Core demands
Ningbo Daxie Investment Promotion International Wharf is located at the junction of Hangzhou Bay and Qiantang River, the golden waterway of the Yangtze River. For this terminal container area located on a 10.7km wide deep water shoreline, the identification of container door orientation is very important, which directly affects the transportation efficiency and transfer handling costs of the terminal. If the outlet container box is in the wrong direction, it needs to use large special machinery for transfer, and the cost of moving it once is as high as tens of yuan. This is a large potential cost for the wharf with thousands of boxes transported every day.
However, traditional terminals mainly rely on manual verification and entry of container information, and the audit efficiency is low. Ningbo Daxie Merchants International Wharf has about 3000 gates a day, and the workload of manual verification is heavy. Therefore, an intelligent verification scheme is urgently needed to improve the efficiency of verification, so as to meet the demand for efficient verification of container door orientation.
Solution
Faced with the picture of foreign ports' automated operation in the "storage yard" 20 years ago, Chinese terminal enterprises once had a vision of "overtaking on the curve" with AI deep learning and other technologies. At present, there are plans and explorations for the construction and transformation of automated terminals at important terminals at home and abroad. Ningbo Daxie Merchants International Terminal has many years of exploration experience as an international port with a container throughput of 3.16 million TEUs. The traditional terminal automation transformation strategy was officially implemented as the IT plan of Ningbo Daxie Merchants International Terminal in 2016.
"It should be 2015, and we have started to transform to intelligence." Chen Wenjie, the information technology system administrator of Ningbo Daxie Merchants International Terminal, said, "The main transformation application scenarios are concentrated on the shore, gate, and yard, and there are special departments and heads to collect funds."
From Chen Wenjie's words, it is easy to see that while some people still regard "AI" as "the sea of stars" in the sky, more people are thinking about how to add "AI" to traditional industries and run out of another Chinese speed of "intelligent" upgrading. The intelligent upgrading of traditional docks has never been so fast.
Chen Wenjie said: "We heard about EasyDL in an industry event, and tried it with a try attitude. We have tried other solutions before, but the results are not very ideal. If we want to use traditional product customization to create a system specifically, the financial input-output ratio is not cost-effective."
Chen Wenjie pointed out that it is the common problem that traditional industries will encounter when carrying out industrial intelligent transformation - with the industrial transformation, the details of each production link may need to be upgraded simultaneously, but due to many factors such as capital, resources, technology, time, etc., it has been postponed in the advance and retreat.
At the 2019 Baidu AI Developers Conference, in the transcript submitted by Baidu with the theme of "industrial intelligence", each AI application case is a realistic proof of the sentence "Do Better, technology is better!" And when Baidu opened up AI technology and further lowered the threshold, how to replace "proficient algorithm" with a simple way So that employees of all enterprises without AI background can use AI technology quickly and become more traceable.
Among them, Ningbo Daxie Merchants International Terminal, through independent training, customized models, and quickly applied the results to practice to solve the problem of "yard" door orientation recognition, is one of the explorations of this "quick customized AI application".
Chen Wenjie said: "It is understood that Baidu Brain EasyDL is an enterprise oriented full process AI model development and service platform, and it also has the advantage of supporting AI model training without algorithm programming." So his team first selected 500 pictures from the historical verification pictures, and conducted classified training on wrong and correct box direction pictures. After only one week of iteration, the model was developed three times, and the identification model of gate box and door was finally completed, with an accuracy rate of about 95% at present.
In addition, EasyDL also provides them with automated model publishing and service deployment synchronously. Just select the verified model version and submit the service publishing application to obtain a service API interface. Moreover, this service interface has a complete flow control, authentication and other security mechanisms, and its stability and availability are consistent with Baidu Brain's own AI application service level. After the model is released, the data function can be called through the cloud, and the actually called data can also be linked to the model training set with one key to achieve data closed-loop and continuously optimize the model effect.
Therefore, when the technical team of Ningbo Daxie Merchants International Terminal faced the review problem of container door turning direction, it only took one week to quickly complete model training and deployment with Baidu Brain EasyDL platform at almost zero budget cost, so that the camera could check the container information instead of the human eye, Real time judge whether the container door of the gate is correctly placed.
It used to require one employee to check more than 3000 pictures every day. Now we can use the EasyDL model to screen the pictures by machine first, and then review them manually. Therefore, only one employee is needed to check 150 pictures to complete the day's audit work, which not only improves the audit efficiency, but also greatly frees the time of auditors.
In the future, it has completed the transformation from a traditional terminal to an intelligent Ningbo Daxie Investment Promotion International Terminal, which can be rapidly upgraded and put more energy into data research and business development.
