深度学习下动态目标识别算法优化仿真  

Optimization and Simulation of Dynamic Target Recognition Algorithm with Deep Learning

作　　者：朱木清 邹欢 ZHU Mu-qing;ZOU Huan(School of Computer Engineering,Guangzhou Huali College,Guangzhou Guangdong 511325,China;Faculty of Mechanical and Electrical Engineering,Yunnan Agricultural University,Kunming Yunnan 650000,China)

机构地区：[1]广州华立学院计算机工程学院,广东广州511325 [2]云南农业大学机电工程学院,云南昆明650000

出　　处：《计算机仿真》2023年第12期321-324,336,共5页Computer Simulation

摘　　要：与静态图像目标识别相比,动态目标识别过程易受复杂背景、未知的运动趋势、障碍物、光照强度等问题的干扰,为了解决上述问题,提出基于深度学习的动态目标识别算法优化研究。采用基于帧间差的高阶统计量算法分割出动态目标的背景区域,采用双向光流预测算法提取动态目标的特征,采用粒子群算法优化深度学习中的BP神经网络模型,将提取的特征输入到模型中,通过模型的训练输出符合要求的目标,完成动态目标的识别。实验结果表明,所提算法的特征提取能力强、识别时间短、识别效果好。Compared with static image object recognition,the dynamic target recognition process is susceptible to interference from complex backgrounds,unknown motion trends,obstacles,light intensity,and other issues.To ad⁃dress these issues,a deep learning based dynamic object recognition algorithm optimization study is proposed.At first,the high-order statistical algorithm based on interframe difference was adopted to segment the background area of dynamic targets.Then,the bidirectional optical flow prediction algorithm was used to extract the features of dynam⁃ic targets.Moreover,particle swarm optimization algorithm was used to optimize the BP neural network model in deep learning.Meanwhile,the extracted features were input into the model.Finally,the target meeting the requirements was output by training the model,thus completing the recognition for dynamic targets.Experimental results show that the proposed algorithm has strong ability in feature extraction,short recognition time,and good recognition effect.

关 键 词：背景分割 目标的二值模板 特征点提取 粒子群优化 神经网络 误差阈值 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]