研究目的
To present a novel deep learning architecture for infrared and visible images fusion problem that achieves state-of-the-art performance in objective and subjective assessment.
研究成果
The proposed deep learning architecture based on CNN and dense block for infrared and visible image fusion exhibits state-of-the-art fusion performance, as shown by experimental results. It can also be applied to other image fusion problems with appropriate fusion layer.
研究不足
The training data of infrared and visible images is insufficient, requiring the use of gray scale images of MS-COCO for training the model.
1：Experimental Design and Method Selection:
The proposed method uses a deep learning architecture combining convolutional layers, fusion layer, and dense block for feature extraction and fusion.
2：Sample Selection and Data Sources:
Infrared and visible image pairs were collected from specific datasets for training and testing.
3：List of Experimental Equipment and Materials:
NVIDIA GTX 1080Ti GPU and Tensorflow as the backend for the network architecture.
4：Experimental Procedures and Operational Workflow:
The network is trained using gray scale images of MS-COCO, with specific learning rates, batch sizes, and epochs.
5：Data Analysis Methods:
The performance is evaluated using subjective and objective quality metrics including entropy, Qabf, SCD, FMI, SSIMa, and MS SSIM.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
