A Real-time Ship Detection System for Large-Scale Optical Remote Sensing Image on Micro-Nano Satellite

Ship detection in optical remote sensing images is of great importance for maritime traffic management. At present, the advanced optical system on the micro-nano satellites has been able to generate large-scale remote sensing images of gigabits data in real-time. However, the image processing system cannot manage such a huge amount of data and finish the ship detection task within the time constraint. To address this issue, this article contributes a large-scale remote sensing image processing system for real-time ship detection on micro-nano satellite. By introducing the heterogeneous System-On-Chip (SoC) and Field Programmable Gate Array (FPGA) processors to the hardware design with distributed memory access architecture, the high throughput requirements of large-scale image acquisition and processing strategies including sliding window crop, grayscale variance calculation and convolutional neural networks are successfully satisfied. The implementation and evaluation of the proposed system demonstrate its effectiveness in real-time ship detection in large-scale remote sensing images. With the large-scale remote sensing image as the input, the designed system achieves up to 3. 2Gbps of image data throughput for ship detection in real-time.