As socio-economic development and modernization accelerate, contemporary humans face increasing stress and challenges stemming from changes in the external environment, such as occupational competition and the educational system, as well as from the contradictions between individual expectations and objective reality. This context has led to the widespread prevalence of non-positive emotions like anxiety, depression, and anger, posing an urgent issue for social mental health. In the field of art and design, emotion visualization research provides an innovative approach to address this issue. This study investigates the expression mechanisms of non-positive emotions in decorative pattern design and explores theories and methods to enhance emotional resonance through visual semantic transformation. Utilizing the Double Diamond Design Model and an empiricist methodology, the research employs techniques such as video recording and AI data analysis to develop a system for the daily collection and visualization of non-positive emotions. Additionally, the study integrates Convolutional Neural Networks (CNN) and Generative Adversarial Networks (GAN) to collect and process relevant image data. The models are then fine-tuned based on preliminary results to improve the quality of the generated images.