Lightweight Hierarchical Dynamic Adaptation Network:Enhancing Few-Shot Object Detection with Robust FeatureRecalibration

Few-shot object detection (FSOD) is a challenging task that aims to accurately locate and identify specific objects with minimal data. Despite recent advances in transfer learning and meta-learning methods, FSOD still faces difficulties in achieving effective feature representation. To address this, we propose a Lightweight Hierarchical Dynamic Adaptation Network (LHDAN), which enhances feature extraction and recalibration processes without significantly increasing the number of training samples. LHDAN leverages often-overlooked intermediate features and introduces a composite loss function to dynamically focus on relevant features during learning. Extensive experimental results demonstrate that our method achieves state-of-the-art performance across multiple FSOD benchmark datasets, with approximately 1.5\% improvement over most existing methods. By redefining the feature extraction and recalibration processes, LHDAN strengthens the model's discriminative ability and robustness, setting a new benchmark for FSOD.