Resumen
In order to improve the accuracy and efficiency of the methods that are used for the detection of impurities in and the breakage rate of harvested corn grains, we propose a classification and identification method using a feature threshold and a backpropagation (BP) neural network that is based on a genetic algorithm. We also constructed a continuous single-layer discrete tile detection system for application to harvested corn grains containing impurities and broken kernels. We conducted an evaluation of the proposed approach with a three-factor and three-level orthogonal experimental design. By setting the working parameters, we realized the continuous single-layer discrete tiling of the grains and 50 grain materials were collected on average in a single picture. In the static test, the error between the system monitoring value and the manual detection value was small, the maximum absolute errors of the breakage and impurity rates were 2.16% and 1.03%, and the average time that was required for each image recognition was 1.71 s. In the experimental environment, the maximum absolute error values of the breakage and impurity rates were 3.48% and 1.78%. The system?s identification accuracy and processing time meet the requirements of the online detection of corn characteristics in grain harvesting.