Abstract
Accurate estimation of cattle weight is essential for effective farm management, health assessment, and productivity optimization. Traditional manual methods for weight estimation, however, are labor-intensive, time-consuming, and prone to inaccuracies. Recent advances in computer vision have facilitated the automation of weight prediction from image data. However, traditional regression models, such as Random Forest and Linear Regression, face challenges in capturing the complex, nonlinear relationships within image data, leading to less accurate predictions. To address these issues, we introduce CattleNet-XAI, a framework designed for both efficiency and explainability, which utilizes a custom Convolutional Neural Network (CNN). For the CNN-based approach, we incorporated advanced image preprocessing techniques, including normalization and histogram equalization, to enhance the input data quality. We compared its performance with other CNN models, the pretrained EfficientNetB3 model, and traditional machine learning methods like Random Forest and Linear Regression. For the traditional methods, we first leveraged the YOLOv5 algorithm for precise feature extraction from the cattle images. All the models were trained and evaluated on a dataset of cattle images and associated weight data, with performance measured by Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE). Our results show that one variation of our custom CNN (3Conv3Dense) model significantly surpasses other conventional models, achieving a low MAE of 18.02 kg and an RMSE of 19.85 kg, which demonstrates superior accuracy. We also present LIME visualization and error case analysis to provide insights into the decision-making process of the model. This study emphasizes the capability of deep learning, especially CNN, in automating and enhancing the precision of livestock weight estimation, offering a modern and effective approach to cattle management.