In order to guide the robots to harvest citrus fruit in complex natural scenes,a method based on image processing is developed to identify citrus fruit in the tree canopy.Due to the complexity of natural environment and the deficiency of identification of citrus fruit by the traditional methods,five narrow band filters in the visible and near-infrared region are selected to capture five images.Spectral angle mapper (SAM) algorithm is attempted to process these images.Experiment results show that citrus can be recognized with an accuracy of 96% in different aspect of illumination angle and intensity.This work demonstrates that spectral imaging technique based on filters combining with SAM algorithm can be used effectively to identify citrus fruit in complex natural scenes.