Cotton plays an essential role in global human life and economic development. However, diseases such as leaf blight pose a serious threat to cotton production. This study aims to advance the existing approach by identifying cotton blight infection and classifying its severity at a higher accuracy. We selected a cotton field in Shihezi, Xinjiang in China to acquire multispectral images with an unmanned airborne vehicle (UAV); then, fifty-three 50 cm by 50 cm ground framed plots were set with defined coordinates, and a photo of its cotton canopy was taken of each and converted to the L*a*b* color space as either a training or a validation sample; finally, these two kinds of images were processed and combined to establish a cotton blight infection inversion model. Results show that the Red, Rededge, and NIR bands of multispectral UAV images were found to be most sensitive to changes in cotton leaf color caused by blight infection; NDVI and GNDVI were verified to be able to infer cotton blight infection information from the UAV images, of which the model calibration accuracy was 84%. Then, the cotton blight infection status was spatially identified with four severity levels. Finally, a cotton blight inversion model was constructed and validated with ground framed photos to be able to explain about 86% of the total variance. Evidently, multispectral UAV images coupled with ground framed cotton canopy photos can improve cotton blight infection identification accuracy and severity classification, and therefore provide a more reliable approach to effectively monitoring such cotton disease damage.