Abstract: It is suggested to use a classication recognition method for worm damaged and intact particles, based on an air-coupled ultrasonic technology. Firstly, record the physical attributes of corn particles, such as elastic modulus, poisson's ratio and the density, which can then be used to calculate the ultrasonic velocity. With that information, determine the frequency of the excitation signal based on the needed detection precision. Secondly, use MATLAB to analyze the collected ultrasonic signal data of the two types of seeds and analyze the impact of the seed thickness and position on the ultrasonic response characteristics. From there, establish the four kinds of recognition models, including K nearest neighbor (KNN), soft independent modeling of class analogy (SIMCA), Fisher linear discriminant (LDA) and decision tree (DT), and test the identification models performance. The results of this study clearly illustrate that the hole depth, embryo thickness and bearing of the corn different, the reflection degree of ultrasonic signal on the seed surface and its propagation path are different, then the ultrasonic signal attenuation degree, the amplitude of received signal and the distribution of sample points in the feature space are different. The four pattern recognition models all could identify both holed and intact particles, and the KNN model has the best performance, with an accuracy rate for damaged and intact particles of 98% and 100%, respectively and an error band of 2% and 0, respectively. The results prove that using the air-coupled ultrasonic technology makes it possible to accurately assess the moth-eaten corn seed testing.