Abstract: In view of the fact that foreign bodies in boxed dumplings seriously endanger consumers' physical and mental health, and that the traditional metal detector can only detect metals, the results cannot be visualized intuitively, we proposed a method of recognizing foreign dumplings based on LeNet convolutional neural networks(CNN) model, which can detect X-ray dumplings image containing five kinds of foreign bodies include metal balls, wires, screws, stones, and glasses. Firstly, the image of dumplings with no foreign bodies and foreign bodies was obtained by using X-ray detection equipment, then the images were processed by noise removing and contrast stretching transformation. Secondly, the CNN model is optimized, trained and validated by using batch normalization(BN) method, Softmax linear regression classifier, ReLu as activation function and Max-Pooling as downsampling method. Using the trained network model to test 100 dumpling images of without foreign bodies and foreign bodies, the experimental results show that the recognition method can accurately identify the foreign body dumplings, and the recognition rate is up to 99.78%. Finally, by extracting the traditional texture features of LBP, HOG and Gabor as the feature vectors for identifying no foreign objects and foreign objects dumplings, using BP neural network, support vector machine(SVM), k-nearest neighbor(KNN) classifier , AdaBoost classifier, Naive Bayes classifier and Decision tree classifier for dumplings recognition. The result is compared with the network model in this paper, which verify the superiority of the algorithm and the effectiveness of the extracted features. This research provides a new idea for the detection of foreign bodies in food and it is conducive to ensuring food safety.

Key words: Boxed dumplings, X-ray, Foreign body recognition, Convolutional neural networks, feature vector, Food safety