Infoblock alpha
cs-cv/vision IB-arxiv-2607.28481-accuracy measured

A fuzzy rule-based neuro-symbolic framework (Swin Transformer perception plus J48-derived fuzzy IF-THEN rules) improved accuracy by 17.9% over image-only classification on a 3,244-image sewer-defect dataset with five imbalanced severity classes.

Active Evidence state is recorded explicitly and may be revised by a later event.

Observed quantities

accuracy_improvement_pct 17.9 · n_images 3244 · n_classes 5

01

Primary source

  1. arxiv · 2607.28481 · v1 · abstract
Title
A Fuzzy Rule-based Neuro-Symbolic Approach for Pipe Severity Prediction in Sewer Networks
Authors
Ngoc Thai Le, Thanh Ma, Umberto Straccia
Version
v1
Check
Support quote found word for word in the arXiv abstract v1, 2026-10-08
Supporting quote from the source
Our results show an improvement of accuracy, balanced accuracy, Macro F1 and MCC by 17.9%, 12.2%, 23.0%, and 17.3%, respectively, over image-only based classification

02

Boundary of this block

No additional public limitation was recorded.

03

Connections

Outgoing

No outgoing public edges.

Incoming

No incoming public edges.