基于B2DPCA与神经网络的MHT检测算法  

Moving Human Object Detection Algorithm Based on b2dpca and Neural Network

作　　者：夏百战[1] 骆昊[1] XIA Bai-zhan;LUO Hao(University of Electronic Science and Technology of China,Zhongshan Institute,Zhongshan Guangdong 710119,China)

机构地区：[1]电子科技大学中山学院,广东中山710119

出　　处：《计算机仿真》2022年第5期348-351,374,共5页Computer Simulation

基　　金：2019年度广东省科技厅重点领域研发计划项目(2019B090910001);2020年度中山市科技局社会公益与基础研究项目(2020KF258-39)。

摘　　要：随着信息化时代地快速发展,目标识别成为人工智能领域的重要组成部分,大量的视频图像数据给人工智能识别带来了极大地挑战,于是提出基于B2DPCA与神经网络的移动人体目标检测算法。先通过剩余网络层中较大的学习率对样本集进行训练、较小的学习率对神经网络进行调整,实现网络模型图像的特征保留,当识别的模糊图像与清晰图像的特征相近时,通过softmax进行辨别提高模糊图像的辨别率。通过2个投射矩阵对人体目标图像矩阵进行特征描述,并采用监督方法对投影矩阵进行学习,增强图像特征矩阵的分类性能,通过迭代求解最终得出最佳的网络参数。建立样本库,将调整量通过反向传播反馈给神经网络的权重,完成网络训练,直到迭代次数达到训练结束要求为止。实验结果表明,基于B2DPCA与神经网络的移动人体目标检测算法对人体目标具有特征保留性,能够有效提高平均识别率和分类识别效果。With the rapid development of the information age,target recognition has become an important part of the field of artificial intelligence.A large number of video image data has brought great challenges to artificial intelligence recognition.Therefore,a moving human target detection algorithm based on b2 dpca and neural network is proposed.Firstly,the sample set was trained by the larger learning rate in the remaining network layer,and the neural network was adjusted by the smaller learning rate,so as to achieve the feature preservation of the network model image.When the features of the recognized fuzzy image were similar to those of the clear image,the recognition of the fuzzy image was carried out by softmax to improve the recognition rate of the fuzzy image.Then,the image matrix of human body was described by two projection matrices,and the projection matrix was studied by the supervision method,and the classification performance of the image feature matrix was enhanced.The optimal network parameters were obtained through iterative solution.Finally,the sample database was established,and the adjustment quantity was fed back to the weight of neural network through back propagation to complete network training until the number of iterations was the end of network training.The experimental results show that the moving human object detection algorithm based on b2 dpca and neural network has the feature retention for human objects,and can effectively improve the average recognition rate and classification recognition effect.

关 键 词：神经网络 投射矩阵 分类性能 网络参数 移动人体目标 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]