ISD-YOLO: A lightweight and efficient feature extraction network for invoice seal detection

Xiaosi Song, Xiaohong Zhao, Caihua Ma, Wenpei Xu, Peng Shen · 2026

Read ISD-YOLO: A lightweight and efficient feature extraction network for invoice seal detection by Xiaosi Song, Xiaohong Zhao, Caihua Ma, Wenpei Xu, Peng Shen (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/plos-pone-0358190.txt.

Description

Xiaosi Song and colleagues published a 2026 computer vision paper in PLOS ONE on finding the red official seals stamped on invoices, a step that matters for automated financial review. Seals vary in size, sit on cluttered backgrounds, and the detector must stay light enough for modest hardware. Their model, ISD-YOLO, adapts a popular detection network with modules that sharpen features at several scales, focus attention on promising regions, and adapt to the seal's shape, behind a slim detection head.

Tested on a mix of public seal images and real invoices, the model beat its baseline and rival detectors on standard accuracy scores while using only a few million parameters. The authors say the balance of accuracy and lightness suits resource-limited settings such as office scanners or edge devices. Broader scenarios and real edge hardware still need testing before the design counts as field-proven.

Catalog record

CREATORXiaosi Song, Xiaohong Zhao, Caihua Ma, Wenpei Xu, Peng Shen
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE84527
LICENSECC_BY
RIGHTSCC BY
SOURCEPLOS

Source: https://doi.org/10.1371/journal.pone.0358190

Browse the full library · About the Tower of Babel