In pursuit of carbon neutrality and advancing energy-efficient practices within the steel and coking industries, the traditional cokemaking process is progressively evolving towards intelligence, with coke quality prediction emerging as a pivotal technology at its core. Nevertheless, the intricacy of the coke production process presents a formidable challenge in accurately forecasting it. This study is the first to propose a novel image expression-driven modeling approach that transforms the numerical coal properties into image expressions and uniquely integrates the utilization of the Convolutional Neural Network (CNN) for predicting coke quality including Coke Strength after Reaction (CSR) and Coke Reactivity Index (CRI). Utilizing the collected 729 Chinese coal properties and corresponding coke quality indexes, the dimensionality reduction technique was employed to transform numerical coal properties into image expressions. A convolutional neural network combined with the random forest model was subsequently developed for learning and prediction, with its performance evaluated on Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R2 metrics. The results suggested that the proposed groundbreaking model outperformed existing numerical properties-based coke quality prediction models and typical regression models, achieving MAE of 1.57, RMSE of 2.22, and 0.86 for R2 metric, along with MAE of 1.82 and RMSE of 2.42 as well as 0.91 for R2 metric in CRI and CSR prediction, respectively. Furthermore, a comprehensive analysis was also undertaken to identify the pivotal factors influencing the efficacy of coke quality prediction based on the proposed approach.