95.8%。In order to improve the accuracy and efficiency of weed identification, a method for cotton-weed recognition was proposed by using the combination technique of features optimization and least squares support vector machine (LS-SVM). After a series of image processing such as graying, filtering and threshold segmenting, six geometric shape features and seven Hu moment invariants were extracted from the single plant leaf. Then, using particle swarm optimization (PSO) algorithm, the extracted features were optimized in order to reduce the size of the training data sets. Finally, the weed was identified by using the trained classifier. The experimental results indicate that this method can effectively compact feature subset and maintain a higher accuracy than using the original feature set, the average correct identification rate is 95.8%. 