This paper describes a recognition method for plant pipeline direction, the rotational angle in the vertical plane and in depth. A neural network is applied using the vision method for the plant maintenance robot. For this purpose, a fiber grating method has been employed. The fiber grating is made of optical fiber through which laser beams form discrete dot patterns without any mechanical scanning method. Thus, the proposed method has the advantage of a shorter processing time compared to that of other methods. Although the proposed method does not provide a fine analysis of image data like the common scene analysis, it is suitable for obtaining rough information from the operating environment of robots. Experimental results are included to verify the proposed method.