A versatile AI-OCR model that can be widely adopted to promote digital transformation in trade operations
Key item reading accuracy of 92%* / Automatically reads 29 items regardless of format

On May 14th, we will begin selling the "Flax Scanner for Trade Document Packing List General-Purpose Model," an original high-precision AI-OCR for our Flax Scanner. The model automatically recognizes 29 items organized by frequency of use and priority and extracts text data from packing lists (P/L), a type of trade document that is widely used around the world but has different formats for each country and company.

In the past, we haveFlax Scanner for trade documents Commercial Invoice general purpose model”, “Flax Scanner for Trade Documents Bill of Lading General Purpose Model" and is focusing on digital transformation in trade operations by expanding its general-purpose form-specialized AI-OCR model that can read non-standard forms.

Packing List = A type of trade document.

The "Flax Scanner for Trade Document Packing List General-Purpose Model" can automatically read 29 frequently appearing items (see attached sheet) from among the items written on the document using AI. In addition, the 16 items with the highest importance among them have a high reading accuracy of 92%* on average for character unit accuracy and 87%* on average for item-by-item accuracy.

*All values are based on our test data.

It can be implemented by a wide range of companies that handle trade documents, such as exporters/importers (manufacturers/trading companies, etc.), shipping companies, airlines, forwarders, insurance companies (non-life insurance), banks, etc., and the system can be easily used using the UI. In addition to providing only the API, we can also provide customization proposals such as detailed tuning settings.

Cinnamon AI will continue to promote DX in the trading industry by developing versatile and highly accurate AI-OCR models that can be quickly introduced without the cost and time required to build a dedicated model. Masu.

*Please note that on April 23rd, we announced our AI-OCR platform for both standard and non-standard text.Flax Scanner HUB"Flax Scanner HUB" can be used with many document-specific general-purpose AI-OCR models, including the "Trade Document Packing List General-Purpose Model" announced today. "Flax Scanner HUB" is an AI-OCR solution that can be introduced in a wide range of business areas, including accounting, procurement, manufacturing, and construction.

▼ Target documents for "Flax Scanner HUB"

■ Background of the development of "Flax Scanner for Trade Document Packing List General-Purpose Model"

One of the trade operations in Japan involves checking and inputting necessary items from trade-related forms (trade documents). The format of the forms differs depending on the country and company, so experience and skills are required to confirm the necessary information, and the time-consuming input process is an issue. Additionally, because it is costly to tune each form and develop a dedicated AI-OCR model, there were a limited number of companies that could introduce it.

To address this issue, Cinnamon AI has so far developed a dedicated AI-OCR system that specializes in forms in the trade industry.Flax Scanner for trade documents Commercial Invoice general purpose model”, “Flax Scanner for Trade Documents Bill of Lading General Purpose Model”, and the “Join the trade consortiumWe have focused on promoting DX by introducing AI in the trading industry, and have built up a track record of implementation and know-how.

In order to enable more companies to adopt AI-OCR, we have now started selling the "Packing List General Purpose Model," following the "Commercial Invoice General Purpose Model" and "Bill of Lading General Purpose Model."

■Main target companies

Exporters/importers (manufacturers/trading companies, etc.), shipping companies, airlines, forwarders, insurance companies (non-life insurance), banks

■High-precision general-purpose AI-OCR that supports various formats realized with advanced AI technology

"Flax Scanner for Trade Documents Packing List General-Purpose Model" is an AI solution that minimizes the labor and cost required for practical AI-OCR implementation for any company that handles trade documents. Most AI-OCRs are developed using a "coordinate definition type" that defines in advance what is written where on a formatted document, but in global trade transactions, there are countless formats, so it is virtually impossible to deal with them using a "coordinate definition type". Developing an AI-OCR that can read a variety of forms and documents requires advanced technology called "feature learning type" that does not require coordinate definition of extracted items in advance, and the "Flax Scanner for Trade Documents Packing List General-Purpose Model" announced this time was developed using this "feature learning type".

"Feature learning type" AI-OCR allows AI to learn from a huge amount of data, finds features (patterns and consistency), and allows the AI to identify what kind of data the target data is. It is possible. Additionally, because it does not require coordinates to be defined in advance, it is the perfect technology for reading trade documents that come in a myriad of formats.

Cinnamon AI has advanced AI technology that allows AI to understand meaning and read data from a wide range of formats like a human, and these are collectively referred to as IDP (intelligent document processing = automatic knowledge extraction technology). Excellent in the technical field called for. IDP is a very important technology that is the basis for realizing hyper-automation in enterprises.

 "Flax Scanner for Trade Document Packing List General Purpose Model" can read 29 items
Based on interviews with a wide range of companies that handle trade documents, such as exporters/importers (manufacturers/trading companies, etc.), shipping companies, airlines, forwarders, insurance companies (non-life insurance), and banks, the following 29 items were determined as target items for reading the Packing List as a general-purpose model.

(Highly important items are *)
Items marked with (details) are read in table format.

◇Reading accuracy (test data value by Cinnamon AI)
Important item accuracy ● Character unit accuracy: Average 91.98% / ● Item-specific accuracy: Average 86.86%
Accuracy of all items ● Character unit accuracy: Average 92.60% / ● Accuracy by item: Average 85.11%

No.	Reading item name
1	Shipper/Exporter Name
2	Consignee/Messrs/Shipped to Name
3	Invoice Date/Date
4	Invoice No.*
5	Contract No.
6	Order No.
7	P/O No.
8	Vessel Name/Shipped per//M/V
9	Voyage No
10	Port of Loading/Port of Shipment/From
11	Shipment Date/Date of Departure/ON
12	Final Destination/Port of destination/Destination/To
13	Country of Origin
14	Payment Terms
15	Case Mark/Marks & Nos./Shipping Mark*
16	Item Name/Description of Goods*
17	Item No./Lot No. (Details)*
18	Container No. (Details) *
19	Item Quantity/Unit Quantity*
20	Unit*
21	Pcs (details)*
22	Net Weight (Details)*
23	Gross Weight (Details)*
24	Measurement/M3 (details)*
25	Total Quantity*
26	Total Pcs*
27	Total Net Weight*
28	Total Gross Weight*
29	Total Measurement/Total M3*