The high noise and low resolution of ultrasonic testing images of large-thickness honeycomb sandwich composite components bring challenges to quality evaluation and defect identification. In this study, image denoising and super-resolution reconstruction methods are investigated using ultrasonic C-scan images of the C919 inner door for the main landing gear. An efficient edge-preserving filter denoising scheme based on a gradient descent algorithm is proposed, effectively removing speckle noise while preserving image details. After denoising, the peak signal-to-noise ratio reaches 37.53 dB and the structural similarity is 0.92, outperforming the digital morphological filter by 12.99 dB and 0.04, respectively. The images reconstructed by the improved super-resolution residual network (ISRResNet) model have high resolution, rich content details, and clear edges. Furthermore, an image with Φ11 mm delamination defects is processed, the signal-to-noise ratio of defects is improved by 6.25 dB on average. Results show that the proposed denoising method and super-resolution model can effectively remove speckle noise, improve the image resolution, and enhance the accuracy of defect quantification. It can support high-quality ultrasonic testing of large-thickness honeycomb sandwich composite components.