Improved Ice Channel Segmentation and Width Calculation Using Drone-Based Optical Sensors

The aim of this study is to investigate the utilization of optical sensors mounted on unmanned aerial vehicles (UAVs) for the identification of inconspicuous ice channels in polar regions, serving as an auxiliary navigation aid for polar vessels. In the previous research, there was no prior work utilizing optical-based physical sensors for the purpose of ice channel perception tasks. An algorithm for semantic segmentation, known as the ICEU-Net (ice channel segmentation network with UNet as main frame), has been proposed. It integrates ResNext and UNet architectures with customized features to enhance feature extraction capabilities. The integration of the efficient channel attention network (ECA-Net) during the multiresolution fusion and skip connection process further enhances the detection of subtle ice channel features. Comparative experiments demonstrate an average improvement of 1.2% in mean pixel accuracy and 1.88% in mean intersection over union (mIoU), confirming the effectiveness of optical sensors in providing prior information about ice channels. The width of the ice channels is determined using the flight parameters of the UAV and optical sensor parameters, after that offering information about the ice conditions for polar navigation vessels. This application marks a groundbreaking use of optical sensors in extracting polar ice channels, holding significant pioneering value.