Wavelength Division Multiplexing (WDM) is one of the core technologies in fiber optic communication system. In WDM system, a number of optical signals with different wavelengths are transmitted simultaneously in a single optical fiber, realizing the multiplexing of the optical signals, and solving a series of problems in large-capacity, high-speed data transmission. At the receiving end of WDM system, the multiplexed optical signals are demultiplexed into a series of single wavelength optical signals, then further detected and identified to usually obtain the peak values of the signals for the next applications. Generally, the peak value of the target single-wavelength optical should be detected and identified in real time. For this purpose, we designed and developed a real-time peak detection system for multi-wavelength optical signals based on multi-window recognition, which realizes the detection and recognition of the peaks for the multi-wavelength optical signals.This work proposes a real-time peak detection system for multi-wavelength optical signals based on multi-window recognition. The optical demultiplexing part uses an Fiber Fabry-Perot Tunable Filter (FFP-TF) to filter out the target single-wavelength signal from the multiple-wavelength optical signals. The driving and controlling circuit of FFP-TF and the low-noise weak optical signal conversing and detecting circuit are designed. The former is used to drive and control the FFP-TF to filter out the optical signals of the target wavelengths and the latter detects and converts the power intensity of the optical signals in real time. The detected optical signal power is converted into voltage intensity. Then the peak position and intensity in the voltage waveform are identified by using a method based on multi-window waveform identification. At last, the corresponding peak intensity and wavelength position of the optical signal waveform is obtained. The system realizes the filtering of the target wavelength optical signal, and the detecting and identifying of the peak intensity in the signal waveform at the target wavelength.Firstly, a standard spectrometer is used to detect the optical signal waveform of the experiment light source. The results are illustrated (Fig.11). Secondly, the detection system proposed and developed in this work is used to determine the same experiment light source. The comparison of the results measured by the standard spectrometer and by the detection system proposed and developed in this work is shown (Fig.12). The corresponding data to Fig.11-12 are shown (Tab.1). It can be seen that the peak points of the signal waveform obtained respectively from the standard spectrometer and the detection system proposed and developed in this work agree very well. The measurement variation of the peak intensity obtained from the detection system is less than 0.01 dBm and the recognition time is less than 3 s. More experiment results illustrated in Fig.13 show that the detection range of the system proposed and developed in this work is 0--60 dBm.In this work, a real-time peak detection system for multi-wavelength optical signals based on multi-window recognition is proposed and developed. The driving voltage of the FFP-TF is controlled to filter out the target wavelength optical signal and the weak optical signal detection circuit converts the target wavelength optical signal into a normal range of voltage. Then the multi-window waveform peak recognition method is used to identify and record the signal peaks in the waveform. The optical signals of different powers from narrowband light source are tested. The results show that the waveforms of the original optical signal and the detected one by the system is consistent. The measurement variations of peak intensity is less than 0.01 dBm, the recognition speed is less than 3 s, and the minimum detectable power of optical signal is as low as -60 dBm.