Objective Early blight is a common destructive disease in the growth process of Solanaceae crops, which can lead to crop failure and serious losses. Traditional crop disease detection methods are difficult to detect disease characteristics in a timely manner during the incubation period of disease, and thus take scientific and effective prevention and control measures. This article obtained hyperspectral images of early blight of peppers at different infection stages through continuous monitoring with a hyperspectral imager. The earliest identifiable time during the incubation period of early blight in peppers (the earliest identifiable time during the incubation period in this experiment was 24 h after inoculation) was determined using the spectral angle cosine-correlation coefficient and Chebyshev distance. Method Taking the symptoms of the latent period of early blight in peppers as the research object, 13 characteristic wavelengths were selected using a genetic algorithm. An identification model of crop disease latent period symptoms based on spectral features was established through optimized combinations of characteristic wavelengths combined with a logistic regression model. Simultaneously, a recognition model of the latent period of early blight in peppers based on image texture features was established using local binary patterns. Result The experiment was tested with 120 samples. The accuracy of the identification model of crop disease latent period symptoms based on spectral features reached over 93% in both the training set and the test set. The accuracy of the identification model of crop disease latent period symptoms based on texture features reached 98.96% and 100% in the training set and test set, respectively. Conclusion The results showed that both spectral features and texture features can be used to detect and identify crop disease latent period symptoms. Texture features more significantly revealed the characteristics of the latent period of the disease compared to spectral features, effectively improving the detection performance of the model. The research results in this article can provide theoretical references for monitoring and identifying other crop disease latent period symptoms.