Title: Ship identification from SAR image using novel deep learning method with reduced false prediction

Authors: J. Anil Raj; Sumam Mary Idicula; Binu Paul

Addresses: Department of Computer Science, Cochin University of Science and Technology, Kochi 682022, Kerala, India; Department of Electronics and Communication, Muthoot Institute of Technology and Science, Kochi, India ' Department of Computer Science, Cochin University of Science and Technology, Kochi 682022, Kerala, India ' Division of Electronics, School of Engineering, Cochin University of Science and Technology, Kochi 682022, Kerala, India

Abstract: Many research works using deep learning techniques for automatic vessel (or ship) detection from SAR images has good detection accuracy. But the main problem in these methods is false detection, primarily due to speckle presence. Therefore, we propose a novel pre-processing and deep learning model for vessel detection to address this problem. First, generate a three-channelled image (SarNeDe image) from a greyscale SAR image. Then, this image is used to train the model to predict the vessel's position in the SAR image. We studied the performance of different models using the SarNeDe technique and designed a lightweight model with the highest detection accuracy. We experimented on the public SAR ship detection dataset (SSDD) and dataset of ship detection for deep learning under complex backgrounds (SDCD) to validate the proposed method's feasibility. The experimental results indicated that our proposed method's vessel detection accuracy has increased with a reduced false detection percentage.

Keywords: convolutional neural network; image processing; ship detection; SAR target detection.

DOI: 10.1504/IJCVR.2022.123901

International Journal of Computational Vision and Robotics, 2022 Vol.12 No.4, pp.411 - 425

Received: 15 Feb 2021
Accepted: 10 Jun 2021
Published online: 04 Jul 2022 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article