<p>The growing commodities economy has made innovative packaging design increasingly essential, as packaging now functions not only as a protective medium but also as a key marketing tool and representation of brand identity. Traditional packaging design methods often struggle to adequately capture consumer attention and preferences, creating a need for more effective, data-driven design approaches. To address this challenge, this study proposes PackInsightNet, a novel packaging design framework based on target identification and deep learning techniques, aimed at generating consumer-centered design insights. A new dataset, PackDesignDB, comprising 5,000 packaging images from the food, beauty, and electronics sectors, was constructed for training and evaluation. The proposed model was systematically compared with established baselines, including Faster R-CNN and Haar-Adaboost. Experimental results demonstrate that PackInsightNet achieves superior performance in visual attention area detection, with accuracies of 92.2% for food packaging, 92% for beauty packaging, and 93% for electronic product packaging. In addition, the model shows stronger alignment with consumer preferences in design element derivation, achieving matching degrees of 87%, 89%, and 86% across the respective categories. These findings indicate that PackInsightNet provides a systematic and practical approach for enhancing the visual appeal and consumer-pleasing qualities of packaging designs, offering valuable decision-support tools for designers and contributing to improved competitiveness within the packaging design industry.</p>