Multimodal Aspect-Based Sentiment Analysis (MABSA) is an important task. Its goal is to predict the emotional polarity. Existing methods mainly realize the association between aspects and multimodal features through question and answer modeling, but there are still limitations in the dynamics of modal fusion, the accuracy of aspect-modal alignment, and the use of external knowledge. Therefore, this article proposes MACA (Multimodal Aspect-Based Sentiment Analysis with Dynamic Adaptive Gating, Contrastive Alignment, and LLM Augmentation): A unified framework for MABSA. In order to solve the problem of modal difference and aspect evidence mismatch at the sample level, a method combining dynamic adaptive gating (DAG), aspect-modal contrast alignment (ACA) and LLM prior dual path integration is proposed. The MACA model maintains consistent modal preferences. It adaptively allocates text/image/description/prior contributions through learnable gating, and uses contrastive learning to reinforce consistency between aspect and evidence region. The decision-making layer introduces temperature calibration and confidence modulation, amplifying their effects when the prior is reliable and automatically contracting when the prior is uncertain. On Twitter-15/17, MACA outperforms the strong baseline in multiple tasks. Ablation shows that the three modules reinforce each other, significantly improving the accuracy and robustness of the model.