Vision-based guitar chord recognition offers a promising alternative to traditional audio-driven methods, particularly for silent practice, classroom environments, and interactive learning applications. While existing research predominantly relies on full-frame image analysis using 2D convolutional networks, the use of structured hand landmarks remains underexplored despite their advantages in robustness and computational efficiency. This study presents a comprehensive comparative analysis of three one-dimensional convolutional neural network architectures—CNN-1D, ResNet-1D, and Inception-1D—for classifying seven guitar chord types using 63-dimensional static hand-landmark vectors extracted via MediaPipe Hands. The methodology encompasses extensive dataset preprocessing, targeted landmark augmentation, Bayesian hyperparameter optimization, and stratified 5-fold cross-validation. Results show that CNN-1D achieves the highest mean accuracy (97.61%), outperforming both ResNet-1D and Inception-1D, with statistical tests confirming significant improvements over ResNet-1D. Robustness experiments further demonstrate that CNN-1D maintains superior resilience under Gaussian noise, landmark occlusion, and geometric scaling. Additionally, CNN-1D provides the fastest inference and most stable computational performance, making it highly suitable for real-time or mobile deployment. These findings highlight that, for structured and low-dimensional landmark data, simpler convolutional architectures outperform deeper or multi-branch designs, offering an efficient and reliable solution for vision-based guitar chord recognition.