Differential pulsewidth pair (DPP) technology effectively improved the spatial resolution of the Brillouin optical time domain analysis (BOTDA) system. However, the signal-to-noise ratio (SNR) of the time domain signal is reduced after differential processing, and the accuracy of Brillouin frequency shift (BFS) also deteriorates. We present a novel approach that combines the DPP technique with the fast and flexible denoising convolutional neural network (FFDNet) to enhance the key performance indicators of BOTDA systems, such as spatial resolution, SNR, and frequency shift accuracy. In the experiment, a 50/40 ns pulse pair and a 45/40 ns pulse were used to reduce the spatial resolution from 4 to 1.12 and 0.66 m, respectively. Without affecting the spatial resolution, the FFDNet denoising method effectively improves the SNR of the system and the extraction accuracy of BFS. In the simulation, we used this method to improve the SNR by 40.71 dB and reduce the BFS uncertainty along the fiber by 4.08 MHz. In the experiment, the method improved the SNR of the signal acquired along the 2 km sensing fiber by up to 24.22 dB, and the BFS uncertainty along the sensing fiber was reduced from 2.13 to 1.08 MHz, a reduction of 1.05 MHz. In addition, the processing speed of FFDNet denoising method is much faster than that of the traditional wavelet denoising (WD) method and non-local mean (NLM) denoising method, taking only 1.37 s, which has great potential in actual fast denoising.