Images play an important role in information communication and information preservation, which can meet the needs of the construction of visual learning resources in the education industry. This paper constructs a deep convolutional generative adversarial network model for text-generated images based on conditional enhancement and attention mechanism, which adopts a bidirectional long and short-term memory network to extract textual features, and enriches the feature information of the text through the conditional enhancement module. Subsequently, textual and visual features are fused and output in the generative adversarial network, and the detailed adjustment of the output features is accomplished based on the attention mechanism to generate educational visual image resources containing important features of textual descriptions. The text-to-generateimage method proposed in this paper obtained excellent IS scores on both CIFAR10 and CelebA datasets, with their mean values higher than 7. The visual learning resources, designed in conjunction with the content of the mathematics curriculum, help to enhance the multifaceted mathematical literacy and learning effectiveness of secondary school students, and have gained the approval of the majority of students, with an average satisfaction score of around 4 on the student questionnaire. The text-image generation method in this paper provides new ideas for the construction of visual learning resources for mathematics, which helps learners to better understand mathematics learning methods and concepts
