Click on “Download PDF” for the PDF version or on the title for the HTML version.

If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.

Abstract. Kernel broken rate has been determined as one of the main technical indicators for a good corn combine harvester. Thus, determining kernel broken rate is extremely helpful for adjusting harvesters to the best working condition. Previous findings suggest that machine vision provides a good alternative to determine corn kernel broken rate by extracting shape, color and textural features or combining these features together. However, errors may arise when object distance or light condition changes. This paper presents a research on the potential of using only contour-based features to discriminate broken kernels from whole kernels. The contour of a corn kernel was sampled with interval angle into a one-dimensional sequence of centroid-contour distance. The normalized sequence as contour-based feature was put into the neural network for classification. The result shows that the method presented in this case has an average accuracy of 91.57% and is promising to be optimized. Moreover, the method is resilient to illumination changes and scaling factors. This research also assesses the effect of different number of sampling points and directions of kernels, laying the foundation for further optimization and future use.