|Paper title||Object detection methods for dark vessel detection and classification using SAR imagery|
|Form of presentation||Poster|
Illegal, unreported, and unregulated fishing vessels pose a huge risk to the sustainability of fishing stocks, marine ecosystems, and also plays a part in heightening political tensions around the globe, (Long et al., 2020) both in national and international waters. Annual global losses have an estimated value between US$10 billion to $23.5 billion, and this figure is even higher when impacts across the value chain and the ecosystems are taken into account. Illegal fishing is often organized internationally across multiple jurisdictions, and as a consequence the economic value from these catches leaves the local communities where it would otherwise belong.
The identification of illegal fishing vessels is a hard problem, that in the past required either data from and Automatic Identification System (AIS) (Longépé et al., 2018), or short range methods such as acoustic telemetry (Tickler et al., 2019). For vessel presence detection, SAR imagery has proven to be a reliable method when combined with traditional computer vision algorithms (Touzi et al., 2004; Tello et al., 2005), and more recently neural networks (Chang et al., 2019; Li et al., 2017). Its big advantage over other methods is that it is applicable in all weather conditions and does not require cooperation from the ships. The biggest hurdle in developing effective identification of illegal vessels was the lack of high resolution, reliably labeled data, as modern neural network based methods rely on the abundance of data for dependable predictions.
The newly released xView3 (xView3 Dark Vessel Detection Challenge, 2021) dataset and the complimentary challenge provides an excellent testing ground for adapting neural network based object detection methods to SAR based dark vessel detection. The open-source dataset contains over 1000 scenes of maritime regions of interest, with VV and VH SAR data from the European Space Agency’s Sentinel-1 satellites, bathymetry, wind speed, wind direction, wind quality, land/ice masks, and with accompanying hand-corrected vessel labels.
Our goal is to find an accurate and practical detection method for dark vessel identification. In order to achieve this we adapt two popular object detection architectures, Faster R-CNN (Ren et al., 2015) and YOLOv3 (Redmon et al., 2018) to the the xView3 data, together with pre- and post-processing steps. The specific architectures are chosen so that the robust high performance of Faster R-CNN can serve as a baseline, while YOLOv3 is considered a good compromise between computational complexity and performance, and so is expected to improve practical usability in near real-time use cases.
Domain specific adaptations to the architecture (such as adapting augmentation methods to SAR data, adjusting anchor sizes, and resizing parts of the network to better accommodate the fewer input channels but smaller output predictions) are expected to show a significant increase in performance, based on preliminary results and past experiments. We perform both quantitative and qualitative evaluation of the outputs, and an ablation study to quantify the effectiveness of different parts of the processing pipelines.
Chang, Y.-L., Anagaw, A., Chang, L., Wang, Y. C., Hsiao, C.-Y., Lee, W.-H., 2019. Ship detection based on YOLOv2 for SAR imagery. Remote Sensing, 11(7), 786.
Li, J., Qu, C., Shao, J., 2017. Ship detection in sar images based on an improved faster r-cnn. 2017 SAR in Big Data Era: Models, Methods and Applications (BIGSARDATA), IEEE, 1–6.
Long, T., Widjaja, S., Wirajuda, H., Juwana, S., 2020. Approaches to combatting illegal, unreported and unregulated fishing. Nature Food, 1(7), 389–391.
Longépé, N., Hajduch, G., Ardianto, R., de Joux, R., Nhunfat, B.,´
Marzuki, M. I., Fablet, R., Hermawan, I., Germain, O., Subki, B. A. et al., 2018. Completing fishing monitoring with spaceborne Vessel Detection System (VDS) and Automatic Identification System (AIS) to assess illegal fishing in Indonesia. Marine pollution bulletin, 131, 33–39.
Redmon, J., Farhadi, A., 2018. Yolov3: An incremental improvement. arXiv preprint arXiv:1804.02767.
Ren, S., He, K., Girshick, R., Sun, J., 2015. Faster R-CNN: Towards real-time object detection with region proposal networks. Advances in neural information processing systems, 28, 91–99.
Tello, M., López-Martínez, C., Mallorqui, J. J., 2005. A novel algorithm for ship detection in SAR imagery based on the wavelet transform. IEEE Geoscience and remote sensing letters, 2(2), 201–205.
Tickler, D. M., Carlisle, A. B., Chapple, T. K., Curnick, D. J., Dale, J. J., Schallert, R. J., Block, B. A., 2019. Potential detection of illegal fishing by passive acoustic telemetry. Animal Biotelemetry, 7(1), 1–11.
Touzi, R., Charbonneau, F., Hawkins, R., Vachon, P., 2004. Ship detection and characterization using polarimetric SAR. Canadian Journal of Remote Sensing, 30(3), 552–559.
xView3 Dark Vessel Detection Challenge, 2021. https://iuu. xview.us/. Accessed: 2021-11-26.