Segmentation of aerial images of crop is an important method for the crop prediction and analysis based on plant-protecting unmanned aerial vehicles (UAVs). The problem is that the accuracy of traditional crop-image segmentation based on a super-green feature is not high enough and, moreover, the corresponding algorithm can reveal great differences in the precision for different crops. In this study, we use a Markov random-field segmentation model for the aerial images taken from the plant-protecting UAVs. A standard K-mean algorithm is employed to mark the observation field and the model parameters are estimated by an expectation-maximization method. For deriving the segmentation results, an iterated conditional-modes algorithm is applied. The experimental results testify that the approach of Markov random fields can segment efficiently the cotton-plant images obtained from the UAVs. Then the segmentation accuracy becomes higher and the algorithm adaptability better. However, the processing time increases and the real-time regime can be questioned when the number of segmented image pixels exceeds a critical value.