Coupling Denoising to Detection for SAR Imagery

Abstract

Detecting objects in synthetic aperture radar (SAR) imagery has received much attention in recent years since SAR can operate in all-weather and day-and-night conditions. Due to the prosperity and development of convolutional neural networks (CNNs), many previous methodologies have been proposed for SAR object detection. In spite of the advance, existing detection networks still have limitations in boosting detection performance because of inherently noisy characteristics in SAR imagery; hence, separate preprocessing step such as denoising (despeckling) is required before utilizing the SAR images for deep learning. However, inappropriate denoising techniques might cause detailed information loss and even proper denoising methods does not always guarantee performance improvement. In this paper, we therefore propose a novel object detection framework that combines unsupervised denoising network into traditional two-stage detection network and leverages a strategy for fusing region proposals extracted from both raw SAR image and synthetically denoised SAR image. Extensive experiments validate the effectiveness of our framework on our own object detection datasets constructed with remote sensing images from TerraSAR-X and COSMO-SkyMed satellites. Extensive experiments validate the effectiveness of our framework on our own object detection datasets constructed with remote sensing images from TerraSAR-X and COSMO-SkyMed satellites. The proposed framework shows better performances when we compared the model with using only noisy SAR images and only denoised SAR images after despeckling under multiple backbone networks.

CCDN-DETR: A Detection Transformer Based on Constrained Contrast Denoising for Multi-Class Synthetic Aperture Radar Object Detection

“…Representative methods include the integration of attention mechanisms (SE [ 29 ]) into the feature extraction part of the backbone network by Hou et al [ 18 ], providing additional attention to the detection regions; L-YOLO [ 30 ] achieves a lightweight, efficient network structure by simplifying convolution operations and proposes a k-means algorithm for clustering anchor boxes; Miao T et al [ 8 ] improved model accuracy by adjusting the backbone network and applying channel and spatial attention. Additionally, integrating SAR-based prior knowledge [ 31 , 32 ] with deep learning helps to provide more stable detection results. Moreover, some specifically optimized loss functions [ 12 , 13 ] have also gained attention and have been demonstrated to perform well in SAR object detection tasks.…”

Special Issue on Image Simulation in Remote Sensing

“…Despite advances in SAR image processing, existing detection technologies still have limitations in boosting detection performance because of their inherently noisy characteristics. Sujin Shin and collaborators [6], in their contribution, propose a novel object detection framework that combines an unsupervised denoising network and a traditional detection network to leverage a strategy for fusing region proposals extracted from both raw SAR images and synthetically denoised SAR images. Changno Lee and Jaehong Oh [7], in their paper, propose sensor level mosaicking to generate a seamless image product with geometric accuracy to meet mapping requirements.…”

A Mixed-Scale Self-Distillation Network for Accurate Ship Detection in SAR Images

“…[30] gradually integrates the semantic strong features and low-level high-resolution features to mitigate false alarms. [31] combines candidate proposals from the raw SAR image and the synthetically denoised SAR image to reduce the impact of noise on ship detection. [32] adopts focal loss to adjust the weights of hard negative and simple samples.…”