Strategies for Marine remote sensing target detection based on deep learning model and internal wave simulation imaging technology

Abstract

As high-resolution optical remote sensing imaging technology increasingly develops, there is an urgent need for a rapid and reliable target detection technology to identify important targets in remote sensing images. Meanwhile, the target detection technology at sea surface plays an important role in the Marine safety inspection and cargo transportation. However, current Marine targets detection still faces challenges. To this end, this paper constructed a target detection model for Marine remote sensing images based on deep learning. First of all, the saliency image of the dataset was obtained using the FAST-MBD saliency fusion algorithm, which shares the label with the original dataset for training. Then, in order to increase the robustness of the model, a variety of data enhancement methods were adopted, and K-means algorithm was introduced, which complete the data preprocessing before model training. Afterwards, the performance of the model was evaluated after the model was fully trained. Concretely, the impacts of RNMS, K-means algorithm, saliency image fusion and data enhancement on the model detection effect were studied by ablation experiments, the influences of the size and span of images after cutting on the model detection effect were analyzed using dataset cutting method, and the ability to detect submarine wake of the model was verified by the submarine wake dataset constructed. Finally, it is concluded according to the results analysis that in the same environment, this model has a extensive ability of target detection as well as certain theoretical significance in this field, which compares with other advanced models.