Abstract: To satisfy the requirement of real-time processing and identification accuracy, a method based on iterative-RELIEF relevance vector machine was proposed. In this method, information of image samples is brought into the module of iterative-RELIEF algorithm, which exports a weight for every feature. Then, the information of image samples with weights is brought into the training module of relevance vector machine (RVM). As a result, an image classifier is made, which can be used to predict the classes of unknown pixels of a image containing a cucumber. In the experiment, the rate of right identification is up to 80% or more, while the rate of false identification is lower than 27%, and the ratio of the two is up to 3.0 or more.  

Key words: harvesting robot, image identification, iterativeRELIEF, relevance vector machine（RVM）