Feature extraction via dimension reduction is an important step in image recognition. Due to the high complexity of the crop disease leaves and the corresponding lesion images, caused by various observed angle, locality and illumination in the real filed scene, many classical dimensional reduction and feature extraction algorithms are not effective to recognize the crop diseases. In this paper, based on discriminant locality preserving projections（DLPP）, a crop leaf recognition method is proposed for crop diseased leaf identification. Firstly, GrabCut algorithm is used to segment the background of the collected leaf image, and then the watershed algorithm is employed to segment the image to obtain the lesion image. Next, DLPP is introduced to project the segmented lesion image into the low-dimensional discriminant space to get the classification features. Finally, K-nearest neighbor classifier is adopted to recognize the disease category. The experimental results on the image dataset of apple leaf diseases show that the method is effective and feasible.