Mesh saliency has been widely considered as the measure of visual importance of certain parts of 3D geometries, distinguishable from their surroundings, with respect to human visual perception. This work is based on the use of convolutional neural networks to extract saliency maps fo large and dense 3D scanned models. The network is trained with saliency maps constructed with a fusion spectral and geometrical analysis generated measures. Extensive evaluation studies carried out, include visual perception evaluation, simplification and compression use cases. As a result, they verify the superiority of our approach as compared to other state-of-theart approaches. Furthermore, performance experiments indicate that CNN-based saliency extraction method is much faster in large and dense geometries allowing its application in low-latency and energy-efficient systems.