Synthetic Aperture Radar (SAR) imagery is widely utilized in military and civilian applications. Recent deep learning advancements have led to improved ship detection algorithms, enhancing accuracy and speed over traditional Constant False-Alarm Rate (CFAR) methods. However, challenges remain with complex backgrounds and multi-scale ship targets amidst significant interference. This paper introduces a novel method that features a context-based decoupled head, leveraging positioning and semantic information, and incorporates shuffle attention to enhance feature map interpretation. Additionally, we propose a new loss function with a dynamic non-monotonic focus mechanism to tackle these issues. Experimental results on the HRSID and SAR-Ship-Dataset demonstrate that our approach significantly improves detection performance over the original YOLOv5 algorithm and other existing methods.

Recently, synthetic aperture radar (SAR) target detection algorithms based on Convolutional Neural Networks (CNN) have received increasing attention. However, the large amount of computation required burdens the real-time detection of SAR ship targets on resource-limited and power-constrained satellite-based platforms. In this paper, we propose a hardware-aware model speed-up method for single-stage SAR ship targets detection tasks, oriented towards the most widely used hardware for neural network computing—Graphic Processing Unit (GPU). We first analyze the process by which the task of detection is executed on GPUs and propose two strategies according to this process. Firstly, in order to speed up the execution of the model on a GPU, we propose SAR-aware model quantification to allow the original model to be stored and computed in a low-precision format. Next, to ensure the loss of accuracy is negligible after the acceleration and compression process, precision-aware scheduling is used to filter out layers that are not suitable for quantification and store and execute them in a high-precision mode. Trained on the dataset HRSID, the effectiveness of this model speed-up algorithm was demonstrated by compressing four different sizes of models (yolov5n, yolov5s, yolov5m, yolov5l). The experimental results show that the detection speeds of yolov5n, yolov5s, yolov5m, and yolov5l can reach 234.7785 fps, 212.8341 fps, 165.6523 fps, and 139.8758 fps on the NVIDIA AGX Xavier development board with negligible loss of accuracy, which is 1.230 times, 1.469 times, 1.955 times, and 2.448 times faster than the original before the use of this method, respectively.

The deep learning object detection algorithm has been widely applied in the field of synthetic aperture radar (SAR). By utilizing deep convolutional neural networks (CNNs) and other techniques, these algorithms can effectively identify and locate targets in SAR images, thereby improving the accuracy and efficiency of detection. In recent years, achieving real-time monitoring of regions has become a pressing need, leading to the direct completion of real-time SAR image target detection on airborne or satellite-borne real-time processing platforms. However, current GPU-based real-time processing platforms struggle to meet the power consumption requirements of airborne or satellite applications. To address this issue, a low-power, low-latency deep learning SAR object detection algorithm accelerator was designed in this study to enable real-time target detection on airborne and satellite SAR platforms. This accelerator proposes a Process Engine (PE) suitable for multidimensional convolution parallel computing, making full use of Field-Programmable Gate Array (FPGA) computing resources to reduce convolution computing time. Furthermore, a unique memory arrangement design based on this PE aims to enhance memory read/write efficiency while applying dataflow patterns suitable for FPGA computing to the accelerator to reduce computation latency. Our experimental results demonstrate that deploying the SAR object detection algorithm based on Yolov5s on this accelerator design, mounted on a Virtex 7 690t chip, consumes only 7 watts of dynamic power, achieving the capability to detect 52.19 512 × 512-sized SAR images per second.