Submitted to: Journal of Food Measurement & Characterization
Publication Type: Peer reviewed journal
Publication Acceptance Date: 10/11/2012
Publication Date: 10/27/2012
Publication URL: naldc.nal.usda.gov/download/57493/PDF
Citation: Pearson, T.C., Moore, D., Pearson, J. 2012. A machine vision system for high speed sorting of small spots on grains. Journal of Food Measurement & Characterization. 6:27-34. DOI: 10.1007/s11694-012-9130-3. Interpretive Summary: A new type of automatic electro-optical sorting system was developed to identify and remove grains with small spots, or blemishes, on them. There currently is no commercially available system that can separate grains having a small blemishes on their surface. However, several food processors has requested development in this area as it would improve food quality and safety. The newly developed system was tested for removing popcorn with a defect called blue-eye, which is caused by a fungus and appears as a small blue blemish on the germ of the kernel. This system was able to remove 89% of the blue-eye damaged popcorn kernels while only rejecting approximately 6% of the un-damaged kernels. Blue-eye infected popcorn results in off tastes so the system will find use with popcorn processors across the country. The system can also be used to separate grains with other types fungal damage or insect damage. Resulting in higher quality and safer food products.
Technical Abstract: A sorting system was developed to detect and remove individual grain kernels with small localized blemishes or defects. The system uses a color VGA sensor to capture images of the kernels at high speed as the grain drops off an inclined chute. The image data are directly input into a field-programmable gate array (FPGA) that performs image processing and classification in real time. Spot detection is accomplished by a combination of color information and a simple, non-linear spatial filter that detects small dips in pixel intensity along an image line. Color information is combined with spatial filtering to achieve a high level of accuracy. Testing was performed on popcorn with blue-eye damage, which is characterized by a small blue blemish on the germ. A two-camera system was developed to inspect the opposite sides of each kernel as they slide off the end of a chute. The chute was designed such that the kernels slide down the chute without tumbling, increasing the probability that a spot will be in the field of view of one of the cameras. The system's accuracy is 89% identification of blue-eye damaged kernels with a 6% false positive rate. The throughput is approximately 180 kernels per second, or 100 Kg/hr.