基于卷积神经网络的图像目标检测优化算法研究  

Research on Image Target Detection Optimization Algorithm Based on Convolutional Neural Network

作　　者：戴振民 孙宽宏 陈浩 DAI Zhenmin;SUN Kuanhong;CHEN Hao(China State Shipbuiding Corporation Limited No.723 Research Institute,Yangzhou Jiangsu 225001,China)

机构地区：[1]中国船舶集团有限公司第七二三研究所,江苏扬州225001

出　　处：《信息与电脑》2023年第18期124-127,共4页Information & Computer

摘　　要：文章提出基于卷积神经网络的图像目标检测优化算法。以最小代价函数作为卷积神经网络的学习目标,构建图像目标检测的深度残差网络模型,通过感兴趣区域网络获取图像目标候选区域,获取图像目标的分类层输出和预测层输出。实验结果表明:该算法检测到图像精度在97.36%以上。This paper proposes an optimization algorithm for image target detection based on convolutional neural network.Taking the minimum cost function as the learning target of convolutional neural network,the depth residual network model of image target detection is constructed,and the candidate regions of image target are obtained through the region of interest network,and the output of classification layer and prediction layer of image target are obtained.The experimental results show that the accuracy of the image detected by this algorithm is above 97.36%.

关 键 词：卷积神经网络 目标检测优化 监督学习 深度残差网络模型 卷积核参数 损失函数 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]