Accurate estimation of carbon stocks is vital for management. While field-based methods often face limitations in coverage, remote sensing approaches present a more effective alternative. Recent advancements in deep learning have enabled the application of Convolutional Neural Networks (CNNs) to analyze high-resolution drone imagery in carbon stock estimation. This study assesses the performance of VGG-16 and ResNet-20 for regression tasks, employing Optuna for hyperparameter optimization to enhance prediction accuracy. The experimental findings reveal that VGG-16 achieved an R² score of 0.645, with lower RMSE and MAE values than ResNet- 20. Furthermore, the study highlights significant challenges, such as dataset imbalance and feature extraction in regions with high carbon stocks. Future research may investigate hybrid learning techniques, ensemble models, and multispectral data fusion to improve model estimation accuracy and generalization.