Abstract：To solve the problem of nondestructive collection of nonlinear distortion images of soybean diseases and the mapping relationship between disease types, the digital image processing technology and the neural network reasoning mechanism were combined to propose the automatic diagnosis model of soybean disease by image correction technology.The nondestructive collection of digital images of soybean diseases was conducted by selfmade calibration templates, and the bilinear projection mapping algorithm was used to correct the geometric distortion of disease images.The shape of lesion area characteristics,the color feature and the texture feature parameter were calculated based on the multidimensional feature index. The inference rules of soybean disease were automatically obtained by applying the strong adaptability of neural network, and the automatic diagnosis model of soybean disease was established.The simulation experiments show that the distortion correction accuracy of soybean diseases is more than 99%, and the diagnostic accuracy of disease types is 98.33%. The automatic diagnosis and the accurate measurement of soybean diseases are realized.
2018, Vol. 39 
