**7. Conclusions**

غ

This paper proposes a novel deep learning network ShadowDeNet for the moving target shadow detection from video SAR images. Five characteristics are used to guarantee ShadowDeNet's superior detection performance, i.e., (1) HESE, which is used to enhance shadow saliency to facilitate feature extraction, (2) TSAM, which is used to focus on regions of interests to suppress clutter interferences, (3) SDAL, which is used to learn moving target deformed shadows adaptively to conquer motion speed variations, (4) SGAAL, which is used to generate optimized anchors to match shadow location and shape, and (5) OHEM, which is used to select typical difficult negative samples to improve background discrimination capacity. Finally, the quantitative and qualitative results reveal the state-ofthe-art detection performance of ShadowDeNet with a 66.01% best *f* 1 accuracy. Specifically, it is superior to the experimental baseline Faster R-CNN by a 9.00% *f* 1 accuracy. It is also superior to the existing best model by a 4.96% *f* 1 accuracy. Moreover, the detection speed sacrifice is very slight. Last but not least, we also conduct extensive ablation studies on the public SNL video SAR data to confirm the effectiveness of each characteristic. We also use ShadowDeNet to detect shadows on another one video SAR data, and the results reveal its universal effectiveness and excellent migration ability. In short, ShadowDeNet can provide high-quality predetection results for subsequent trackers, of grea<sup>t</sup> value.

Our future work is as follows:


**Author Contributions:** Conceptualization, T.Z.; methodology, T.Z.; software, J.B.; validation, X.X.; formal analysis, J.B.; investigation, J.B.; resources, T.Z.; data curation, J.B.; writing—original draft preparation, T.Z. and J.B.; writing—review and editing, T.Z., J.B. and X.Z.; visualization, X.Z.; supervision, X.Z.; project administration, T.Z.; funding acquisition, X.Z. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported in part by the National Key R&D Program of China under Grant 2017YFB0502700 and in part by the National Natural Science Foundation of China under Grants 61571099, 61501098, and 61671113.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** No new data were created or analyzed in this study. Data sharing is not applicable to this article. The public video SAR data provided by Sandia National Laboratories (SNL) is available from https://www.sandia.gov/app/uploads/sites/124/2021/0 8/eubankgateandtrafficvideosar.mp4 (accessed on 20 November 2021) to download for scientific research. China Aerospace Science and Industry Corporation (CASIC) 23 research institute video SAR data is not public because of copyright restrictions.

**Acknowledgments:** The authors would like to thank the editors and anonymous reviewers for their valuable comments that greatly improve our manuscript. The authors would like to thank Sandia National Laboratories (SNL) and China Aerospace Science and Industry Corporation (CASIC) 23 research institute for providing the video SAR data.

**Conflicts of Interest:** The authors declare no conflict of interest.
