深度机器视觉平臺 (ART/240CP)

深度机器视觉平臺 (ART/240CP)

深度机器视觉平臺 (ART/240CP)
ART/240CP
平台
16 / 11 / 2017 - 15 / 11 / 2019
14,721

蔡振荣博士

1. Deep machine vision software platform: Vision measurement software a. GUI based IDE for vision application R&D b. Runtime processing module for vision application deployment c. 2D/3D vision measurement module d. Computer aided parameter tuning module Specifications: Runtime environment: Windows / Linux / ARM Development environment: Windows Speed: < 0.2s for template matching @4M camera I/O: Serial port / TCP / UDP / GPIO 2. Deep machine vision software platform: Defect detection software a. Data pre-processing algorithms b. Deep learning training system for machine vision c. Defect detection algorithms Specifications: Runtime environment: Windows / Linux / ARM Speed: < 200ms @ 4M camera Accuracy: > 95% (10-fold) 3. Deep machine vision software platform: Defect classification software a. Deep learning based vision classification algorithms b. Defect classification algorithms Specifications: Runtime environment: Windows / Linux / ARM Speed: < 500ms @ 4M camera with < 50 subjects Accuracy: > 95% (10-fold) 4. 2.5/3D cover glass inspection system a. High stability glass conveyer module for inspection b. Curved glass illumination module c. Multi-view image capture module d. Deep machine vision software system integration & optimization for 2.5D/3D cover glass inspection system e. Prototype pilot-run at cooperation party factory Specifications: Glass size: 3-8 inch Capturing views: front-view, side-view Defect size: > 0.015mm Speed: < 5s (2.5D), <8s (3D) 5. (A) Contract service deliverables for BICI: Deep learning vision classification training system Specification: System accuracy: 90% TPR (True Positive Rate) @ 5% FPR (False Positive Rate) (B) Contract service deliverables for SAE: Customize a smart machine vision software for the Customer’s wafer/chip surface defect inspection Specifications: I Smart machine vision software (alpha version), comprising of: a) Positioning module b) Detection module (alpha version) c) Point-line module d) Calibration module e) Image IO (input-output) module f) System logging module g) Communication module II Smart machine vision software (beta version), comprising of: a) All features in item 1 b) Camera capture module & extending interface c) Detection module (beta version) d) Image pre-processing module e) Plugin-module III Smart machine vision software (final version), comprising of: a) All features in item 2 b) GUI (Graphical User Interface) defect model design module c) Customized defect detection model d) Defect detection module IX A standard operation procedure in PowerPoint slides format (C) Contract service deliverables for Prevision: Design and prototype of dual line 2.5D & 3D cover glass inspection system

北京协同创新研究院香港有限公司
日本电气香港有限公司
品图视觉科技有限公司
新科实业有限公司
深圳英诺敏科技有限公司


近年来,机器视觉技术飞跃发展,在发达国家的各种自动化生產在线中得到成功应用,该项技术的应用促使中国企业大量取代装配生產线上的操作工人(例如广东东莞实行的“机器换人”战略)。目前,由於消费者对產品品质要求不断提高,製造商必然僱佣更多工人进行表面缺陷检查,导致人力成本上升。而现有的大多数视觉技术是以尺寸测量等功能而设计的,儘管一些专业软体供应商也试图开发用於表面瑕疵检测模组,但是这类別模组通常是基於模板匹配或是基於人工编码等不灵活的方法。同时,这类型软体由於鲁棒性低、机器试执行时间长、以及对专业领域知识水准要求高等因素,通常不能满足表面瑕疵检查的要求。而机器视觉技术正是解决该问题的良方。 因此,本项目提出了深度机器视觉平台技术来加强缺陷检测的视觉应用。项目团队提出以下受知识產权保护的平台技术,以挖掘市场的需求,包括:(1)快速视觉应用开发技术对常用的视觉应用进行快速开发;(2)深度学习缺陷检测技术,使得瑕疵检测的方法可以直接从视像数据中学习得到,而无需视觉工程师进行手动设计;(3)基於深度学习的缺陷分类技术可以对瑕疵进行进一步的分析;(4)2.5D及3D的玻璃检测照明和採集技术。 本项目提出的平臺技术可以广泛应用於不同工业领域,如消费电子製造,半导体製造等。尤其是消费电子製造行业,大部分的生產商位於中国,同时他们年营业额也在快速增长。以手机显示屏为例,超过81%的產品是在中国製造的。本项目將要开发的平台技术可以帮助製造商以快速、低成本的方式构建视觉检测的解决方案。而且深度机器视觉平台技术可以帮助用户节省50%的视觉应用试运行时间,並降低30%的总成本。