In scientific research and practical engineering projects, it is necessary to measure the displacement, deformation, vibration and strain of optical rough surfaces with high precision. Electronic speckle interferometry is a high-precision non-destructive testing technique for optical rough surfaces, but in the measurement process, the interference fringe image inevitably contains a large amount of coherent speckle noise and environmental noise, which seriously affects the subsequent phase demodulation and unwrapping accuracy and accuracy. Therefore, an experimental model of image noise reduction based on DnCNN neural network model was designed, hardware system was designed and built, and software was written to realize the acquisition and enhancement of images. The computer simulated speckle interference fringe image was used to generate the dataset, and the design and training of the neural network were completed. Experimental results show that the design method can effectively denoise, reduce white noise, and achieve the best fitting effect of the network. Experimental tests show significant advantages over traditional filtering methods in the evaluation of denoising performance, among which the signal-to-noise ratio and speckle index are outstanding.