Aimed at the problem of low processing speed and poor anti-interference in existing algorithms for the visual navigation, an algorithm based on improved genetic algorithm (IGA) was designed for navigation line detection. Since the trend of crop row in image was approximate to a line, two points from the bottom and top sides of image were randomly selected to code as chromosome. By multiple genetic evolutions, the highest fitness individual was chosen as the crop row line, so as to obtain the navigation line. To increase search efficiency and accuracy, IGA adopted the method of combination of probability preservation and optimal preservation strategy as the selection operator. The probabilities of crossover and mutation were adjusted automatically to improve the convergence speed and global search ability. The experimental results showed that, compared with Hough transform and Generic algorithm, IGA had the advantages of rapid detection and strong anti-interference. When the velocity of the vehicle was 0.6m/s, the maximum lateral deviation and average lateral deviation were 76 mm and 33.1 mm respectively, which could meet the requirement of farm machinery navigation.