The accuracy of current deep learning algorithms has certainly increased. However, deploying deep learning networks on edge devices with limited resources is challenging due to their inherent depth and high parameter count. Here, we proposed an improved YOLO model based on an attention mechanism and receptive field (RFA-YOLO) model, applying the MobileNeXt network as the backbone to reduce parameters and complexity, adopting the Receptive Field Block (RFB) and Efficient Channel Attention (ECA) modules to improve the detection accuracy of multi-scale and small objects. Meanwhile, an FPGA-based model deployment solution was proposed to implement parallel acceleration and low-power deployment of the detection algorithm model, which achieved real-time object detection for optical remote sensing images. We implement the proposed DPU and Vitis AI-based object detection algorithms with FPGA deployment to achieve low power consumption and real-time performance requirements. Experimental results on DIOR dataset demonstrate the effectiveness and superiority of our RFA-YOLO model for object detection algorithms. Moreover, to evaluate the performance of the proposed hardware implementation, it was implemented on a Xilinx ZCU104 board. Results of the experiments for hardware and software simulation show that our DPU-based hardware implementation are more power efficient than central processing units (CPUs) and graphics processing units (GPUs), and have the potential to be applied to onboard processing systems with limited resources and power consumption.

Real-time detection of satellite remote sensing images is one of the key technologies in the field of remote sensing, which requires not only high-efficiency algorithms, but also low-power and high-performance hardware deployment platforms. At present, the image processing hardware acceleration platform mainly uses an image processing unit (GPU), but the GPU has the problem of large power consumption, and it is difficult to apply to micro-nano satellites and other devices with limited volume, weight, computing power, and power consumption. At the same time, the deep learning algorithm model has the problem of too many parameters, and it is difficult to directly deploy it on embedded devices. In order to solve the above problems, we propose a YOLOv4-MobileNetv3 field programmable gate array (FPGA) deployment scheme based on channel layer pruning. Experiments show that the acceleration strategy proposed by us can reduce the number of model parameters by 91.11%, and on the aerial remote sensing dataset DIOR, the average accuracy of the design scheme in this paper reaches 82.61%, the FPS reaches 48.14, and the average power consumption is 7.2 W, which is 317.88% FPS higher than the CPU and reduces the power consumption by 81.91%. Compared to the GPU, it reduces power consumption by 91.85% and improves FPS by 8.50%. Compared with CPUs and GPUs, our proposed lightweight algorithm model is more energy-efficient and more real-time, and is suitable for application in spaceborne remote sensing image processing systems.