Model and method for food material online incremental prediction

The invention discloses a model and method for food material online incremental prediction which is applied to the technical field of food material detection. The model for food material online incremental prediction is an order CNN-based multi-task model, and the method for food material online incremental prediction comprises the following steps of constructing a general CNN model combined witha specific order and multi-class classification; using the NEC-based negative sampling to replace the multi-class full negative sampling in the general model, and training the large-scale class typesin a food material supply chain; and training the noise of a non-order type by using an online incremental prediction algorithm based on the order model and the multi-classification universal model, and gradually predicting the food material types in a manner of deleting the possible types when the food material reception is finished. According to the invention, a shareable potential task matrix is used, which is very help