To access the growth information of crops fast, non-destructively in real-time, a four band multi-spectral sensor of crop was developed based on spectral monitoring mechanism of crop growth indicators. Optical filtering technique was used to improve the input signal to noise ratio of optical radiation. Based on crop canopy characteristics and operating environment in the field, the suitable structure parameters of detection lens was designed, ensuring that sensitivity and resolution effect of multi-spectral sensor. The weak spectral information amplifying circuit was built by T-type resistance network. Through calibration, the relational model of multi-spectral sensor and FieldSpec Pro FR2500 spectrometer was obtained, the coefficient of determination of which was 0.8028, 0.8068, 0.8185 and 0.8900 respectively. The wheat test results showed that the average measurement error of the sensor were 5.6%, 4.6%, 1.4% and 4.5% respectively. The sensor could realize real-time online tests of crop canopy reflectance spectra to provide strong support for the development of the crop growth monitoring equipment.