一种改进残差深度网络的多目标分类技术  

A Multi-objective Classification Technique Based on Improved Residual Deep Network

作　　者：陈超 吴斌[1] CHEN Chao;WU Bin(School of Information Engineering,Southwest University of Science and Technology,Mianyang 621010,China;Numerical Simulation Key Laboratory of Sichuan University,Neijiang 641000,China)

机构地区：[1]西南科技大学信息工程学院,四川绵阳621010 [2]四川省高等学校数值仿真重点实验室,四川内江641000

出　　处：《计算机测量与控制》2023年第7期199-206,共8页Computer Measurement &Control

基　　金：国家自然科学青年基金项目(11502121);四川省项目应用基础研究计划(2021JY0108);四川省科技研究计划项目(njfh20-003)。

摘　　要：由于受场景、视角、光照、尺度变化以及局部变形等因素的影响,对重叠目标、拥挤目标、小目标的识别精度较低,提出了一种改进多支路的残差深度卷积神经网络来提高多目标识别的准确度;在第一个卷积残差块layer1后保留恒等映射的同时,增加一个1×1的短接分支尽可能多的保留原始特征;再平行嵌入一个修改激活函数ReLU6的空间_通道注意力机制模块(CBAM);融合以上3个特征图;融合后的特征层着重关注空间和通道中比较显著的信息,从而增强特征图的特征表达能力,以至于卷积神经网络(CNN)获得更多的判别特征,从而大大提高物体识别精度;在FashionMNIST和Cifar10两个数据集的对比性实验显示改进的resnet50算法是准确性-速度较为折中的目标识别模型。Due to the influences of scene,visual angle,illumination,scale change and local deformation,the recognition accuracies of overlapping target,crowded target and small target are low,an improved multi-branch Resnet50 convolutional neural network is proposed to improve the accuracy of multi-objective recognition.While retaining the constant maps after the first convolutional residual block Layer1;The short branch of one by one is added to preserve as many original features as possible;A space_channel attention mechanism module(CBAM)is embed to modify the activation function ReLU6 in parallel;The last three feature graphs are fused.The fused feature layer focuses on the significant information in space and channels,thus enhancing the feature expression ability in the feature graphs,so that the convolutional neural network(CNN)can obtain more discriminant features,thus greatly improving the accuracy of object recognition.The comparative experiment on the FashionMNIST and Cifar10 data sets shows that the improved Resnet50 algorithm is suitable for a target recognition model with a medium accuracy and speed.

关 键 词：残差深度卷积神经网络 短接分支 CBAM 激活函数ReLU6 多目标分类 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置]