In large-scale image classification involving complex design semantics, designers often face challenges in balancing structured data processing, semantic alignment, and image generation accuracy. While general-purpose large models possess cross-domain capabilities, they often exhibit insufficient generalization in scenarios requiring specific design styles, functional characteristics, and contextual applications. To address this issue, this paper proposes an optimized multimodal classification approach tailored to design-oriented images. Specifically, we introduce a curated dataset comprising award-winning works from the RedDot Design Award, which has been filtered and categorized to establish a classification system centered on design semantics, style, and functionality. A specialized multimodal classification model is then constructed to overcome limitations in accuracy and domain-specific performance seen in existing methods. Experimental results demonstrate that our approach significantly improves the precision and professionalism of design image classification, offering new perspectives and technical frameworks for AI applications in the design field.