Recently, self-supervised contrastive learning has shown great potential in Graph Neural Network (GNN) through the use of contrastive learning to distinguish each node or instance in a dataset. However, a common issue in contrastive learning is the presence of "false-negative samples" where some negative samples are actually positive samples, leading to biased model learning. To address this issue, this paper proposes a new self-supervised contrastive learning framework called FD4GCL, which progressively detects and corrects false-negative samples. Specifically, in terms of attribute aware, an adaptive attribute threshold calculation method is established, which detects false-negative samples based on the similarity of attributes between nodes according to the attribute threshold. In terms of structure aware, considering the invariance of graph structure, a static structure threshold is set. Structural similarity between nodes is calculated, and false-negative samples are detected based on the structure threshold. The correction of a large number of false-negative samples by FD4GCL contributes to learning more generalized node representations with stronger generalization performance. Multiple experiments were conducted on various datasets, evaluating the model with node classification as the downstream task. The experimental results show that FD4GCL outperforms State-Of-The-Art methods with an average improvement of 1% in node classification accuracy on three citation datasets, and an average improvement of 0.5% on other datasets compared to the baseline methods.