Automatic Identification of Asian Rice Plant-Hopper Based on Image Processing

Abstract. To realize the automatic forecasting of rice plant-hoppers, a rice plant-hopper species recognition method is proposed using the windowed data of two-dimensional (2-D) Fourier spectrum of back images of insects and support vector machine. A self-made insect image acquisition device was used in field to collect the back images of rice plant-hoppers in natural environment. According to the statistical analysis of image pixels, we chose the blue component, B=140, as the color threshold to binary the original images. The segmented and morphologically filtered insect images were first obtained, and then logical AND operation was applied to them, generating the images of back regions of rice plant-hoppers. Then 2-D Fourier transform was used to obtain the 2-D Fourier spectrum of rice plant-hoppers‘ back images. We extracted the data from an l*l (l = 1, 2, ..., 9) windowed data of 2-D logarithmic spectrum to describe the features of a rice plant-hopper‘s back by analysis. Then we developed a classification model for rice plant-hoppers based on support vector machines. Experimental results showed that a discrimination model using the data from a 3x3 windowed data of 2-D Fourier spectrum can achieve a recognition rate up to over 90%for rice plant-hoppers.