Semi-Supervised Learning-Driven New Strategy for YOLO Sample Less Target Detection

Insufficient labelled data is an outstanding challenge in the field of target detection. In this paper, we propose a semi-supervised learning-based YOLO few-sample target detection framework, which constructs an efficient few-sample detection system through pseudo-label generation and refinement, feature enhancement and multi-scale detection optimisation. A hybrid supervised training strategy, a cross-domain knowledge migration mechanism and a semi-supervised active learning method are designed to effectively improve the model's detection performance and generalisation ability in data-scarce environments, as well as reduce the annotation cost and enable edge device deployment.