This paper deals with a recognition method of plant pipeline direction (rotational angle in the vertical plane and in depth) by the neural network applications using the vision method for the plant maintenance robot. For this purpose, a fiber grating (F. G.) method is employed in this study. The F. G. is itself made of optical fiber, through which the laser beams form discrete dot patterns without any mechanical scanning method. Thus the proposed method has the following advantage: the processing time is generally shorter than that of other methods. Although the proposed method does not give a fine analysis method of image data like the common scene analysis, it is suitable to obtain rough information on the working environment of robots. This paper introduces the applicability of the neural network to the discrete dot pattern by F. G. for the recognition of plant pipeline direction (rotational angle in the vertical plane and in depth). Some experimental results are also shown here to verify the proposed method.