Automated corn kernel quality inspection using machine vision

A machine vision system was designed and developed for automatically inspecting corn kernels. This system consisted of a motor driven kernel delivery apparatus, two light-beam-based object position sensors, a strobe light, a control circuit, a video camera and a PC computer with image processing hardware and software.; Stroboscopic illumination was used to virtually eliminate image blur due to motion. Images were acquired with an image processing system. The image frame grabber was set to enable image capture when an incoming trigger signal was received. The horizontal drive and the vertical drive signals of the video camera were utilized to generate a strobe-triggering pulse which was synchronized with the start of the next image field after the frame grabber was triggered. Synchronization of digitizing video images under strobe lighting was 100% error-free. A vibratory bowl feeder was used for kernel singulation to present individual kernels to the camera. The success rate of the kernel singulation operation was 97.7%.; Machine vision inspection was developed for whole-versus-broken kernel percentages, size grading, and crown-end of kernel discrimination.; Corn kernels were inspected using machine vision for whole-versus-broken percentages. A four-layer neural network classifier was used with a total of 34 shape features as its inputs. On-line tests of eight corn samples showed success classification rates between 87% and 95% in comparison to human inspection. The processing time of the machine vision classifier was between 0.94 to 1.04 second per kernel and the entire cycle time was about 3.9 seconds per kernel including placement and processing time.; A corn kernel size grading algorithm was developed using machine vision. Measurements of three dimensions of corn kernels were obtained using on-board hardware-based image processing operations. A linear discriminator classifier was employed with kernel length, width, thickness, and projected area as the inputs. The algorithm accuracy was evaluated by the comparison of the results of machine vision to mechanical sieving. The system provided correlation coefficients from 0.86 to 0.89 between predicted and actual sieving results. The processing time was between 2.03 to 2.09 seconds for each kernel.; A machine vision algorithm was developed to discriminate the shape characteristics of the crown end of corn kernels. Corn kernels were classified as rounded/flat and dent based on their crown end shape. Dent corn kernels were further classified into smooth dent and non-smooth dent kernels. Kernel crown end images were acquired through a 45{dollar}spcirc{dollar} plane mirror under the camera. Image enhancing and filtering were employed to reduce noise. Multiple line profiles were used to obtain the needed three-dimensional information from the analysis of the different shape patterns of the one-dimensional profile signals. The number of zero-crossings in the gradient of the line profiles was used as the classification feature. The algorithm provided an average accuracy of approximate 87% when compared to human inspection. The processing time was between 1.53 and 1.76 seconds per kernel.