Abstract: Due to superior performance of the wavelet scattering network, it is rapidly and widely used in the fields of image and audio. This new type of network structure is used to extract the features of the image and combines similarity measurement methods to achieve the application of this feature extraction in image retrieval. In addition, the eigenvalues obtained by the wavelet scattering network are taken as its new features, taking the mean value and the variance as the new features, so as to realize the dimensionality reduction of large-scale image retrieval. Finally, the experimental algorithm is used to verify the feasibility and superiority of the above dimensionality reduction method.