In order to realize the rapid, comprehensive and nondestructive identification of imperfect grains of wheat, image processing and CNN recognition methods can be applied by collecting the imperfect grains and perfect grains mentioned in the national quality standard of wheat. Image processing can make the output image have better effect, which is convenient for image analysis and recognition. Image preprocessing includes image acquisition, graying, median filtering, image segmentation and so on. The classical classification network LeNet-5 in convolutional neural network (CNN) takes the preprocessed image as the input and adds batch normalization (BN). BN algorithm can speed up the decline speed of training gradient, increase the convergence speed of the model, and increase the stability of the model. It can recognize the image and evaluate the performance with the accuracy of the test set. It avoids complex feature extraction steps and effectively improves the recognition rate of wheat grains, which is of great significance to the intelligent detection and recognition of wheat.