Semantic-guided Nematode Instance Segmentation

Nematodes are widely recognized as an important biological model. Accurate nematode segmentation and monitoring of nematode morphological changes are crucial in genetic research and pharmaceutical development. However, the presence of impurities with similar characteristics in the culture environment poses a challenge to nematode segmentation. To address this issue, we propose a semantic-guided transformer network that effectively differentiates between nematodes and background elements with similar features. Our network employs high-level semantic features within the cross-attention mechanism of the decoder, enhancing the network’s ability to capture nematode-specific features while mitigating background interference. To validate the efficacy of our approach, we conducted comparative experiments on two nematode datasets (BBBC010 and CSB-1). The results demonstrate that our method successfully addresses complex background interference and improves segmentation accuracy.