The detection of the type of soil surface where a robotic vehicle is navigating on is an important issue for performing several agricultural tasks. Satisfactory results in activities such as seeding, plowing, fertilizing, among others depend on a correct identification of the vehicle environment, specially its contact interface with the ground. In the this work, the implementation of a supervised image texture classifier to recognize five different classes of typical agricultural soil surfaces is presented and analysed. The sensing device is the Microsoft Kinect for Windows V2, which allows to acquire RGB, IR and depth data. Only IR and depth data were used for the processing, since color information becomes unreliable under different illumination conditions. Two data acquisition modes allowed to validate and to apply the system in real operation conditions. The accuracy of the classifier was assessed under different configuration parameters, obtaining up to 93 percent of success rate, in ideal conditions. Real field conditions were simulated by placing the sensor over a moving wagon, obtaining up to 86 percent of success rate, showing in this way the usability of a low cost sensor such as the Kinect V2 for agricultural robotics.