A Lightweight Cutting Tool Recognition Model Based on LResNet-ELA

Introduction: In intelligent industrial manufacturing, the rapid and accurate identification of cutting tools is crucial for production efficiency and processing quality. However, practical applications often face challenges such as the scarcity of dedicated tool datasets and the difficulty in balancing recognition accuracy with lightweight deployment on embedded devices. Methods: This paper proposes a lightweight cutting tool recognition model based on the LResNet- ELA architecture. The model integrates MobileNet's depthwise convolution to reduce parameters and computational costs, ResNet's residual structure to address deep network training issues, and the ELA mechanism to focus on key tool features. A dataset of 4207 cutting tool images was collected using industrial cameras, covering typical tool types, including end mills, face mills, taps, and drills. Results: The proposed LResNet-ELA model achieved a recognition accuracy of 99.48%, with a parameter count of only 0.23 million, ensuring both high performance and efficiency. These results indicate that the architecture is well-suited for real-time deployment in embedded devices. Discussion: The results demonstrate that the LResNet-ELA model offers an effective solution for cutting tool identification, balancing high recognition accuracy with resource-efficient deployment. However, the model's performance could be further validated under different industrial environments and tool types to ensure generalizability. Conclusion: This lightweight architecture provides a feasible solution for real-time tool identification on embedded devices in machine tool environments, offering a balance between high recognition accuracy and resource-efficient deployment.