V.A. Pavlov1, A.A. Belov2

1,2 Peter the Great St. Petersburg Polytechnic University (St. Petersburg, Russia)

1 pavlov_va@spbstu.ru; 2 belov@spbstu.ru

Currently, neural network (NN) approaches have become the dominant tool for detecting objects in images, demonstrating high efficiency and accuracy. However, their successful application usually depends on the availability of a large amount of annotated training data, which is necessary for training NN models. This requirement creates significant difficulties, especially in areas where the amount of such data is limited or difficult to access. In the case of optical images, there are many annotated datasets that allow neural networks to be trained to solve various computer vision problems. However, in the field of radar image processing, the situation is much more complicated: there is much less annotated data available, which significantly limits the possibilities of training and using neural network models. Optical and radar images have fundamentally different physical nature. Optical images are formed based on reflected natural light, which makes it easy to distinguish colors and textures of objects. At the same time, radar images are based on the reflection of radio waves emitted by a radar transmitter, which makes them insensitive to color and independent of lighting and weather conditions. However, at sufficiently high spatial resolutions (on the order of 1 meter and better), the images of objects on optical and radar images begin to show structural similarities. This similarity is evident in the size, shape of the contour, the relative position and orientation of the corners and other elements of an object. This structural similarity opens up prospects for using optical images in training neural networks designed to work with radar data. In this paper, we investigated the possibility of using neural networks trained exclusively on optical images for automated ship detection on radar images. The results obtained were positive, which confirms the potential of using optical images both for direct neural network training and for augmentation of existing radar data sets. Special attention in the study was paid to the problem of speckle noise, which is inevitably present in radar images and significantly complicates the process of detecting objects. Speckle noise is a strong multiplicative noise that distorts images and makes them difficult to analyze. Various filters have been applied to reduce this noise, including traditional adaptive filters such as Lee and Frost filters, as well as neural network filters trained on optical images with artificially added noise. The use of filters to suppress speckle noise resulted in a significant improvement in detection quality, reducing the number of false detections to about 18%. At the same time, neural network filters have demonstrated better performance compared to traditional methods, which highlights the promise of using neural networks not only for object detection, but also for image preprocessing. Thus, this study demonstrates that using optical images to train neural networks can significantly expand the possibilities of automated object detection in radar images. This opens up new horizons for the development of computer vision techniques in conditions of limited availability of annotated radar data and highlights the importance of developing effective image preprocessing techniques to improve detection quality

Pavlov V.A., Belov A.A. On using optical images in training of neural networks for radar image processing. Radiotekhnika. 2025.
V. 89. № 3. P. 98−108. DOI: https://doi.org/10.18127/j00338486-202503-09 (In Russian)

1. https://www.iceye.com/.
2. Vinogradova N.S., Dorosinskij L.G. Raspoznavanie radiolokacionnyh izobrazhenij, formiruemyh radiolokacionnymi sistemami s sintezirovannoj aperturoj. Ural Radio Engineering Journal. 2021. T. 5. № 3. S. 258–271 (in Russian).
3. Pavlov V.A., Belov A.A., Tuzova A.A. Implementation of synthetic aperture radar processing algorithms on the Jetson TX1 platform. 2019 IEEE International Conference on Electrical Engineering and Photonics (EExPolytech). St. Petersburg. Russia. 2019. Р. 90-93.
4. Pavlov V.A., Belov A.A., Tuzova A.A. Investigation of the influence of speckle noise on the accuracy of object detection by convolutional neural networks. 2021 International Conference on Electrical Engineering and Photonics (EExPolytech). St. Petersburg, Russian Federation. 2021. Р. 47–50.
5. Pavlov V.A., Belov A.A., Volvenko S.V., Rashich A.V. Application of convolutional neural networks trained on optical images for object detection in radar images. Computer Optics. 2024. V. 48. № 2. P. 253–259.
6. Zhang L., Yin H. Research on ship detection method of optical remote sensing image based on deep learning. 2022 International Conference on Sensing, Measurement Data Analytics in the era of Artificial Intelligence (ICSMD). 2022. P. 1–6.
7. Xie G., Li J., Xiao F., Li T. Optical remote sensing image ship detection method for improving EfficientDet. 2021 IEEE 3rd International Conference on Frontiers Technology of Information and Computer (ICFTIC). 2021. P. 340–343.
8. Qin C., Wang X., Li G., He Y. An improved attention-guided network for arbitrary-oriented ship detection in optical remote sensing images. IEEE Geoscience and Remote Sensing Letters. 2022. V. 19. P. 1–5.
9. Zhuang Y., Liu Y., Zhang T., Chen H. Contour modeling arbitrary-oriented ship detection from very high-resolution optical remote sensing imagery. IEEE Geoscience and Remote Sensing Letters. 2023. V. 20. P. 1–5.
10. Jianxin G., Zhen W., Shanwen Z. Multi-scale ship detection in SAR images based on multiple attention cascade convolutional neural networks. 2020 International Conference on Virtual Reality and Intelligent Systems (ICVRIS). 2020. P. 438–441.
11. Hou Z., Cui Z., Cao Z., Liu N. An integrated method of ship detection and recognition in sar images based on deep learning. 2020 IEEE International Geoscience and Remote Sensing Symposium. 2020. P. 1225–1228.
12. Zhao X., Zhang B., Tian Z., Xu C., Wu F., Sun C. An anchor-free method for arbitrary-oriented ship detection in SAR images. 2021 SAR in Big Data Era (BIGSARDATA). 2021. P. 1–4.
13. Fu X., Wang Z. Fast ship detection method for SAR images in the inshore region. 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS. 2021. P. 3569–3572.
14. Wang X., Zhou H., Chen Z., Bai J., Ren J., Shi J. Few-shot SAR ship image detection using two-stage cross-domain transfer learning. IGARSS 2022 - 2022 IEEE International Geoscience and Remote Sensing Symposium. 2022. P. 2195–2198,
15. Wang C., Bochkovskiy A., Liao H. YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors. 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2023. P. 7464–7475.
16. Liu L., Ouyang W., Wang X., et al. Deep Learning for Generic Object Detection: A Survey. Int. J. Comput. Vis. 2020. № 128.
    P. 261–318.
17. Wei S., Zeng X., Qu Q., Wang M., Su H., Shi J. HRSID: A high-resolution SAR images dataset for ship detection and instance segmentation. IEEE Access. 2020. V. 8. P. 120234–120254.
18. Wang Z., Bovik A.C., Sheikh H.R., Simoncelli E.P. Image quality assessment: from error visibility to structural similarity. IEEE Transactions on Image Processing. 2004. V. 13. № 4. P. 600-612.
19. Xue W., Zhang L., Mou X., Bovik A.C. Gradient Magnitude Similarity Deviation: A Highly Efficient Perceptual Image Quality Index. IEEE Transactions on Image Processing. 2014. V. 23. № 2. P. 684-695.
20. Pavlov V.А., Belov A.А., Nguen V.T., Jovanovski N., Ovsyannikova A.S. Comparison of neural networks for suppression of multiplicative noise in images. Computer Optics. 2024. V. 48. №. 3. P. 425‒431.
21. Gonzalez R.C., Woods R.E. Digital image processing. upper saddle river. New Jersey: Pearson Education, Inc. 2008.
22. Lee J.S. Digital image enhancement and noise filtering by using local statistics. IEEE Transactions on Pattern Analysis and Machine Intelligence. 1980. V. PAMI-2. № 2. P. 165–168.
23. Frost V.S., Stiles J.A., Shanmugan K.S., Holtzman J.C. A model for radar images and its application to adaptive digital filtering of multiplicative noise. IEEE Transactions on Pattern Analysis and Machine Intelligence. 1982. V. 4. № 2. P. 157–166.
24. Perona P., Malik J. Scale-space and edge detection using anisotropic diffusion. IEEE Transactions on Pattern Analysis and Machine Intelligence. 1990. V. 12. № 7. P. 629–639.