Class increment continuous learning method based on classification layer unification and improved EWC

The invention discloses a class increment continuous learning method based on classification layer unification and improved EWC, and the method comprises the steps: training a neural network through a training set, obtaining an identification model, and identifying the class of an object in a picture through the identification model, the training comprises old task training and at least one new task training, and in the new task retraining process, the classification of the object in the new task training is completed. Model parameters of an old task are used as base points, and offset of the parameters is limited. According to the class increment continuous learning method based on classification layer unification and improved EWC, the new class deviation problem is solved, and the average accuracy of the model is improved. 本发明公开了一种基于分类层统一化和改进EWC的类增量持续学习方法，通过训练集对神经网络进行训练，获得辨识模型，通过辨识模型辨识图片中事物的类别，所述训练包括旧任务训练和至少一次新任务训练，在新任务再训练过程中，以旧任务的模型参数为基点，限制参数的偏移。本发明公开的基于分类层统一化和改进EWC的类增量持续学习方法，解决了新类别偏向问题，提升了模型的平均准确度。