Optical fiber temperature sensor has attracted extensive attention in the extreme environment, cable monitor, and biological field due to the ability of anti-electromagnetic interference, high biocompatibility, flexibility, and easy configuration. To enhance the sensitivity, various interferometer structures such as Mach-Zehnder interferometer (MZI), Sagnac ring, Fabry-Perot interferometer (FPI) are proposed. The perceived signal is modulated in the spectrum, convolution demodulation method is based on the optical spectrometer analyze (OSA) or Field Programmable Gate Array (FPGA). However, the convolution method needs to spectral scan, so the demodulation time is extremely limits. Using disperse fiber could transform the spectrum to time domain. In time domain, the signal could be detected and analyzed real time. But the spectrum and time domain data are not strictly corresponding, so there is a challenge on data analysis. In this paper, we propose a dual peanut shaped MZI sensor structure, use disperse fiber to transform the sensor data to time domain and construct a classification method to identify the specific temperature based on 1D CNN. We aimed to realize the ultrafast, accurate optical fiber temperature interrogation.