基于智能芯片的舰船目标智能识别系统设计及实现

Design and Implementation of Ship Target Intelligent Detection System Based on Neural Process Unit

传统的海面目标检测识别方法在复杂背景下存在目标检测率低、目标特征依赖人工设计等问题,很难满足实际应用的要求。本文以深度学习、智能芯片等技术为基础,针对可见光、遥感等多波段传感器成像,基于主流深度学习框架建立图像识别、目标检测网络模型,实现智能目标位置检测、目标分类及关键部位识别,采用通用智能芯片NPU(Neural Process Unit)搭建完成嵌入式环境下可见光场景舰船目标智能识别系统,实现智能目标识别算法在硬件资源受限环境下的高效处理,初步验证智能技术在飞行器上应用的可行性。

Traditional sea surface target detection and identification methods have shortness in complex environment,such as low detection rate of the target and the dependence of target features on artificial design,which are difficult to meet the requirements of practical application.Based on deep learning and AI chip technology,aiming at multi-band sensor imaging such as visible light and remote sensing,image identification and target detection network model based on mainstream deep learning framework are established.This model realizes intelligent target location detection,target classification and key position identification.By adopting universal intelligent chip NPU(Neural Process Unit),the intelligent identification system of ship target in visible scene is built based on embedded environment,and the efficient processing of intelligent target recognition algorithm in hardware resource-constrained environment is realized.The feasibility of intelligent technology in aircraft application is preliminarily verified.