基于CBAM-DDcGAN的锌渣红外与可见光图像融合

Infrared and visible images fusion of zinc slag based on CBAM-DDcGAN

对热镀锌生产线锌池中的锌渣分布进行准确识别是捞渣作业智能化的前提。红外图像易损失锌渣的纹理特征,可见光图像又容易受到光照等因素的影响,红外与可见光图像的融合是提高锌渣识别精度的有效手段。本文提出一种结合卷积注意力机制模块(CBAM)和双判别器生成对抗网络(DDcGAN)的红外与可见光图像融合算法。首先,生成器内部采用编解码网络,编码器采用密集连接方式,与CBAM相结合,在最大程度地保留图像特征的同时还能突出关键特征,抑制无用信息,提升融合效果。其次,利用两个判别器与生成器分别进行对抗训练,可同时保留两种源图像的信息。采用锌渣图像数据集进行对比实验,结果表明,本文算法所得融合图像在主观视觉效果和客观定量指标上均有不同程度的改善。

Accurate identification of zinc slag distribution in the zinc pool of hot-dip galvanizing line is the premise of intelligent slag-removing operation.Infrared images are likely to miss the texture features of zinc slag,and visible images are susceptible to illumination and other factors.Infrared and visible images fusion is an effective means to raise the recognition precision of zinc slag.Therefore,an infrared and visible images fusion algorithm was designed by combining dual-discriminator generative adversarial network(DDcGAN)and convolutional block attention module(CBAM).Firstly,the ge-nerator used a codec network inside,and the encoder used the dense connection pattern,which was integrated with CBAM to maximize the retention of image features while highlighting key features and suppressing useless information,so the fusion effect was improved.Secondly,two discriminators were used to perform adversarial training with the generator respectivily,which helps retain information from both source images.Comparative experiments were performed on zinc slag image dataset,and the results show that the fusion images by the proposed method are improved by varying degrees in terms of subjective visual effect and objective qualitative indicators.