|Cetin, A. Enis|
Submitted to: Computers and Electronics in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/20/2012
Publication Date: 6/1/2012
Publication URL: http://naldc.nal.usda.gov/download/55218/PDF
Citation: Yorulmaz, O., Pearson, T.C., Cetin, A. 2012. Detection of fungal damaged popcorn using covariance features. Computers and Electronics in Agriculture. 84:47-52. Interpretive Summary: Popcorn is especially vulnerable to fungal infections at harvest time because drying of this grain must be gradual, otherwise it will not pop. One type of fungal infection is called "blue-eye" since it causes a small blue-gray blemish on the germ of the kernel. These infected kernels can have a very undesirable off taste, resulting in lower consumer acceptance of this snack food. The blemish associated with blue-eye damaged popcorn is so small that current commercially available sorting machines are not able to detect and remove these infected seeds. In this study, image processing techniques were developed to detect blue-eye damaged popcorn with accuracies over 95%. While these techniques are fairly advanced, they can be implemented on high speed sorting machines to detect and remove blue-eye damaged popcorn. Furthermore, the technique is adaptable to detecting blemishes on other commodities, such as wheat, beans, and corn, such as those caused by insect damage and other species of fungi.
Technical Abstract: Covariance-matrix-based features were applied to the detection of popcorn infected by a fungus that cause a symptom called “blue-eye.” This infection of popcorn kernels causes economic losses because of their poor appearance and the frequently disagreeable flavor of the popped kernels. Images of kernels were obtained to distinguish damaged from undamaged kernels using image-processing techniques. Features of kernel images were extracted from the covariance matrices of the pixel properties of the acquired images. The covariance matrices were formed using different property vectors that consisted of the image coordinate values, their intensity values, and the first and second derivatives of the vertical and horizontal directions of different color channels. Support vector machines (SVM) were used for classification purposes. An overall recognition rate of 96.0% was achieved using these covariance features. The recognition rates for the undamaged and damaged kernel images were 97.6% and 95.1%, respectively.