The invention discloses a
cascade neural network image classification method supporting class
incremental learning, and belongs to the field of
hatching egg intelligent detection, and the method comprises the following steps: S1, collecting a
hatching egg candling image, and obtaining an initial two-stage
data set and a new class
data set; s2, constructing a
cascade neural
network model in which two stages of models are connected in series; s3, each level of model integrates a feature extractor, a feature expander and a recursive linear device module, and transfer learning is carried out on a self-built
hatching egg
data set; s4, when a category is newly added, on the basis of an analytic category
incremental learning mechanism, completing new category
knowledge acquisition by recursively updating the model weight, and meanwhile, retaining the original classification capability to realize efficient incremental updating of the model; and S5, preprocessing hatching egg images to be classified, inputting the images into the two-stage
cascade neural
network model, and outputting a hatching egg
classification result. By adopting the method, the problem of insufficient fine
granularity of hatching egg classification in a hatching factory is solved, and efficient and accurate classification of the hatching eggs and low-cost iteration of the model are realized.