Table 4 Shows that the proposed RCA-UNet performs better on the two tongue image datasets. In terms of the MIoU score, RCA-UNet improves by 0.62% on our dataset compared with UNet and improves by 2.44% on the public dataset BioHit. In terms of the dice score, the proposed RCA-UNet outperforms UNet by 0.36% and 1.62%, respectively. Additionally, RCA-UNet decreases by 0.11 and 0.4 on MHd, proving that RCA-UNet segmentation produces results that are more similar to those of the manually labeled mask model. These data indicate that the introduced RCBAM has learned more tongue features, which effectively improves the network’s ability to distinguish between the target and the background. The changes in MHd indicate that RCA-UNet, compared with UNet, retains more information, thus compensating for the feature loss in the network.

From: Tongue shape classification based on IF-RCNet