IMPLEMENTING A COMPUTER VISION SYSTEM FOR CORN KERNEL DAMAGE EVALUATION

Abstract

A computer vision system was developed for evaluation of the total damage factor used in corn grading. Majorcategories of corn damage in the Midwestern U.S. grain market were blueeye mold damage and germ damage. Sevenhundred twenty kernels were obtained from officially sampled Federal Grain Inspection Service (FGIS) corn samples andclassified by inspectors on the Board of Appeals and Review. Inspectors classified these kernels into blueeye mold,germdamaged, and sound kernels at an 88% agreement rate. A color vision system and lighting chamber were developedto capture replicate images from each sample kernel. Images were segmented via input of red, green, and blue (RGB) valuesinto a neural network trained to recognize color patterns of blueeye mold, germ damage, sound germ, shadow in sound germ,hard starch, and soft starch. Morphological features (area and number of occurrences) from each of these color group areaswere input to a geneticbased probabilistic neural network for computer vision image classification of kernels into blueeyemold, germ damage, and sound categories. Correct classification by the network on unseen images was 78, 94, and 93%,respectively. Correct classification for sound and damaged categories on unseen images was 92 and 93%, respectively.