創新及科技解決方案

解決方案編號

S-0862

解決方案名稱

AI 收據資料提取——從收據照片到結構化紀錄

解決方案描述

Coders currently read receipts by hand and key in item description, outlet name/type, purchase date, amount and delivery charge.

Our solution puts a record layer under the process: receipt images (through scanning or uploading) are processed by OCR and vision/language models into the published field set, each field is carrying a confidence score; coders review and confirm against the source image instead of manually keying from scratch; confirmed data will be exported as CSV/JSON/XLSX or via API in line with the existing coding workflow.

The base is Busabase, an open-source system of record — every record keeps its source image, provenance and review history, and the system can be deployed in an environment designated by the organisation.

應用領域

工商業

人口

使用的技術

人工智能

電腦視覺

數據分析

圖像識別

機器學習

自然語言處理

使用例子

A coder receives a batch of receipt photos submitted by survey respondents. The system extracts item description, outlet name and type, purchase date, amount spent and delivery charge into structured fields, flagging low-confidence items. The coder opens the review queue, checks flagged fields against the receipt image, confirms or corrects, and exports the batch as CSV/XLSX in the existing coding format.

A field-level accuracy report against human-coded samples is produced for the organisation. The same pipeline accepts new document templates, supporting the expansion to other document types stated in the need.

若政府部門欲對創科方案進行PoC試驗或技術測試,請聯絡Smart LAB。