A Survey on SAR ship classification using Deep Learning

TL;DR

This review addresses the problem of classifying ships in SAR imagery using deep learning, compiling 187 papers and deriving a four-dimensional taxonomy spanning DL architectures, datasets, augmentation methods, and learning techniques. It highlights the dominant role of CNNs, the pivotal benefits of data augmentation and transfer learning under data scarcity, and the value of integrating handcrafted features and polarization information. The OpenSARShip and FUSARShip datasets serve as key benchmarks, but class imbalance and limited high-resolution data remain central challenges. The paper offers concrete guidance on dataset curation, model design, training strategies, and evaluation practices, and identifies future directions including standardized metrics, interpretability, novel architectures, and stronger interdisciplinary collaboration to advance maritime surveillance capabilities.

Abstract

Deep learning (DL) has emerged as a powerful tool for Synthetic Aperture Radar (SAR) ship classification. This survey comprehensively analyzes the diverse DL techniques employed in this domain. We identify critical trends and challenges, highlighting the importance of integrating handcrafted features, utilizing public datasets, data augmentation, fine-tuning, explainability techniques, and fostering interdisciplinary collaborations to improve DL model performance. This survey establishes a first-of-its-kind taxonomy for categorizing relevant research based on DL models, handcrafted feature use, SAR attribute utilization, and the impact of fine-tuning. We discuss the methodologies used in SAR ship classification tasks and the impact of different techniques. Finally, the survey explores potential avenues for future research, including addressing data scarcity, exploring novel DL architectures, incorporating interpretability techniques, and establishing standardized performance metrics. By addressing these challenges and leveraging advancements in DL, researchers can contribute to developing more accurate and efficient ship classification systems, ultimately enhancing maritime surveillance and related applications.

Figures (14)

• Figure 1: Screening flow.
• Figure 2: Taxonomy of DL methods for SAR ship classification. The taxonomy organizes prior work into four dimensions: (i) architectures, (ii) datasets, (iii) image augmentation, and (iv) learning techniques. $^{\ast}$ denotes CNN variants specifically designed by the respective authors.
• Figure 3: Overview of DL techniques applied to SAR ship classification. The methods are grouped into major categories, including core Custom CNN Architectures, architecture enhancements, temporal and hybrid approaches, lightweight and NAS-based models, metric and few-shot learning, training strategies, and data-centric methods. Representative studies for each category are indicated by citation keys.
• Figure 4: Overview of handcrafted feature integration approaches in SAR ship classification. The methods are organized into four categories: (i) fusion of handcrafted descriptors with deep features, (ii) use of classical and signal-domain descriptors, (iii) advanced learning strategies incorporating handcrafted cues, and (iv) metadata- and structure-driven feature designs. Representative studies for each category are indicated by citation keys.
• Figure 5: Overview of methods that explicitly exploit SAR imaging properties for ship classification. Approaches are grouped into four categories: (i) motion and phase information from complex-valued data, (ii) polarimetric fusion of VV/VH channels, (iii) scattering- and signal-conditioning techniques, and (iv) resolution- and regime-specific strategies. Representative studies for each category are indicated by citation keys.