PIS-Net: A Novel Pixel Interval Sampling Network for Dense Microorganism Counting in Microscopic Images

Jiawei Zhang, Chen Li, Hongzan Sun, Marcin Grzegorzek

Abstract: A novel Pixel Interval Sampling Network (PIS-Net) is applied here for dense microorganism counting. The PIS-Net is designed for microorganism image segmentation with encoder to decoder architecture, and then the connected domain detection is applied for counting. The proposed method has good response for edge segmentation between tiny objects. Several classical segmentation metrics (Dice, Jaccard, and Hausdorff distance) are applied for evaluation. Experimental result shows that the proposed PIS-Net has the best performance and potential for dense tiny object counting tasks, which achieves \(96.88\%\) counting accuracy on the dataset with 420 yeast cell images. By comparing with the state-of-the-art approaches like Attention U-Net, Swin U-Net, and Trans U-Net, the proposed PIS-Net can segment the dense tiny objects with clearer boundaries and fewer incorrect debris, which shows the great potential of PIS-Net in the task of accurate counting tasks.