Detection of embryo based on independent components for kernel RGB images in maize

The characteristics of maize embryo are important agronomic traits of maize, which are mainly measured by hand. In order to implement automatic extraction of the features of maize embryo by computer vision and image processing method, a new method for measuring embryo based on independent component analysis (ICA) was developed, and its testing model was also established. RGB images of 40 maize kernels were scanned with 600 DPI resolutions using a flat scanner. After segmenting embryo part from other parts of maize kernels using the independent component with the maximum entropy, the embryo area and the other 8 embryo characteristics of these maize kernels were extracted. Compared with the manual measured results as ground-truth reference, the area error rate for our proposed method was 0.7%, and determination coefficient of the manual regression to the predicted reached 0.984, and error rates of other 8 characteristics were generally below 2%. When compared with citations of those based on the region growing of color models, our proposed method significantly increased detection accuracy. Obviously, the proposed method based on ICA is accurate and reliable, and can be used to automatic detection of maize embryo.