Segmentation of the Germ from Corn Kernel CT Image Based on the Mask-RCNN

The drying cracks were predominantly generated in the endosperm and there was no crack in the germ, thus the germ should be segmented from corn kernel computed tomography (CT) images for studying the formation and propagation of the drying crack. The CT images of the corn kernel were often manually segmented, which was time-consuming and poor in accuracy. A segmentation model based on a Mask region convolutional neural network was developed for segmenting the CT images of the germ. The average precision and mean of pixel accuracy were 1 and 0.985 respectively, and the processing rate was 3-4 images/s, which indicated that the segmentation model could segment effectively the germ with high segmentation efficiency.