Mapping method and device based on deep learning

Abstract

The invention provides a mapping method and device based on deep learning, and relates to the field of image processing. The method comprises: detecting a foreground mask area of a target image to obtain a foreground mask image; inputting the target image, the foreground mask image and a background image into a convolutional neural network, and performing fusion training on the target image and the background image to obtain an output image; calculating the linear regression loss of the output image and judging whether a preset condition is met or not, and if not, reversely transmitting the linear regression loss; adjusting a network weight parameter of the convolutional neural network according to the reversely transmitted linear regression loss; and performing fusion again according to the adjusted network weight parameters, and calculating linear regression loss of the output image after fusion again until the linear regression loss meets a preset condition. The feature data of themapping target is automatically calculated and calibrated, and the network weight parameters are adjusted to enable the target features to be automatically iterated, so that the mapping quality requirement is met, and the mapping efficiency is improved.

Description

technical field [0001] This application relates to the field of image processing, in particular, to a deep learning-based mapping method and device. Background technique [0002] In the past ten years, deep learning has been an important breakthrough in the field of artificial intelligence technology, especially in the field of image recognition and processing, and has achieved great success. [0003] Deep learning is a method based on data representation learning in machine learning. Its purpose is to establish or simulate the neural network of the human brain for analysis and learning, and to imitate the recognition process of the human brain for images, sounds and texts. [0004] In deep learning, manual calibration is usually used to calibrate and iterate the data of the object or content that needs to be recognized, and manual calibration is a very slow process, which is prone to category mislabeling and target position deviation. A series of problems, such as shifting...

AI Technical Summary

Problems solved by technology

[0004] In deep learning, manual calibration is usually used to calibrate and iterate the data of the object or content that needs to be recognized, and manual calibration is a very slow process, which is prone to category mislabeling and target position deviation. A series of problems such as shifting and slow calibration speed reduce the quality of the texture and the efficiency is not high

Images