The tunable diode laser absorption spectroscopytechnology is widely used in flow field detection. However, the harsh environment of on-site application puts higher requirements on the test system’s environmental adaptability and anti-noise interference ability. Direct absorption spectroscopy technology can obtain all the information on the absorption spectrum and is suitable for many types of test environments, such as wind tunnels and engine flow fields. However, due to the influence of a harsh on-site test environment, the signal-to-noise ratio of the obtained signals is usually poor, and it isn’t easy to achieve high-precision detection. This article analyzes the data processing and analysis algorithm of direct absorption spectroscopy technology and proposes a high-precision temperature analysis method based on transmission curve fitting for this problem. A temperature detection system based on direct absorption spectroscopy technology was developed using the absorption spectra of water vapor near 7 185.6 and 7 444.3 cm^-1. The processing and analysis of test signals in the temperature range of 573～1 173 K were carried out. The analysis results of the baseline-absorption line fitting algorithm(BALF) and the transmission curve fitting(TCF) algorithm were compared. When the detection temperature was 773 K, compared with the BALF algorithm, the peak-to-peak fitting error near 7 444.3 cm^-1 of the TCF algorithm can be reduced by about 30%, and the peak-to-peak fitting error near 7 185.6 cm^-1 can be reduced by about 16%. In the temperature range of 573～1 173 K, the maximum error of the TCF algorithm for temperature inversion is 13 K, which is reduced by 23 K compared with the BALF algorithm. When random noise signals with different amplitudes were added to the test signals, the standard deviation of test results using these two algorithms increased with noise amplitude. Moreover, the higher the temperature, the greater the standard deviation of test results. When the peak-to-peak noise value was 20, 60, and 100 mV respectively, for different temperatures, the minimum standard deviation of test results using the BALF algorithm was 18 K, and the maximum was 313 K. The minimum standard deviation of test results using the TCF algorithm is 4 K, and the maximum is 44 K. Experimental analysis results show that compared with the BALF algorithm, the TCF algorithm can correct the baseline fitting error of laser power, improve the fitting accuracy of transmission signals, and reduce temperature analysis errors. Comparing the temperature analysis results of test signals under different noise levels shows that the TCF algorithm can achieve higher detection accuracy and precision under noise interference and has stronger anti-noise interference ability.