
• handcrafted features
• traditional methods focus