Image fusion method for industrial X-Ray dual energy computed tomography based on VGG-19 network for improved image quality

Currently, industrial X-ray computed tomography (CT) is widely used to test the quality of production results. However, the main problem with conventional CT is testing multi-material samples which produce such image imperfections as artifacts, noise, lack of contrast, and missing parts of detail. In this research, image acquisition for multi-material samples uses dual-energy X-ray computed tomography (DECT) to obtain low- and high-energy images. The image fusion method used to combine the two source images uses the VGG-19 network architecture. The source image fusion procedure, namely the VGG-19 network created, is used to extract in-depth features in the detailed part of the combined image, while the basic image features are combined directly by averaging the weights on the basic features. The proposed method is compared with other commonly used image fusion methods, namely discrete wavelet transform (DWT) and squeeze-and-excitation spatial frequency (SESF). Projection images acquired by self-developed 80 kV-160 kV DECT were used to evaluate the performance of these methods. Evaluation metric of fusion results is carried out using several image quality measurement parameters including structural similarity index measure (SSIM), feature mutual information (FMI), and noise added by the fusion (N_abf). The sample that has the best average metric value is the spark plug with the highest value of FMI_dct evaluation metric is 2.96×10^-1, FMI_pixel value is 9.70×10^-1, FMI_w value is 3.69×10^-1, SSIM value is 9.92×10^-1, and the lowest N_abf value is 3.82×10^-3. Experimental results show that the proposed method can produce fused images that are generally better than other comparison methods. The proposed method can improve image quality, the resulting tomographic image has less noise, good contrast, and can maintain object structure. This research implies that imaging using industrial X-ray CT can obtain more informative and accurate image results, especially for testing multi-material samples.