In order to reduce the influence of insufficient light and improve the extraction effect of slightly yellow leaves in the segmentation of wheat image in the field environment,a new method of wheat image segmentation combining white balance adjustment,local homomorphic filtering preprocessing and color naming algorithm based on probabilistic latent semantic analysis(PLSA)model was proposed.Firstly,the wheat image collected in the field was adjusted by the white balance to get the image without color deviation;then the local homomorphic filtering of luminance component I in HSI color model was executed for reducing the influence of insufficient light on the image;finally,the corresponding areas of green and yellow pixels from RGB image were extracted as the target area using the color name RGB value dictionary constructed by PLSA model. The results showed that the F1 value increased by 1.61 percentage points after the white balance adjustment;the F1 value of the insufficient light image segmentation increased by 12.43 percentage point after the local homomorphic filtering,and the effect of segmentation was improved obviously;the F1 values of green,slightly yellow and insufficient light wheat images were respectively 96.39%,97.29% and 96.22% by the method,which achieved good segmentation effect;compared with K-means clustering algorithm,although there were more point noise and small holes in this method,the F1 value for slightly yellow wheat image was increased by 4.42%,the overall segmentation effect was better and the stability was stronger.