Navigation mark detection based on deep learning models from UAV images

Abstract Prosperous waterway economics require rigorous safety measures. Unmanned aerial vehicle (UAV) offers massive images of inland waterways, within which navigation mark detection plays a critical role in ensuring waterway safety. This paper proposes a deep learning-based method for detecting navigation marks in UAV images. Firstly, a dataset of inland waterway navigation marks is constructed from UAV aerial images, which includes data collection, image enhancement, sample creation, and sample annotation. Secondly, a deep learning network model is developed, which uses ResNet-50 as the backbone, incorporates Coordinate Attention and Large-Scale Selective Kernel Attention mechanisms, integrates a Feature Pyramid Network (FPN) for feature enhancement, and uses Distance Intersection over Union (DIoU) as the loss function. Thirdly, the model is trained and evaluated on the constructed dataset, followed by precision assessment and post-processing. This paper explore a deep learning network model for small object detection in UAV images and establish a comprehensive workflow for detecting inland waterway navigation marks, thereby providing technical support for waterway safety.