Patch and Shuffle: A Preprocessing Technique for Texture Classification in Autonomous Cementitious Fabrication

TL;DR

The paper tackles texture classification in autonomous cementitious fabrication, where semantic cues can mislead models. It introduces a patch-and-shuffle preprocessing step that segments images into patches, shuffles them, and reconstructs a jumbled input before the classifier, encouraging reliance on local textures. Using a ResNet-18 backbone and a controlled comparison, the method substantially improves accuracy (best 90.64% vs 72.46%; average 86.84% vs 68.71%), with strong gains on several texture classes. The findings suggest that disrupting global structure enhances low-level texture learning and may generalize to other domains requiring fine-grained texture discrimination, with potential for real-time monitoring and automated adjustments in fabrication workflows.

Abstract

Autonomous fabrication systems are transforming construction and manufacturing, yet they remain vulnerable to print errors. Texture classification is a key component of computer vision systems that enable real-time monitoring and adjustment during cementitious fabrication. Traditional classification methods often rely on global image features, which can bias the model toward semantic content rather than low-level textures. In this paper, we introduce a novel preprocessing technique called "patch and shuffle," which segments input images into smaller patches, shuffles them, and reconstructs a jumbled image before classification. This transformation removes semantic context, forcing the classifier to rely on local texture features. We evaluate this approach on a dataset of extruded cement images, using a ResNet-18-based architecture. Our experiments compare the patch and shuffle method to a standard pipeline, holding all other factors constant. Results show a significant improvement in accuracy: the patch and shuffle model achieved 90.64% test accuracy versus 72.46% for the baseline. These findings suggest that disrupting global structure enhances performance in texture-based classification tasks. This method has implications for broader vision tasks where low-level features matter more than high-level semantics. The technique may improve classification in applications ranging from fabrication monitoring to medical imaging.

Figures (4)

• Figure 1: Texture Windows - Unmodified from 2020_paper
• Figure 2: Classification Pipeline
• Figure 3: Original image (left) compared to transformed image (right)
• Figure 4: Proposed Classification Pipeline