Study and Design of Corn Hybrid Variety Identification System by Computer Vision

摘要

Variety purity is the most important quality index of corn hybrid seed, corn variety purity has a great influence on corn production and quality, but variety identification is very difficult. Electrophoretic method was a new variety identification method developed in recent years. It was rapid, simple, accurate and reproducible. Because the electrophoretogram recognition relied on human completely, this demanded users having high technique level, It was difficult to spread and apply. In this paper, a computer vision system was developed for corn hybrid variety identification by corn seed electrophoretogram processing and analysis. The method of qasi-wavelet edge multiscale detection based on Gaussian filter and vector differential operator was used to extract feature of characteristic bands successfully. A BP neural network classifier was designed for electrophoretogram of corn hybrid classification, the test results demonstrated that classification accuracy is 96%. Based on the proposed theories and methods, a set of applied software was developed for corn hybrid variety identification. It was tested with 30 kinds of corn hybrid, the results demonstrated that the system provided an accuracy of 95% compared to experts inspection.