In order to detect potato late blight quickly and compare the difference of spatial spectrum information, hyperspectral imaging technology was used to compare the spatial spectrum of potato late blight in order to find the best discriminant method. A hyperspectral camera was used to collect the hyperspectral images of 0?6 days of disease infection. At the same time, the hyperspectral data of typical late blight diseases on the 6th day were selected as the research object. Second-order derivative combined with principal component analysis and second-order principal component analysis were used to extract features from spectral and spatial aspects respectively. Then, K-nearest neighbor classification algorithm, BP neural network and decision tree algorithm were established based on the reflectance of characteristic band and the gray value of principal component image to identify diseases in different periods. The recognition rate of the model was 96.6% based on the gray value of the secondary principal component image and BP neural network. The experimental results showed that the model based on the gray value of the secondary principal component image and BP neural network had good effect on the identification of potato late blight. The three models based on the gray value of the principal component image reduce the redundancy of the band and improve the recognition rate, which provides a reference for researching and developing real-time on-line testing equipment and instruments..