基于华为昇腾Atlas 200DK的黑白图片上色实验  

Black-and-White Image Coloring Experiment Based on Huawei’s Ascend Atlas 200DK

作　　者：武畅 夏玉林 杨小漫 王宏 WU Chang;XIA Yulin;YANG Xiaoman;WANG Hong(School of Information and Communication Engineering,University of Electronic Science and Technology of China,Chengdu 611731,China)

机构地区：[1]电子科技大学信息与通信工程学院,成都611731

出　　处：《实验科学与技术》2025年第2期14-22,共9页Experiment Science and Technology

基　　金：教育部-华为“智能基座”产教融合协同育人项目;电子科技大学研究生教研教改重点项目(JYJG2021401);电子科技大学本科实验教学和教学实验室建设研究项目(BK-SYJY-202423);人工智能技术赋能本科教学教改项目(2024AIXM007)。

摘　　要：随着人工智能(AI)技术的发展,大学教学中相关课程和技术应用的比重不断上升,使学生对人工智能的实验需求爆发式增长。但是,目前的人工智能实验体系并不完善,实验形式主要以软件实现为主,缺乏基于硬件的平台和实验项目,与实际工程应用和行业需求脱节。结合CDIO(构思、设计、实现、运作)模式,针对实验中的深度学习网络推理速度缓慢、耗时过长等问题,提出了以华为昇腾Atlas 200DK硬件平台为基础,利用深度学习网络,完成黑白图像上色的硬件人工智能实验。通过该实验,学生可以学习人工智能的原理、神经网络的结构和推理过程,搭建软件开发环境和配置硬件系统,实现软硬件协同的神经网络的部署,完成功能和相关指标测试。该实验有助于学生立足于实际应用需求理解神经网络的运行机理,掌握其在实际硬件平台上的应用方法,从而全面提升软硬件的工程实践能力。With the development of artificial intelligence(AI)technologies,the proportion of related courses and technological applications in university curricula has been increasing,leading to a surge in students’demand for AI experimentation.However,the current AI experimental systems are not well-established,with most experiments being software-based and lacking hardware platforms and projects,which results in a disconnect from real-world engineering applications and industry needs.This experimental instruction integrates the CDIO(conceive,design,implement,operate)model,and to address issues such as slow inference speeds and long processing times in deep learning networks during experiments,an AI experiment based on the Huawei Ascend Atlas 200DK hardware platform is proposed,which utilizes deep learning networks to perform colorization of grayscale images.Through this experiment,students can learn about the principles of AI,the structure and inference processes of neural networks,set up software development environments,configure hardware systems,and deploy neural networks that integrate both software and hardware.The experiment also involves the completion of functional and performance testing.This hands-on experience helps students understand the operational mechanisms of neural networks from a practical application perspective,master their application methods on actual hardware platforms,and thereby enhance their overall engineering and practical skills in both software and hardware.

关 键 词：人工智能 Atlas 200DK 卷积神经网络 图像上色 产教融合 

分 类 号：G642[文化科学—高等教育学] TN85[文化科学—教育学]