Stripe rust and powdery mildew are two kinds of the most destructive foliar diseases in wheat grown and have a significant impact on the production of wheat. They differ in the pathogenesis and prevention, so it is necessary to distinguish and identify the two diseases, which can help to improve the development of agricultural information technology and automation. For the problem that stripe rust and powdery mildew lesions are similar in color features, as well as the shape features are not obvious, it is difficult to distinguish each disease using traditional features. However, the spots of two diseases have a significant difference in the trend of the directional distribution of the leaves of wheat. With respect to this characteristic, this paper proposed an orientation coherence feature based on the directional kernel convolution (DKC) method, and applied this feature to the identification of stripe rust and powdery mildew. In detail, the DKC method used several directional kernels to convolve with image to generate direction maps and edge maps which were used to calculate the directional distribution histogram. Then, the standard deviation of the histogram was used to describe the consistency of the directional distribution in the image and regarded as an orientation coherence feature. The orientation coherence feature could be used to describe the orientation dispersion of disease. If the orientation coherence feature of a sample was large, the disease of the sample was more likely to be stripe rust. Otherwise, it is more likely to be powdery mildew. To verify the effectiveness and the noise resistibility of proposed orientation coherence feature, two experiments were performed, and the results were compared with edge orientation histograms (EOH) based method. Firstly, the DKC and the EOH based orientation coherence feature were extracted for synthetic images with different noise levels. The results inferred that the noise had little effect on the DKC based orientation coherence feature which could best describe the directional information of noise images than traditional method. Secondly, the experiment for identification of stripe rust from powdery mildew indicated that the proposed orientation coherence feature could distinct the wheat stripe rust and powdery mildew much better than EOH based feature, and the accuracy could be up to 99%. In addition, the proposed orientation coherence feature could be treated as a new description for other plant diseases and it provides a new idea for crop recognition and detection, which is important in the field of computer vision based technology for agriculture.