基于GoogLeNet-WT-Canny的紫外成像仪中的图像配准与融合

Image Registration and Fusion of High Sensitive Ultraviolet Imager Based on GoogLeNet Model and WT-Canny

摘要:

针对目前紫外成像仪中紫外光图像与可见光图像配准叠加精度低、电晕放电定位偏差大等问题,提出一种基于GoogLeNet模型、小波变换(Wavelet Transform,WT)和Canny算子相结合的紫外与可见光图像配准融合方法,并将其应用于高灵敏紫外成像仪中。首先,引入迁移学习的思想,利用预训练的GoogLeNet模型自主挖掘可见光图像和紫外图像的特征；其次,将提取出来的特征作为预测变量,输入极限学习机(Extreme Learning Machine,ELM),以空间变换参数为指导监督模型训练,实现高精度紫外与可见光图像配准；最后,利用二维小波变换与Canny算子对配准后的图像进行多分辨率分析与边缘检测,实现无紫外信息损失的图像融合。实验结果表明,所提方法的紫外与可见光图像配准精度高,融合效果良好,具有很好的工程实用价值。

Abstract:

Aiming at the problems of low registration and superposition accuracy of ultraviolet image and visible image and large deviation of corona discharge location in ultraviolet imager, a UV and visible image registration and fusion method was proposed based on the GoogLeNet model, Wavelet Transform (WT) and Canny operator, and it was applied in highly sensitive UV imagers. First, by introducing the idea of migration learning, the pre-trained GoogLeNet model was used to autonomously mine the characteristics of visible light and ultraviolet images. Then, the extracted features were input into extreme learning machine (ELM) as predictive variables, and supervised model training was guided by spatial transformation parameters to achieve high-precision UV and visible image registration. Finally, the multi-resolution analysis and edge detection of the registered image were carried out by using two-dimensional wavelet transform and Canny operator to realize the image fusion without UV information loss. The experimental results show that the proposed method has high registration accuracy of UV and visible images, completing good fusion effect and engineering applicability.