大规模互联网图像自动识别技术研究

发布时间：2018-05-25 04:33

本文选题：图像识别 + 卷积神经网络　；参考：《浙江大学》2017年硕士论文

【摘要】：随着深度学习的迅猛发展,计算机视觉在交通、医疗、安防、广告、个人图像视频分析等领域用,其中作为计算机视觉基础的图像识别起到了关键的作用。基于深度学习的图像识别需要大量的标签数据用于模型的训练,而人工收集标定数据需要花费大量的财力和时间。与此同时,随着图像搜索引擎和个人图像分享网站的发展,互联网上能获取到大量图像数据,这些数据有着数量大噪声多的特点,难以直接用于训练。基于目前的问题,本文提出了一种利用互联网上的海量数据训练图像识别模型的方法:利用爬虫自动爬取互联网上的图像数据;根据图像的深度视觉特征自动清洗数据;在预训练的模型上高效地训练模型。针对海量的数据处理,本文进一步利用云计算平台将服务搭建在分布式集群上,加快了整个流程地处理过程。最后,根据本文研究内容,实现了基于互联网和云计算的图像识别原型系统,并通过实验验证了本文方法的有效性。
[Abstract]:With the rapid development of in-depth learning, computer vision is used in the fields of transportation, medical treatment, security, advertising, personal image and video analysis, among which image recognition, which is the basis of computer vision, plays a key role. Image recognition based on deep learning requires a lot of tag data for model training, while manual collection of calibration data takes a lot of money and time. At the same time, with the development of image search engine and personal image sharing website, a large amount of image data can be obtained on the Internet. Based on the current problems, this paper proposes a method to train the image recognition model using the massive data on the Internet: crawler automatically crawls the image data on the Internet, automatically cleans the data according to the depth visual features of the image, and proposes a method to train the image recognition model based on the massive data on the Internet. Train the model efficiently on the pre-training model. For the massive data processing, this paper further uses cloud computing platform to build services on the distributed cluster, which speeds up the whole process of processing. Finally, according to the research content of this paper, the prototype system of image recognition based on Internet and cloud computing is implemented, and the effectiveness of this method is verified by experiments.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP183

【相似文献】