Image recognition method for deep learning network structure search based on sparse coding

The invention discloses an image recognition method for deep learning network structure search based on sparse coding. The method comprises the following steps: optimizing a differentiable network model structure based on gradient; constructing a network model for performing structure search on a low-dimensional space mapped after the original high-dimensional space is compressed; enabling the compressed solution of the low-dimensional space to correspond to the sparse solution of the original space through a sparse coding technology, wherein the optimized network model structure during searchis the structure during retraining and is applied to search-retraining two-stage image recognition and search-retraining merging one-stage image recognition. According to the method, the network in the search stage has sparsity, the finally converged structure in the search training stage is the finally searched structure, network structure search is more efficient and reasonable, and image recognition performance is excellent.