<p>Unsupervised few-shot learning faces challenges such as "sampling bias" and "class collision," which make it difficult to extract effective features from limited samples and map them into a separable feature space, thus weakening the model’s generalization ability. To address this issue, this paper proposes a novel unsupervised few-shot learning method—Enhanced Contrastive Self-supervised Learning (ECSL). ECSL includes a multi-task training framework, a General Feature Enhancement (GFE) module, a Projection Distance Metric Algorithm (PDMA), and an Adjustable Loss function (AL). The method aims to enhance the model’s understanding of samples through complex task collaboration, enabling it to extract more generalized semantic information and thereby establish an effective mapping from samples to a separable feature space. Experimental results show that ECSL outperforms existing methods such as PsCo and CPNWCP on the miniImageNet and tieredImageNet datasets. In the 5-way 1-shot task, ECSL improves classification accuracy by 2.10% and 2.11%, respectively, compared to CPNWCP, and in the 5-way 5-shot task, it improves by 0.66% and 1.55%. These results show significant competitiveness among unsupervised few-shot learning methods, particularly in the miniImageNet 5-way 1-shot experiment, where ECSL even exceeds traditional supervised methods.</p>