Ship Detection Image Dataset

Description

The dataset used in this study consists of 6,468 images collected from multiple publicly available internet resources, including Roboflow public datasets, Kaggle image repositories, the Airbus Ship Detection Challenge dataset, Google image search, and additional images obtained by contacting authors of relevant published studies. All images were manually annotated and labeled using two classes: Object (1) and Background (0). The Object class represents any visible vessel present in the scene, while Background indicates images containing only sea surface without detectable ships. The dataset includes images captured from Unmanned Aerial Vehicles (UAVs) and Unmanned Surface Vehicles (USVs) under diverse environmental conditions. These conditions include different weather scenarios, varying illumination levels, and both daytime and nighttime observations. The images contain multiple vessel categories such as cargo ships, passenger ships, fishing vessels, and other maritime vessels. The purpose of constructing this dataset is to support the development and evaluation of robust maritime object detection models capable of identifying ships on sea surfaces under realistic operational conditions. The diversity of viewpoints, lighting conditions, and vessel types ensures that the dataset reflects real-world maritime monitoring environments and improves the generalization capability of the trained models.

Steps to reproduce

The dataset was constructed through a multi-stage image collection and filtering process aimed at supporting maritime object detection research. Image Collection Images were collected from several publicly available internet sources, including Roboflow public datasets, Kaggle repositories, the Airbus Ship Detection Challenge dataset, and Google image search. Additional images were obtained by contacting authors of relevant maritime computer vision studies and requesting sample images. Data Filtering and Cleaning All collected images were manually inspected to remove duplicates, corrupted files, and images unrelated to maritime environments. Only images clearly showing sea surface scenes were retained. Scene Diversity Selection The dataset was designed to include diverse maritime conditions, including: Different weather conditions Daytime and nighttime observations Different viewpoints, including images captured from UAVs and USVs Various vessel types such as cargo ships, passenger ships, and fishing vessels Dataset Structure The final dataset consists of 6,468 images representing maritime scenes. Some images contain vessels, while others contain only sea surface.

Institutions

• Southampton Solent University

  England, Southampton

Categories

Licence