The invention discloses a multispectral detection-based waveband selection method. Through a multispectral imaging system, reflectance characteristics of an object at a plurality of spectral bands can be obtained, and by the reflectance characteristics as features, waveband selection is realized. The multispectral imaging system adopts a narrowband optical filter and calibrates light source brightness output. The multiple photographed and calibrated multispectral images comprise positive samples and negative samples needed by waveband selection. Reflectance characteristics of the positive and negative samples at different wavebands are calculated and extracted and are used as features and by the features, waveband selection is carried out. The multispectral detection-based waveband selection method is an iterative computation method based on one-dimensional histogram distance measurement. The multispectral detection-based waveband selection method overcomes defects such as high complexity and large calculated amount of high-dimensional spatial distance measurement, and improves waveband selection efficiency.