Non-Contact Measurement of Pregnant Sows’ Backfat Thickness Based on a Hybrid CNN-ViT Model

Backfat thickness (BF) is closely related to the service life and reproductive performance of sows. The dynamic monitoring sows’ BF a critical part production process in large-scale pig farms. This study proposed application hybrid CNN-ViT (Vision Transformer, ViT) model for measuring address problems high measurement intensity caused by traditional contact low efficiency existing non-contact models BF. introduced depth-separable convolution lightweight self-attention, mainly consisting Pre-local Unit (PLU), Lightweight ViT (LViT) an Inverted Residual (IRU). could extract local global features images, making it more suitable small datasets. was tested on 106 pregnant sows with seven randomly divided results showed that had Mean Absolute Error (MAE) 0.83 mm, Root Square (RMSE) 1.05 Percentage (MAPE) 4.87% coefficient determination (R-Square, R2) 0.74. Compared LviT-IRU, PLU-IRU PLU-LviT, CNN-ViT’s MAE decreased than 12%, RMSE 15%, MAPE 15% R² improved 17%. Resnet50 ViT, 7%, 13%, 7% R2 15%. method better meet demand automatic actual provide technical support intelligent management