<p>To address the issue of insufficient and homogeneous background samples in the training datasets for robotic arm target detection tasks, this study proposes an efficient recognition framework named Adaptive Weighted Contrastive Fusion Pyramid Network (AWCFPN). In this framework, a novel foreground enhancement module, Adaptive Weighted Contrastive Module (AWCM), is designed. This module generates adaptive weights by computing the contrast between optimized target regions and background regions, and utilizes these weights to adjust attention features extracted from both foreground and background. This process enhances the alignment of key regions with foreground objects. Furthermore, the framework integrates spatial feature reconstruction and pyramid contextual extraction to promote multi-scale feature fusion, enabling the model to better understand the relationships between scenes and objects in an image. Finally, feature distillation is employed to meet the lightweight and deployment requirements of practical applications. The proposed method was trained and tested on both the public PASCAL VOC dataset and a self-built dataset.Meanwhile, in a vision-guided robotic arm grasping scenario, target detection and grasping experiments were conducted with a coaxial dual-rotor UAV as the dynamic target.Experimental results show that the method achieves significant improvements in foreground enhancement tasks, and the strategy increases the robotic arm’s grasping success rate.</p>