This paper proposes a novel approach for soil estimation and classification in autonomous excavation exploiting key features from bucket velocity and energy signatures considering the interaction dynamic of the digging process. A real-time energy-based method is proposed for estimating the dynamic friction force arising during soil-tool interaction. The method relies on a novel technique for measuring the force and displacement variables which allows the on-line determination of the bucket velocity and dissipation energy along the full excavation profile. It is shown that these measurements can be effectively used for the on-line identification and classification of different types of soil encountered during the excavation process. The proposed method is insensitive to noise and can be easily implemented in practice using an excavation arm and hydraulic actuators. Various experimental results are presented supporting the practicality of the proposed method.