Ship Detection in Images Obtained from the Unmanned Aerial Vehicle (UAV)

Background/Objectives: The main objective of this work is the creation of a new algorithm for detecting ships in the high resolution images obtained by unmanned aerial vehicle. Method: To achieve the objective the developed ship detection algorithm uses two cascades: shape-based cascade and cascade based on automatically extracted features. Deep neural network autoencoder is applied for automatic feature extraction. Findings: Images from Google Maps, Yandex Maps, and a small fraction of the images taken from the Internet were used for training and testing the developed approach. In total, the dataset consists of 1000 images containing ships. On average, each image contains 5.1 ships. The dataset was divided into training (50%) and test (50%). Detection quality was estimated using the Recall and Precision metrics. Our study has shown that the combined use of shape-based cascade and cascade based on automatically extracted features allows achieving high quality (Recall: 0.95, Precision: 0.94) and performance in the problem of ship detection by UAV. The results demonstrate that the proposed method is comparable or slightly outperforms the state-of-the-art methods. Improvements/Applications: Applying the detector in real video from the UAV, it would be possible to improve performance by parallelization of this approach on the GPU.