Kailing Technology · AI OCR Enterprise loginFree trial
AI OCR Visual Capability Platform / Customs declaration recognition: multi-page declarations automatically aggregated into one document

Customs declaration form recognition: multi-page customs forms automatically aggregated into one declaration

Scanned copies of customs export declarations are uploaded as a whole package. AI extracts the declaration number/date/consignee/commodity line amount. Multiple pages of the same declaration are automatically merged, and amounts are summarized in separate columns by currency such as USD/CNY/MNG.

Register for free trial (10 times per module)Get the solution and quote

Core Capabilities

Automatic multi-page aggregation

Automatically group multi-page 1/2, 2/2 into one ticket by the 18-digit customs declaration number, with header fields based on the first page and amounts accumulated across pages.

Automatic currency aggregation

Product line total prices are listed and counted by currency (USD/CNY/MNG/EUR, etc.), directly usable for export tax rebate reconciliation.

Full-package batch recognition

Supports automatic decompression of zip package uploads, 8-concurrent batch recognition, and real-time progress and throughput display.

Automatic portrait orientation correction

Horizontally oriented customs documents placed vertically during scanning are automatically rotated upright, with no manual preprocessing required.

Extractable fields

Customs Number (18-digit Declaration Number)Export dateOverseas consigneeContract/agreement numberTransport vehicle/vehicle numberProduct line total price and currency systemPage (N/M)

Typical scenarios

Common Questions

How is a multi-page invoice order closure handled?
The system automatically groups by customs declaration number, sorts by page number, takes header fields from the first page, accumulates commodity line amounts across pages, and outputs one invoice per line.
Can it recognize customs declarations photographed with a phone?
Yes. The system automatically rotates tilted and vertically placed images upright. It is recommended to take clear photos and avoid strong reflections.
How long does it take to recognize one?
Typically 3-8 seconds for a single page, using the local recognition engine fast track; when the layout is complex, it automatically falls back to the large model for deep recognition, preferring slower over errors.
Is the data secure?
Supports private deployment, with recognition data not leaving the enterprise's internal network; the SaaS version uses HTTPS encrypted transmission throughout.