| 摘 要: 针对红外与可见光图像融合算法中细节模糊、对比度低等问题,提出一种基于深度特征提取的融合算法。该算法模型具有较好的泛化性,通过迁移学习可应用于水利工程无人机自动巡检系统的安全监测。首先,利用滚动引导滤波对源图像进行多级分解。其次,针对基础层引入PCANet网络指导融合,针对细节层引入显著测量方法和引导滤波指导细节层融合。最后,将各融合部分进行叠加重构,得到融合图像。实验结果表明,该算法在主观视觉上具有细节信息丰富、高对比度的特点,在客观评价指标上也有较好的结果,尤其是标准差指标,较以往大部分方法领先幅度超过40%。 |
| 关键词: 红外与可见光图像;滚动引导滤波;PCANet;无人机;特征提取 |
|
中图分类号: TP391
文献标识码: A
|
|
| Infrared and Visible Image Fusion Based on Depth Feature Extraction |
|
ZHANG Wei1, WANG Leidan1, LIU Xiaoliang1, LI Kunhuang2
|
(1.Guangdong Yue Gang Water Supply Co., Ltd., Shenzhen 518021, China;
2.Shenzhen Highly Innovative Technology Co., Ltd., Shenzhen 518000, China)
506303235@qq.com; wangld@163.com; liuxliang@163.com; kunhli@163.com
|
| Abstract: Aiming at the problems of blurred details and low contrast in the fusion algorithm of infrared and visible images, a fusion algorithm based on deep feature extraction is proposed. The algorithm model has good generalization and can be applied to the safety monitoring of water conservancy engineering UAV(Unmanned Aerial Vehicle) automatic inspection system through transfer learning. Firstly, the source image is decomposed at multiple levels using rolling guided filtering. Then, PCANet network is introduced to guide fusion in the basic layer, and significant measurements and guided filtering are introduced to guide fusion in the detail layer. Finally, the fusion parts are overlaid and reconstructed to obtain the fused image. The experimental results show that the algorithm has rich detail information and high contrast in subjective vision, and also has better results in objective evaluation index, especially on the standard deviation indicator, which is more than 40% ahead of most previous methods. |
| Keywords: infrared and visible image; rolling guided filtering; PCANet; UAV; feature extraction |
