This study employs a deep learning strategy and a CNN model with three convolutional modules to improve the accuracy and applicability of existing mechanical performance prediction models for pervious concrete. The coarse and fine aggregate, water, admixture, cement, fly ash, and silica fume content are the eight input variables used in the model to predict the 28-day compressive strength of pervious concrete. There are 111 sample sets in the dataset, with an additional 12 sets added for robustness. Contrasted with Backpropagation (BP) brain organizations, the CNN model shows a higher coefficient of assurance (0.938) and a mean outright rate mistake of 9.13%, demonstrating predominant precision and general material.