Mobile terminal light weight image super-resolution reconstruction method based on convolutional neural network
By employing an equivalent transformation technique based on convolutional neural networks, the model structure is simplified and made suitable for mobile devices. This solves the problems of high computational resource consumption and model incompatibility in existing technologies, achieving fast and high-quality image super-resolution reconstruction.
CN115965527BActive Publication Date: 2026-07-07EAST CHINA NORMAL UNIV
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- EAST CHINA NORMAL UNIV
- Filing Date
- 2022-12-21
- Publication Date
- 2026-07-07
Smart Images
Figure CN115965527B_ABST
Abstract
The application discloses a kind of mobile terminal light weight image super-resolution reconstruction method based on convolutional neural network, comprising the following steps, first, obtain training dataset.Then, construct the light weight image super-resolution network suitable for mobile terminal, including training time and inference time network, inference time network is converted from training time network using equivalent conversion method.Equivalent conversion method replaces time-consuming operator in mobile terminal with less time-consuming convolution.In training time, using the above training dataset and the light weight super-resolution network constructed, compare the loss of original picture and generated high-resolution picture in dataset, based on loss back propagation calculation until training is finished.In inference time, using the more concise network of equivalent conversion, model volume is small, and the output rate is fast.The advantages of the application are: based on the scene of mobile terminal is optimized and adapted, using equivalent conversion method constructs a simple, efficient super-resolution network.
Need to check novelty before this filing date? Find Prior Art