Study of feature extraction based visual invariance and species identification of weed seeds

Aiming at the seeds of biological stability genetic character, we present a method to feature extraction based on visual invariance. By analyzing the weed seeds and hilum shape characteristics, nine shape features and seven moment invariants of visual invariance were extracted. Back Propagation (BP) Neural Network was used to identify weed seeds, and the relationship between the change of features dimension, recognition rate and recognition time was analyzed. The experimental results prove that the proposed features have good visual invariance. The recognition rate of the 16 dimensions eigenvalue is up to 96%. The method could meet the requirement of the detection of weed seeds, and is rapid and accurate.