In farmland with complex environment, guidance line recognition of agricultural machinery based on machine vision is subjected to illumination variation, weed noise, etc. In addition, the conventional path detection algorithms have the drawbacks of low processing speed and poor anti-interference. The visual navigation path detection under natural environment was conducted. Firstly, to reduce the influence of illumination changes on the quality of image segmentation, Cg component was constructed on the base of YCrCb color mode and the 2Cg—Cr—Cb factor was selected to preprocess the image. Secondly, the clustering segmentation of the image was performed based on improved K-means algorithm to achieve the respective clusters of soil and green crop information. Then, the weed interference information in the binary image was eliminated by morphological filtering algorithm so as to obtain the complete and clear crop information. Finally, according to the characteristics of the crop rows in the image, linear equation constraints of crop rows were established. An algorithm of crop lines detection based on particle swarm optimization (PSO) was designed. Experiment results showed that the image segmentation based on 2Cg—Cr—Cb gray image can effectively identify crop from soil background under different illumination conditions. The segmentation images were less affected by change of illumination and no background noise was contained. The guidance line recognition method based on PSO can quickly and accurately detect the navigation line. Furthermore, it had good fitness for different crops and nice adaptability for different crop growth stages in the farmland. Compared with conventional guidance line recognition algorithms, the designed algorithm had the advantages of high speed and good robustness.