A multi-focus and multi-exposure target recognition approach using bionic compliant mechanisms

Abstract For laser direct-imaging (LDI) machines, out-of-focus-plane (OFP) blur, abnormal illumination, and target defects degrade recognition accuracy and alignment performance. Existing single-focus and single-exposure approaches struggle to handle coupled disturbances under complex industrial conditions effectively. To address this, a multi-focus (MF) and multi-exposure (ME) target recognition approach is proposed, in which a pigeon-inspired bionic compliant mechanism (BioCM) provides high-precision, focal-plane modulation for acquiring MF image sequences. A guided filtering (GIF)-based MF fusion method then restores all-in-focus structural information and suppresses the out-of-focus-plane (OFP) blur. Building on the fused MF outputs, a mask-guided ME fusion method based on structural patch decomposition (SPD) enhances robustness against abnormal illumination and target defects while preserving background cleanliness. The restored images are subsequently localized via RRHT. A dataset with coupled disturbances was constructed and evaluated on a prototype LDI machine. Under severe coupled disturbances, the target-background clarity rating (TBCR) improves by 83.9% and 89.2% compared with the rigid-mechanism (RM) and BioCM, respectively. The localization error is 3.0 px under minor and moderate target defect conditions.