Development of a duck weight estimation system using Vision Transformer and weighted random sampling

Abstract

This study aims to enhance the efficiency of a duck farm by developing a deep learning-based weight estimation system. We utilized Convolution Neural Network (CNN) and Vision Transformer (ViT) models, applying weighted random sampling (WRS) to address dataset imbalance. RGB images of ducks were used for training. The ViT model with WRS achieved the highest accuracy, with a mean absolute error (MAE) of 0.11, mean squared error (MSE) of 0.02, and a coefficient of determination (R2) of 0.7554. These findings demonstrated that WRS significantly improved the accuracy of regression AI models, enabling precise weight estimation from imbalanced data, thereby contributing to the productivity and efficiency of duck farming.