More than 6300 km of the Yangtze River, the Qiantang River in Hangzhou Bay intersection, vast stirred waterways and coastline of gold at one point forming a 10.7 km deep-water coastline, the waters of broad - Ningbo Daxie China Merchants International Terminal sits Here. On the pier, a ship docked at the shore of the huge outer, combined with the truck crane methodically loading and unloading containers. No black smoke, loud noise, no crowded wave after wave of handling personnel, busy and orderly behind the marina, self-development and exploration intelligent dock for many years.
Figure: Ningbo Daxie China Merchants International Terminal
"It should be 2015, we began the transition to the intelligent." IT system administrator Ningbo Daxie China Merchants International Terminal of the Kid said, "The main focus in the transformation scenario shore, gate, yard three, and and there is a special department responsible for co-ordination. "Not difficult to see, when people are still part of the" time of AI (artificial Intelligence) "as far away in the sky" Star ocean ", but it also has more people thinking about how the "AI" filling in traditional industries, but also ran a Chinese speed "intelligent" upgrade.
Intelligent upgrading of traditional dock landing had never been so fast.
In the field of container terminals, container door toward recognition it is important, which directly affects the efficiency of transport and dock turned handling costs. If the container box to begin the wrong direction, we need to use large dedicated machine be turned around, move up to a cost of tens of dollars. For every day thousands of tank transport dock, this is not a small potential costs, so the need for efficient container door toward reconciliation.
Traditional Dock mainly rely on manual check of container information and input, low efficiency audit. Ningbo Daxie China Merchants International Terminal one day amount to begin in about 3000, manual checking workload, thus an urgent need for an intelligent matching program to improve audit efficiency.
When Daxie China Merchants International Terminal to face the problem, only a week to the cost of investment is almost zero budget, use Baidu platform brain EasyDL quickly complete the training and deployment model to achieve the human eye camera replacement of container information checking, real-time judge a shutter container door in the correct direction.
Figure: Ningbo Daxie China Merchants International Terminal
"我们是在一次业内活动中，听说的EasyDL，抱着试一试的态度，去尝试了下。"宁波大榭招商国际码头技术信息技术系统管理员陈文杰介绍道："此前我们也尝试过其他的解决方案，但结果不是很理想。而如果要用传统的产品定制化去专门做一套系统出来的话，财务上的投入产出比又不划算。"陈文杰指出的，正是传统产业在进行产业智能化转型时会遇见的普遍问题——随着产业转型，每个生产环节的细节可能都需同步升级，然而却由于资金、资源、技术、时间等众多因素，在进退维谷中被暂缓。
还记得在今年的百度AI开发者大会上，百度以"产业智能化"为主题交出的成绩单中，每个AI应用案例都写实地印证着那句"Do Better，科技为更好！"而当百度将AI技术开放开源、进一步降低门槛后，如何通过一种简单的方式来代替"精通算法"从而快速让非AI出身的各个企业从业人员都可以使用AI技术，也变得更加有迹可循。这之中，宁波大榭招商国际码头通过自主训练、自定义模型，并将结果快速应用于实践解决"堆场"箱门朝向识别问题的例子，就是这种"快速自定义AI应用"的探索之一。
图：百度EasyDL定制化训练和服务平台
据了解，百度大脑EasyDL是一个面向企业的全流程的AI模型开发和服务平台，而且可支持不需要算法编程的AI模型训练。陈文杰的团队最开始先从历史核查图片中，挑选了500张，将错误和正确的箱体方向图片素材，分别进行分类训练。经过短短1周迭代出3次模型，最终完成了闸口箱门识别模型，准确率在当下就已高达95%左右。此外，EasyDL还同步提供给他们自动化的模型发布和服务部署，只需选择校验通过的模型版本，提交服务发布申请，即可获得一个服务API接口。而且这个服务接口有完整的流控、鉴权等安全机制，稳定性与可用性与百度大脑应用于百度自身的AI应用服务水准保持一致。而在模型发布后，通过云端调用数据功能，还可以一键将实际调用的数据关联模型训练集，实现数据闭环，持续进行模型效果的优化。
图：通过EasyDL进行集装箱箱门识别
原本，需要1名员工每天核对3000余张图片，现在用EasyDL模型就可先进行机器筛选，再由人工有针对性地复审即可。因此，现在这项工作只需1名员工辅助核对150张图片就可以完成一天的审核工作了。这样不仅提升了审核效率，也大大释放了审核人员的时间，增加了投入到更多的数据研究和业务讨论中的时间。
面对20年前国外港口在"堆场"实现自动化作业的画面，中国码头企业曾有过的憧憬，现已随着AI深度学习等技术加速进行着"弯道超车"。目前，国内外重要码头均有新建和改造自动化码头的计划与探索，而宁波大榭招商国际码头作为集装箱吞吐量316万标箱的国际大港更是已有多年探索经验。"传统码头自动化转型战略于2016年作为宁波大榭招商国际码头的IT规划开始实施"。
链接：
http://baijiahao.baidu.com/s?id=1640014838351860071&wfr=spider&for=pc