基于深度神经网络的高效人脸检测算法设计与实现  

Design and Implementation of an Efficient Face Detection Algorithm Based on Deep Neural Networks

作　　者：张佳颖 李爱军[2] Zhang Jiaying;Li Aijun(School of Software,Yunnan University,Kunming Yunnan 650091,China;Xi’an Aeronautics Computing Technology Research Institute,AVIC,Xi’an Shaanxi 710068,China)

机构地区：[1]云南大学,云南昆明650091 [2]西安航空计算技术研究所,陕西西安710068

出　　处：《山西电子技术》2025年第1期41-44,共4页Shanxi Electronic Technology

摘　　要：为解决大数据背景下,训练和测试数据日益庞杂,神经网络模型规模不断扩大,权重参数数量猛增,网络结构日益复杂,最终导致模型效率下降等问题,本文提出了一种人脸检测算法,充分利用了稀疏的小规模多尺度卷积核的优势来提高模型效率,并且加入了1x1的过滤器来降低权重维度。经实验证明,与现有模型相比,算法在不损失精度的情况下,权重数量大幅减少。To address the challenges posed by the increasing complexity of training and testing data in the context of big data,where neural network models expand in size,the number of weight parameters surges,and network structures become more intricate that resulting in a decline in model efficiency,this paper proposes a facial detection algorithm.The algorithm makes effective use of the advantages of sparse small-scalemulti-scale convolutional kernels to enhance model efficiency and incorporates 1x1 filters to reduce the dimensionality of weights.Experimental results demonstrate that,compared to existing models,the algorithm significantly reduces the number of weights without compromising accuracy.

关 键 词：机器学习 深度学习 深度神经网络 人脸检测 模型压缩 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]