<p>As an endogenous fracture of coal-rock, cleat has an important effect on the permeability and mechanical properties of coal-rock and is a key parameter in coal seam logging evaluation. Under high-stress conditions in deep coal formations, cleats exhibit low aperture angles and relatively small scales, resulting in weak logging responses. The effectiveness of conventional theoretical methods built based on electrical and acoustic parameters, such as resistivity, invasion characteristics, wave velocity, and transit time ratio, are limited. The most effective approach is to evaluate jointing through multivariate regression methods that jointly analyze multiple physical parameters, but the results still fail to meet the requirements. Although cleat apertures are narrow, their increasing development enhances the heterogeneity of coal rock structure that is often ignored in modeling. This study proposes using fractal theory to characterize the degree of coal heterogeneity. By applying fractal analysis to the logging curves, heterogeneity features are extracted and then integrated with conventional logging responses. These features are then combined and input into a Deep Belief Network to establish a logging prediction method for cleat development degree in deep coal seams. The results demonstrate that, in deep coal formations, as cleat development degree increases, the fractal dimension of coals increases. Extracting fractal features from logging curves and integrating them into the deep learning model improves the accuracy of the logging prediction of cleat development degree. Based on the combined inputs of logging curves and fractal characteristics, the deep Belief network algorithm reduces the average relative error of cleat development degree prediction from approximately 37% to 8%.</p>