In the realms of computer vision and robotics, the concept of visual place recognition holds significant prominence. Its objective revolves around equipping a model with the capability to identify, from a provided query image, the most analogous images within a database, thereby identifying the place represented by the query image. Current visual place recognition models typically rely on triplet loss functions for training, but such loss functions have limitations. Traditional triplet loss functions only classify database samples into positive and negative classes, without considering the importance ranking among samples. Some samples may be more similar to the query image and contain more useful information, thus deserving more attention during training. In tackling this problem, our approach introduces a novel loss function termed the distance-ranking-based weighted triplet loss. This unique loss function assigns weightage to triplets by evaluating the spatial gap separating positive instances and the queried image, thereby intensifying the emphasis on pivotal samples. Within the framework of place recognition tasks utilizing the NetVLAD pipeline, our method achieves approximately a 1% improvement in both the Recall@l and Recall@5 compared to traditional triplet loss function.