Current Issue Cover

肖振久, 林渤翰, 曲海成(辽宁工程技术大学)

摘 要
目的 针对合成孔径雷达(synthetic aperture radar, SAR)图像噪声大, 成像特征不明显, 尤其在复杂场景更容易出现目标误检和漏检的问题,提出了一种融合多重机制的SAR舰船检测方法,用于提高SAR舰船检测的精度。方法 在预处理部分,设计了U-net Denoising模块,通过调整噪声方差参数L的范围来抑制相干斑噪声对图像的干扰。在YOLOv7主干网络构建MLAN_SC结构,加入通道注意力机制 (selective kernel,SK)至下采样阶段,增强关键信息提取能力和特征表达能力。为解决MP结构中上下分支特征不平衡的问题,改善误检情况,融入上下文信息提取模块(contextual transformer block ,COT),利用卷积提取上下文信息,将局部信息和全局信息结合起来,使图像的特征更有效的提取出来。在头部引入SPD卷积(space-to-depth convolution,SPD-Conv),增强小目标的检测能力。用WIoU损失函数替换CIoU损失函数,运用动态聚焦机制,在复杂图片上加强对目标定位能力。结果 在SSDD(SAR ship detection dataset)数据集和HRSID (high-resolution SAR images dataset)数据集上进行了实验对比,结果表明,改进后的方法相比于YOLOv7,AP可达到99.25%和89.73%,分别提升了4.38%和2.57%,准确率和召回率为98.41%,93.24%和94.79%,81.83%,优于现有的主流方法。结论 本文通过融合多重机制改进YOLOv7方法,提升了对目标的定位能力,显著改善了SAR舰船检测中复杂舰船的误检和漏检情况,进一步提高了SAR舰船检测精度。
SAR ship detection method optimized by multiple mechanisms

Xiao Zhenjiu, Lin Bohan, Qu Haicheng(Liaoning Technical University)

Objective In recent years, the efficacy of synthetic aperture radar (SAR) has been increasingly recognized in the field of maritime surveillance and vessel detection, thanks to its remarkable all-weather and day-night imaging capabilities. The ability of SAR systems to penetrate through clouds and fog has enabled high-quality imaging of the sea surface in various weather conditions. Nevertheless, SAR imaging is often hindered by excessive noise and unclear imaging features, which can lead to erroneous detections in complex maritime environments. In response to these challenges, this study presents an innovative approach that combines state-of-the-art deep learning and computer vision techniques to improve the accuracy of SAR ship detection. By incorporating several critical enhancements to the YOLOv7 algorithm, the proposed method aims to enhance the capability of SAR systems to accurately identify and track vessels on the sea surface. The potential of this method is significant for maritime security and surveillance systems, as the accurate and reliable detection of vessels is paramount to ensuring the safety and security of shipping lanes and ports worldwide. Method The present study proposes a novel method that offers significant improvements to the YOLOv7 algorithm for SAR ship detection. Specifically, a U-net Denoising module is designed in the preprocessing stage to suppress coherent speckle noise interference by leveraging deep learning techniques to model the range of parameter L. Moreover, building MLAN_SC structure in YOLOv7 backbone network. To enhance key information extraction and deep feature expression abilities, the proposed method also introduces SK (selective kernel) attention mechanism to improve the false detection rate. COT (contextual transformer) block is integrated into the backbone network to solve the problem of unbalanced upper and lower branch features in the MP structure and improve the false detection situation. COT block using convolutional operations and combine local and global information for more effective feature extraction. Additionally, SPD convolution is incorporated into the detection head to enhance small object detection capabilities. The study further replaces the CIoU loss function with the WIoU loss function and applies a dynamic focusing mechanism to enhance target localization performance on complex images. Result We employed the network weights of ImageNet for training our model. The experimental data utilized in this study was selected from the SAR ship detection dataset (SSDD). The SSDD contains a total of 1160 SAR images and 2456 ship targets. The dataset primarily includes data from the RadarSat-2, Sentinel-1, and TerraSAR-X sensors. The target area is cropped to 500 pixels in four polarization modes: HH (horizontal), VV (vertical), HV (cross), and VH (cross), and labeled in the PASCAL VOC format. Our deep learning framework is implemented in Python, and the input image size is adjusted to 640 ? 640 during training. The momentum parameter is set to 0.93, and the starting learning rate is set to 0.001. We employ the cosine annealing method to attenuate the learning rate. The NVIDIA GeForce RTX3060 GPU device is used to accelerate the stochastic gradient descent (SGD) learning process and iterate the model. Multiple ablative experiments were conducted to validate the effectiveness of the proposed module improvements on the SSDD dataset, using the original YOLOv7 network as a baseline for comparison. The baseline algorithm achieved an accuracy of 94.87%, while the addition of the denoising module resulted in a more precise extraction of targets in complex backgrounds, leading to an improvement in accuracy. Incorporating the SK and SOCA attention mechanism modules significantly impacted SAR ship detection, enhancing the representation of deep-level features and extraction of key information, reducing false positives, and further improving detection accuracy. Integration of the SPD-Conv module and WIoU loss function helped the model focus on targets in complex scenes, improving localization performance and enhancing the detection ability of small dense targets in deep sea. The proposed method achieved the best AP@0.5 (99.25%) and AP@0.5:0.95 (71.21%) on the SSDD dataset, respectively, 4.38% and 9.19% higher than the YOLOv7 baseline, demonstrating the effectiveness of the proposed module improvements. Comparative experiments were conducted with YOLOv7 and other popular deep learning-based object detection algorithms such as SSD, Faster R-CNN, RetinaNet, CenterNet, FENDet, and Tan et al, in terms of accuracy, recall, average precision, and time. Results showed that the proposed method had a recall rate 16.18% higher than that of Faster R-CNN and an accuracy rate 14.45% higher than that of RetinaNet. Furthermore, the proposed method exhibited high performance in handling missed and false positives. The PR curve comparison indicated that the proposed algorithm had excellent detection performance and a stable PR curve. The detection results of different algorithms on the SSDD dataset were presented, and while several methods effectively detected ship targets, the proposed method achieved the highest accuracy in ship detection and demonstrated superior performance in handling missed and false positives. Overall, the proposed algorithm has high feasibility and practicality in SAR ship detection. Conclusion In this paper, we propose an optimized version of the YOLOv7 algorithm for improving the accuracy of SAR ship detection. Our approach integrates multiple mechanisms to enhance information extraction and overcome challenges associated with noisy and complex images. Specifically, we introduce a noise removal module, which effectively suppresses noise interference. The integration of attention mechanism and self-attention mechanism strengthens feature extraction and enhances the discriminative learning ability of deep features. Additionally, we incorporate an SPD convolution and optimize the loss function to improve target location ability, resulting in significant improvements in false detection and missed detection rates for ships with complex backgrounds and dense small targets in the near-shore SAR ship detection.