一种前景和背景关联共生的红外图像迭代生成方法
By using a generative adversarial network iterative generation method, the problem of insufficient data for detecting small infrared targets in the air was solved, generating diverse infrared image datasets, supporting deep learning model training, and improving the accuracy of infrared imaging system development and testing.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING INST OF TECH
- Filing Date
- 2022-12-26
- Publication Date
- 2026-07-17
AI Technical Summary
In existing technologies, there is a lack of effective datasets for the detection of weak infrared targets in the air, especially for images of weak infrared targets against complex backgrounds, which makes it difficult to support the training of deep learning models. Furthermore, there is limited research on diverse methods for generating infrared images in the detection of weak infrared targets in the air.
A generative adversarial network-based approach is adopted to iteratively generate diverse infrared images by training a generator and a discriminator. The MUNIT model is used to generate foreground targets, and the CycleGAN model is used to generate backgrounds, ensuring that the generated backgrounds and foregrounds coexist and are diverse, thus constructing a multimodal infrared image dataset.
It enables the generation of diverse infrared images of weak targets based on existing datasets, meets the training requirements of deep learning models, improves the accuracy of infrared imaging system development and testing, and promotes the development and practical application of infrared imaging guided weapons.
Smart Images
Figure CN115830169B_ABST