Data-augmented vision system for maritime object detection

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Robust and versatile detection of maritime vessels present in aerial images is a considerable challenge. While neural networks, particularly convolutional neural networks (CNNs), have revolutionized object detection and classification across many industries by enabling machines to learn complex patterns and features from large datasets, maritime vessel detection continues to pose challenges. One challenge is the limited quantity and diversity of training data required by AI/ML systems. In this paper, we present a system which uses multiple sensors in conjunction with salient data augmentation techniques and multiple convolutional neural network (CNN) architectures to test cross-sensor object detection resiliency. Our system is composed of six main subsystems: Image Acquisition, Image Processing, Data Augmentation, Model Creation, Object-of-Interest Detection and System Validation. We show that the data augmentation subsystem improves cross-sensor vessel detection precision by over 10%, paving the way for the design of similar systems which can prove robust across maritime applications, sensors and dataset sizes.