Diagnosis of corn leaf diseases by FTIR spectroscopy combined with machine learning

Abstract

Corn is among the world's most vital crops, yet its yield and quality are often compromised by leaf diseases. Timely and accurate detection of such diseases is thus crucial. In this study, Fourier-transform infrared (FTIR) spectra (4000–400 cm⁻¹) were obtained for leaves afflicted by northern corn leaf blight (NCLB) and gray leaf spot (GLS), alongside spectra from healthy corn leaves as controls. Various machine learning-based classification models were then developed to facilitate precise disease diagnosis. To reduce redundancy and extract pertinent spectral information, the variable importance projection (VIP) algorithm and random leapfrog (RF) method were employed for feature selection. The resulting spectral features were subsequently used as inputs for the classification models. Of the twelve models evaluated, the VIP-KNN model demonstrated the most exceptional performance. While the original FTIR spectrum comprised 1867 data points, the VIP-KNN model achieved classification using only 615 critical data points, delivering an accuracy of 97.46 %, sensitivity of 96.08 %, and precision of 95.96 %. This highlights how the feature selection approach mitigated overfitting and substantially enhanced the model's classification accuracy. The findings of this research underscore the potential of combining FTIR spectroscopy with machine learning for the effective diagnosis of corn leaf diseases, the accuracy of this detection method is high, and the average accuracy of the model is as high as 93.41 %.