The ultra-fast all-optical solid-state framing camera (UASFC) is a new type of X-ray ultra-fast diagnostic technology. It uses X-ray excitation to change the refractive index distribution of the ultra-fast detection chip, and time-tuned multi-wavelength probe light for ultra-fast detection on the order of picoseconds or less. Due to the uneven intensity of the probe light wavelength and spatial diffraction, the noise of the detection image is too high, which directly affects the spatial and temporal resolution of the system. To improve the detection performance of the UASFC system, we adopted NLM image optimization technology based on a convolutional neural network, using 50 shots of the reticle image as the learning set and iterating the weight of the NLM image optimization filter. The CCD image obtained by the four-channel wavelength spectroscopy system is noise-reduced and optimized, which greatly improves the image contrast and edge definition, reduces image noise, and further improves the time and space resolution of the UASFC system. © 2021, Springer Nature Switzerland AG.