<p>Engineering experience indicates that discontinuities are crucial in the mechanical properties and behavioral characteristics of rock masses. The systematic investigation of discontinuities constitutes an essential component in engineering geological exploration. The acquisition of discontinuity parameters, particularly joint frequency, remains a critical research focus in rock engineering. This study reframes the extraction of joint frequency from rock mass exposure images as a “variable-to-image” regression problem within deep learning frameworks. A dataset consisting of 215 discontinuity measuring points was collected from 11 distinct rock engineering sites using the ShapeMetriX3D system. A convolutional neural network logistic regression (CNN-LR) model was developed to capture the characteristic features of rock mass exposure images, and perform linear fitting between these features and joint frequency. Additionally, 64 testing samples were used to evaluate the performance of the proposed method. The results indicate that the proposed method can effectively estimate joint frequency from rock mass exposure images. However, there is a certain discrepancy between the true joint frequency and the predicted values. The errors produced fall within an acceptable range, and strongly correlate with the distribution of the learning samples. The main contribution of this study is the proposal of a deep learning-based framework for the automatic estimation of joint frequency from rock mass exposure images. The performance of the proposed method can be further improved by increasing the quantity and quality of learning samples.</p>