The paramountcy of English in the contemporary global landscape necessitates the enhancement of English language proficiency, especially in academic settings. This study addresses the disparate levels of English proficiency among college students by proposing a novel approach to evaluate English text readability, tailored for the higher education context. Employing a deep learning (DL) framework, the research focuses on developing a model based on convolutional neural networks (CNNs) to assess the readability of English texts. This model diverges from traditional methods by evaluating the difficulty of individual sentences and extending its capability to ascertain the readability of entire texts through adaptive weight learning. The methodology's effectiveness is underscored by an impressive 72% accuracy rate in readability assessment, demonstrating its potential as a transformative tool in English language education. The application of this DL-based text readability evaluation model in college English training is explored, highlighting its potential to facilitate a more nuanced understanding of text complexity. Furthermore, the study contributes to the broader discourse on enhancing English language instruction in higher education, proposing a method that not only evaluates text comprehensibility but also aligns with diverse educational needs. The findings suggest that this approach could significantly support the enhancement of English teaching methodologies, thereby promoting a deeper, more accessible learning experience for students with varying levels of proficiency.