Automated Recognition of Corn Embryos for Selective Breeding Tanner Holte Michael Davies Dr. Baskar Ganapathysubramanian
Corn Embryos
Selective Breeding Project
Machine Learning Supervised Approach Data Model Prediction
Creating Training Set Point in Polygon Manual 1,529 sets created
Point-in-Polygon Algorithm Recreate the polygon using data on vertices Iterate through each pixel in the image If the pixel falls on the inside of the lines, color the pixel white, else color it black
Training Convolutional Autoencoder Iowa State’s CyEnce Cluster Compares colors of pixels Generates model based on patterns Once the model is trained, it can be applied to unlabeled images to generate the label
Testing -3.66% -4.03% +49.34% -8.86% -4.51% -8.44%
Consequences of the Project Faster labeling Labels are generally more consistent and accurate
Personal Experience Difficulty of understanding machine learning Use of a new language, Python Improving my own adaptability