Due to the limited power of underwater intelligent device

acquired image data cannot be processed in real time.Consider characteristics of low power consumption

high computing power and high flexibility of field programmable gate array（ FPGA）.FPGA-based parallelism and pipeline technology

use powerful image processing capability of convolutional neural network（ CNN）

a low power consumption image real-time recognition system is designed.The system is verified uses underwater video images from three different water regions.Experimental results show that the system achieves the same precision as the PC workstation

and realize processing of 1 920 ×1 080 resolution and 25 FPS frame rate image to meet the needs of underwater recognition tasks.