UAV imaging for spectral characterization of Coffee Leaf Miner (Leucoptera coffeella) infestation in the Cerrado Mineiro region

Abstract

Brazil, the world's largest coffee producer, faces challenges managing the coffee leaf miner (Leucoptera coffeella), a significant pest. This study suggests remote sensing for pest control decisions. Two experimental areas in the Cerrado region of Minas Gerais State were analyzed to spectrally characterize infested plants and estimate the number of mines per plant. Results show the ability to differentiate infested plants with greater reflectance variance in the near infrared at 850nm. The performances of the three machine learning algorithms were compared. Determining the number of mines in the group of most infested plants demonstrated slightly higher precision, achieving an RMSE of 22.69% using the Support Vector Machine algorithm. Conversely, the group of least-infested plants obtained the best result with the Random Forest algorithm, achieving an RMSE of 32.47%. These promising results indicated that CLM can be detected using aerial multispectral imaging data.