引用本文:赵静,王弦,王奔,等.基于神经网络的多类别目标识别[J].控制与决策,2020,35(8):2037-2041
【打印本页】   【HTML】   【下载PDF全文】   查看/发表评论  【EndNote】   【RefMan】   【BibTex】
←前一篇|后一篇→ 过刊浏览    高级检索
本文已被:浏览次   下载 本文二维码信息
码上扫一扫!
分享到: 微信 更多
基于神经网络的多类别目标识别
赵静{1,2, 王弦1, 王奔1, 蒋国平1,2, 谢非3, 徐丰羽1,2
(1. 南京邮电大学自动化学院、人工智能学院,南京210023;2. 江苏省物联网智能机器人工程实验室, 南京210023;3. 南京师范大学电气与自动化工程学院,南京210023)
摘要:
随着智能化时代的到来,深度学习在图像处理领域的应用越来越广泛,为复杂的图像处理带来了新的解决方法.但是,深度学习在带来高准确推理结果的同时,往往会造成运算量、推理时间以及处理器内存的增加,受限于数据的缺失以及对高性能处理器的依赖.目前市场上有关深度学习技术的相关产品还未被广泛的应用.基于上述问题,提出一种具有广泛应用前景的基于深度学习的多类别目标识别方案,并在国产龙芯派平台下进行测试验证.首先,获取待处理图像数据集,进而在计算机平台下搭建并训练神经网络模型,利用得到的训练参数在龙芯派平台下建立优化后的神经网络结构,并对目标图像进行处理及识别,最后在用户界面显示目标图像所属类别.本系统利用深度学习能够自动从大数据中学习特征的优势,在龙芯派平台下实现上千类生活中常见对象的自动分类,识别准确率高达$96%$以上,识别速度在3s以内,且该方案具有优秀的可扩展性.
关键词:  龙芯派  图像识别  深度学习  AlexNet网络模型  自动分类
DOI:10.13195/j.kzyjc.2019.0577
分类号:TP273
基金项目:国家自然科学基金项目(61601228);江苏省自然科学基金项目(BK20161021);江苏省六大人才高峰项目(JY-081).
Multi-category target recognition based on neural network
ZHAO Jing1,2,WANG Xian1, WANG Ben1,JIANG Guo-ping1,2,XIE Fei3,XU Feng-yu1,2
(1. College of Automation & College of Artificial Intelligence,Nanjing University of Posts and Telecommunication,Nanjing 210023,China;2. Jiang Su Engineering Laboratory for Internet of Things and Intelligent Robotics, Nanjing 210023,China;3. College of Electrical and Automation Engineering,Nanjing Normal University,Nanjing 210023,China)
Abstract:
With the coming of intelligent era, deep learning algorithm has been applied more and more widely in the field of image processing, which brings new solutions to complex image processing. However, while deep learning can produce highly accurate inference results, it often leads to computational complexity, the increase of inference time and processor memory. Due to the lack of data and reliance on high-performance processors, deep learning related products have not been widely used in real life scenarios. Based on the above questions, this paper proposes a multi-category image recognition scheme based on deep learning with broad application prospects and carries out test verification under the Loongson platform. Firstly, the image data set of the category to be identified is obtained, and then the neural network model is built and trained under the computer platform. The optimized neural network structure is established under the Loongson platform using the obtained training parameters, and then the target image is processed and recognized. Finally, the category of the target image is displayed in the user interface. The system utilizes the advantages of deep learning to automatically learn big data features and realizes automatic classification of thousands of common objects in life under the Loongson platform. The recognition accuracy can reach more than 96%, and the recognition time is limited to 3s or less. The scheme also demonstrates excellent scalability.
Key words:  Loongson Pi  image recognition  deep learning  AlexNet  automatic classification

用微信扫一扫

用微信扫一扫