A method for joint detection of diseased leaves and healthy leaves and extraction of effective characteristic wavelengths for potato late blight was proposed. Principal component analysis was performed on the spectral images of healthy and late blight leaves, and the weight coefficient curves of principal component images were analyzed to extract six characteristic bands of healthy leaf and diseased leaf. Based on the six characteristic wavelengths, the second principal component analysis was performed, and the optimization was reduced to three characteristic wavelengths of 712.19, 749.70 and 841.47 nm. Based on these three characteristic wavelengths, principal component analysis was used to identify the diseased area with the most contrasting image of the main component, and the recognition rate was 100%. The characteristic wavelength of potato late blight disease could be achieved by combining healthy leaves with diseased leaves and secondary principal components. This technology provides a reference for the development of related equipment for potato leaf disease detection.