SMR-Net:Robot Snap Detection Based on Multi-Scale Features and Self-Attention Network

TL;DR

This paper introduces SMR-Net, a self-attention-based multi-scale detection algorithm designed to improve snap detection and localization accuracy in robotic assembly, especially under complex visual conditions.

Contribution

The paper proposes a novel attention-enhanced multi-scale feature fusion architecture for snap detection, significantly improving robustness and precision over traditional methods.

Findings

IoU improved by 6.52% and 5.8% on two datasets

mAP increased by 2.8% and 1.5%

Superior performance in complex snap detection tasks

Abstract

In robot automated assembly, snap assembly precision and efficiency directly determine overall production quality. As a core prerequisite, snap detection and localization critically affect subsequent assembly success. Traditional visual methods suffer from poor robustness and large localization errors when handling complex scenarios (e.g., transparent or low-contrast snaps), failing to meet high-precision assembly demands. To address this, this paper designs a dedicated sensor and proposes SMR-Net, an self-attention-based multi-scale object detection algorithm, to synergistically enhance detection and localization performance. SMR-Net adopts an attention-enhanced multi-scale feature fusion architecture: raw sensor data is encoded via an attention-embedded feature extractor to strengthen key snap features and suppress noise; three multi-scale feature maps are processed in parallel with standard and dilated convolution for dimension unification while preserving resolution; an adaptive reweighting network dynamically assigns weights to fused features, generating fine representations integrating details and global semantics. Experimental results on Type A and Type B snap datasets show SMR-Net outperforms traditional Faster R-CNN significantly: Intersection over Union (IoU) improves by 6.52% and 5.8%, and mean Average Precision (mAP) increases by 2.8% and 1.5% respectively. This fully demonstrates the method's superiority in complex snap detection and localization tasks.