Study On Rice Injured Kernel Detection Method Based On Image Processing

Rice is one of the important crops in China’s grain production. During storage, rice’squality is susceptible to affect by various factors, the worm’s remnants can also pollute therice. Therefore, it’s need to detect the injure kernel to reduce grain losses during storage.Traditional injure kernel detected by the naked eye observation, test results are inaccurateand subjectivity. Injure kernel detection technology based on image processing with highidentification accuracy, speed, High-volume processing，it has a broad applicationprospects.This paper presents a new method for injure kernel recognition based on imageprocessing. Calculate the image boundary coordinate point’s value, and then compute thestatistical value as characteristic parameters, input them to classifier, implement the injurekernel recognition. Extract the individual injure kernel of the image, using a variety oftraditional threshold method to segment the image, comparative analysis which is mostappropriate binarization method, using morphological methods to denoising binary image,removal of redundant information, and then extracted using morphological erosion injurekernel boundary, and boundary coordinate values along the freeman chain code direction totake, not synchronized long conditions, the use of angle calculation formula to calculate theboundary coordinate point vector angle value sequence, statistics, their characteristicparameters: mean, variance, third moment, entropy, and finally using the Fisher classifierand Bayesian classifier to classify the characteristic parameters of the input test sampleidentification, and comparative analysis of the two classifiers to identify the effect. On thisbasis, the Fisher classifier discriminant point selection method to improve, and thencompared with the traditional classifier to recognize the effect of the classification results,and the contrast is not synchronized long conditions. The experimental results show that:the improved Fisher classifier in step length2is most effect, injure kernel’s recognition ratereaches95.65%.