Performance Evaluation of Data Enhancement Methods in SAR Ship Detection

Zhang, Chi; Zhang, Xi; Zhang, Jie; Gao, Gui; Zhang, Jingke; Liu, Genwang; Jia, Yongjun; Wang, Xiaochen; Zhang, Yi; Dai, Yongshou

doi:10.1109/jmass.2022.3211256

Cited by 6 publications

(2 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In Info-NCE loss, the instance-wise discrimination can be regarded as the similarity or distance between two features generated by the nonlinear projection module. We use the cosine similarity formulated as (6) to measure the distance, where the vectors u, v are Euclidean normalized.…”

Section: A Pre-training the Network With Contrastive Learningmentioning

confidence: 99%

“…Nowadays, numerous high-resolution SAR images are available due to extensive satellite exploration activities, making automatic SAR image interpretation an urgent necessity. Target classification is a fundamental SAR interpretation task that can automatically offer target category information for other advanced applications, such as hostile target identification, high-value target surveillance [6]- [8], town planning [9], oil spill detection [10], etc.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Self-Supervised Feature Representation for SAR Image Target Classification Using Contrastive Learning

Pei,

Su,

et al. 2023

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Nowadays, the developed deep neural networks (DNN) have been widely applied to synthetic aperture radar (SAR) image interpretation, such as target classification and recognition, which can automatically learn high-level semantic features in data-driven and task-driven manners. For the supervised learning methods, abundant labeled samples are required to avoid the over-fitting of designed networks, which is usually difficult for SAR image applications. To address these issues, a novel two-stage algorithm based on contrastive learning (CL) is proposed for SAR image target classification. In the pretraining stage, to extract self-supervised representations (SSRs) from an unlabeled train set, a convolutional neural network (CNN)-based encoder is first pre-trained using a contrasting strategy. This encoder can convert SAR images into a discriminative embedding space. Meanwhile, the optimal encoder can be determined using a linear evaluation protocol, which can indirectly confirm the transferability of pre-learned SSRs to downstream tasks. Therefore, in the fine-tuning stage, a SAR target classifier can be adequately trained using a few labeled SSRs in a supervised manner, which benefits from the powerful pre-trained encoder. Numerical experiments are carried out on the shared MSTAR dataset to demonstrate that the model based on the proposed self-supervised feature learning algorithm is superior to the conventional supervised methods under labeled data constraints. In addition, knowledge transfer experiments are also conducted on the openSARship dataset, showing that the encoder pre-trained from the MSTAR dataset can support the classifier training with high efficiency and precision. These results demonstrate the excellent training convergence and classification performance of the proposed algorithm.

show abstract

Section: A Pre-training the Network With Contrastive Learningmentioning

confidence: 99%