Ship Detection and Classification in Adverse Weather Conditions

The use of Unmanned Aerial Vehicles (UAVs) is quite common today. Equipped with various tools, these vehicles are actively utilized in both civilian and military fields. Due to their high-altitude photography, images captured from UAVs often have a wide field of view. However, adverse weather conditions can reduce visibility, making it impossible for the operator controlling the UAV to see all the details. To analyze the images obtained from the UAV and improve environmental conditions, an automated system is needed. A different approach is required to mitigate the negative effects of weather and to detect small objects, which may appear smaller due to high-altitude captures. In this study, the adverse weather conditions faced by UAVs were identified, and subsequently, the Pix2Pix model was used to improve the results caused by these conditions. Enhancements were made in algorithms that were insufficient for detecting small objects. After correcting for adverse weather conditions, object recognition algorithms such as YOLOv8 and Faster R-CNN were used separately for ship detection and classification. It was found that after correcting for adverse weather conditions. there was an increase in mAP values of approximately 2% with YOLOv8 and 1% with Faster R-CNN.