Efficient Edge-Compatible CNN for Speckle-Based Material Recognition in Laser Cutting Systems

TL;DR

The paper addresses the need for accurate, real-time material recognition in laser cutting using speckle-based sensing. It proposes a compact CNN with depthwise separable convolutions and global pooling, achieving 95.05% accuracy on 59 material classes in the SensiCut dataset while requiring only 0.34M parameters and enabling edge deployment at ~295 fps. Grouping into material families yields near-perfect recall, directly mapping recognition to laser presets and improving safety and efficiency. The work demonstrates that domain-tailored, resource-efficient models can outperform larger backbones in speckle-based classification and paves the way for autonomous, material-aware fabrication systems.

Abstract

Accurate material recognition is critical for safe and effective laser cutting, as misidentification can lead to poor cut quality, machine damage, or the release of hazardous fumes. Laser speckle sensing has recently emerged as a low-cost and non-destructive modality for material classification; however, prior work has either relied on computationally expensive backbone networks or addressed only limited subsets of materials. In this study, A lightweight convolutional neural network (CNN) tailored for speckle patterns is proposed, designed to minimize parameters while maintaining high discriminative power. Using the complete SensiCut dataset of 59 material classes spanning woods, acrylics, composites, textiles, metals, and paper-based products, the proposed model achieves 95.05% test accuracy, with macro and weighted F1-scores of 0.951. The network contains only 341k trainable parameters (~1.3 MB) -- over 70X fewer than ResNet-50 -- and achieves an inference speed of 295 images per second, enabling deployment on Raspberry Pi and Jetson-class devices. Furthermore, when materials are regrouped into nine and five practical families, recall exceeds 98% and approaches 100%, directly supporting power and speed preset selection in laser cutters. These results demonstrate that compact, domain-specific CNNs can outperform large backbones for speckle-based material classification, advancing the feasibility of material-aware, edge-deployable laser cutting systems.

Figures (6)

• Figure 1: Proposed lightweight CNN architecture for speckle-based material classification. The design balances accuracy and efficiency using depthwise and pointwise convolutions, global average pooling, and fully connected layers.
• Figure 2: Training and validation accuracy across epochs. The model converges stably, reaching a validation accuracy of 93.76% at epoch 493.
• Figure 3: Training and validation loss curves. The model shows no evidence of overfitting, confirming the effectiveness of augmentation and regularization.
• Figure 4: Confusion matrix for the 59-class classification task. Strong diagonal dominance demonstrates high per-class accuracy, with misclassifications primarily occurring among similar materials such as hardwoods, tinted acrylics, and felts.
• Figure 5: Confusion matrix for the nine-family grouping.