Sequential Pattern Learning of Visual Features and Operation Cycles for Vision-Based Action Recognition of Earthmoving Excavators