Automatic Deployment of Convolutional Neural Networks on FPGA for Spaceborne Remote Sensing Application

In recent years, convolutional neural network (CNN)-based algorithms have been widely used in remote sensing image processing and show tremendous performance a variety of application fields. However, large amounts data intensive computations make the deployment CNN-based challenging problem, especially for spaceborne scenario where resources power consumption are limited. To tackle this paper proposes an automatic CNN solution on resource-limited field-programmable gate arrays (FPGAs) applications. Firstly, series hardware-oriented optimization methods proposed to reduce complexity CNNs. Secondly, hardware accelerator is designed. accelerator, reconfigurable engine array with efficient computation architecture accelerate algorithms. Thirdly, bridge optimized CNNs compilation toolchain introduced into solution. Through conversion from models instructions, various networks can be deployed real-time. Finally, we improved VGG16 YOLOv2 Xilinx AC701 evaluate effectiveness The experiments that only 3.407 W 94 DSP consumption, our achieves 23.06 giga operations per second (GOPS) throughput 22.17 GOPS YOLOv2. Compared related works, efficiency by 1.3–2.7×.