According to color distribution characteristics of Hlyphantria cunea larva nets, RGB color space was selected and the data differences of each channel were analyzed for net curtains, leaves and branches. Furthermore, R—B color model with the Otsu method and threshold algorithm were used to segment images. The region labeling and Freeman coding methods were adopted to calculate the area of each region. The double threshold value was determined and residual noise was removed by using the mean and standard deviation of a plurality area. According to the differences between area sizes, fine white and white regions were compensated by using improved expansion corrosion method. Experimental results showed that the accuracy of net curtain image recognition was above 85% and single image processing time was less than 40ms.