Ship sign image super-resolution method based on semantic information and gradient supervision

Abstract

The invention discloses a ship sign image super-resolution method based on semantic information and gradient supervision. The method comprises the following steps: 1 collecting and preprocessing an original ship image; 2 constructing a feature extraction network and a super-resolution reconstruction network for ship sign characters, and forming a generative network model; then carrying out adversarial learning pre-training through the DF2K data set, and obtaining a generative network pre-training model; 3 after a BLSTM structure is introduced into the generative network pre-training model, adopting a ship sign character data set is for training; and 4 when adversarial learning training is carried out, due to the fact that a ship sign character area has certain sharpness, in order to better guide the network to generate super-resolution ship sign characters, adding a character gradient loss function to enhance supervision on the generated network. According to the method, the clearness of ship sign characters can be obviously improved, the ship sign information can be rapidly identified manually, the traffic control of sea and river shipping is facilitated, and the safety of waterway shipping is improved.

Description

technical field [0001] The invention belongs to the technical field of deep learning, image processing, intelligent management and monitoring of ships in shipping channels, and super-resolution reconstruction, and relates to a super-resolution method for ship plate images based on generative confrontation networks and guided by semantic information and gradient supervision. Background technique [0002] China's shipping waterway transportation system is developed, the inland river network structure is perfect, and the ports are densely distributed, which has greatly promoted the development of the cargo shipping industry. Inland waterway shipping has become an important part of the modern comprehensive transportation system and one of the main contents of the rational development and comprehensive utilization of water resources. According to statistics, in October 2020 alone, the total volume of national waterway cargo transportation reached 706.59 million tons, and the carg...

AI Technical Summary

Problems solved by technology

[0004] At present, there are two main solutions based on traditional methods and deep learning, but these solutions have many problems in the super-resolution of ship plate text: 1) The traditional image super-resolution method is relatively simple due to the algorithm poor performance on

2) In the deep learning method, the method of constructing the training data set with bicubic linear interpolation cannot meet the application of the actual scene

3) The super-resolution method of deep learning lacks the research on the semantic characteristics of the text area of ​​the ship plate, which makes the effect of the algorithm for text super-resolution unsatisfactory

4) Existing deep learning methods are still lacking in dealing with text sharpening effects

Therefore, there is still a lack of super-resolution methods for ship plate text

Images