Oil Pollution Detection in Aquatic Ecosystems Using UAVS and Multispectral Imaging Based on Deep Learning Technologies

This paper presents a deep learning-based algorithm for identifying oil pollution on water surfaces using multispectral images from a 5-channel camera obtained from unmanned aerial vehicles (UAVs). The algorithm, based on the Unet architecture with the efficientnet-b0 encoder, demonstrates high segmentation accuracy and is part of an environmental monitoring system. Using data on natural and controlled oil spills, as well as organic discharges, the method has been field tested on various water bodies, which confirms its efficiency and reliability in the prompt detection of pollution. Particular attention in the article is paid to the accuracy and speed of the algorithm. The developed method has a high data processing speed and can be successfully applied in various climatic conditions. The results demonstrate that the proposed algorithm is able to automatically detect even minor pollution of water surfaces, which allows for a prompt response to environmental disasters and minimize their consequences. The proposed algorithm has shown high results. With the selected model configuration, the Dice Loss metrics were achieved at the level of 0.00265 and the IoU Score equal to 0.9971. These high values confirm the reliability and accuracy of the proposed approach, ensuring accurate identification of oil spills.