The coffee leaf miner (Leucoptera coffeella) is a key coffee pest in Brazil that can cause severe defoliation and a negative impact on the productivity. Thus, it is essential to identify initial pest infestation for the sake of appropriate time control to avoid further economic damage to the coffee crops. A fast non-destructive method is an important tool that can be used to monitor the occurrence of the coffee leaf miner. The present work aims to identify the occurrence of coffee leaf miner infestation through a new vegetation index, using multispectral images from the Sentinel-2 satellite and the Google Earth Engine platform. Coffee leaf miner infestation was measured in the field in four cities in the state of Minas Gerais. The largest infestations occurred in September, October, and November but particularly in October 2021, in which the rate of infestation reached 85%, followed by September 2020 with a maximum infestation of 76%. The calculation steps of the vegetation indices and mappings were carried out in the Google Earth Engine cloud processing platform through the development of a script in JavaScript programming language. Combinations of two sensitive bands were selected to detect coffee leaf miner infestation, and from these, the “Coffee-Leaf-Miner Index” was developed, which was compared with other existing vegetation indices in terms of their performance for coffee leaf miner detection. The combination of the NIR–BLUE and NIR–RED bands was more sensitive for the detection of coffee leaf miner infestation; therefore, the NIR, BLUE, and RED bands were selected to develop the new index. The “Coffee-Leaf-Miner Index” presented the best performance among those evaluated, with a coefficient of determination of about 0.87, a root mean square error of 4.92% coffee leaf miner infestation, accuracy of 89.47%, and kappa coefficient of 95.39. The R² range of other spectral indices which exist in the literature and which were used in this study was from 0.017 to 0.867, and the root mean square error ranged from 4.996 to 13.582% coffee leaf miner infestation. The machine learning method was then adopted using the supervised Random Forest and Support Vector Machine algorithms to recognize patterns of coffee leaf miner infestation in the field, only the Coffee-Leaf-Miner Index was used for the identification test of the coffee leaf miner infestation. The Support Vector Machine with linear Kernel type was applied to establish a discrimination model. The number of trees for the Random Forest classifier was 100. The Support Vector Machine presented a lower performance than the Random Forest algorithm, but the performance of both were above 80% for user and producer precision. Three bands (Blue, Red, NIR) were selected for the creation of the new index, which showed capacity for remote detection of coffee leaf miner infestation on a regional scale. Thus, “Coffee-Leaf-Miner Index” can identify coffee leaf miner infestation thanks to all the complexity involved in detecting pests via orbital remote sensing.

The coffee leaf miner (Leucoptera coffeella) is a key pest in coffee-producing regions in Brazil. The objective of this work was to evaluate the potential of machine learning algorithms to identify coffee leaf miner infestation by considering the assessment period and Sentinel-2 satellite images generated on the Google Earth Engine platform. Coffee leaf miner infestation in the field was measured monthly from 2019 to 2023. Images were selected from the Sentinel-2 satellite to determine 13 vegetative indices. The selection of images and calculations of the vegetation indices were carried out using the Google Earth Engine platform. A database was generated with information on coffee leaf miner infestation, vegetation indices, and assessment times. The database was separated into training data and testing data. Nine machine learning algorithms were used, including Linear Discriminant Analysis, Random Forest, Support Vector Machine, k-nearest neighbors, and Logistic Regression, and a principal component analysis was conducted for each algorithm. After optimizing the hyperparameters, the testing data were used to validate the model. The best model to estimate miner infestation was RF, which had an accuracy of 0.86, a kappa index of 0.64, and a precision of 0.87. The developed models were capable of monitoring coffee leaf miner infestation.

Coffee is among the primary agricultural commodities in international trade; however, mapping coffee crops in mountainous regions faces limitations due to high spectral variability and complex canopy structures. This study hypothesized that optimized spectral band combinations focused on the visible spectrum may outperform configurations including near-infrared (NIR) for coffee crop segmentation. This work aimed to evaluate how different spectral band combinations affect the performance of the U-Net for segmenting coffee crops in mountainous regions. Seven PlanetScope images (4 m resolution) from Matas de Minas, Brazil, covering different phenological stages in 2023–2024, were divided into 316 training patches and 25 test patches of 256 × 256 pixels and used to train U-Net models across five spectral band combinations: (B, G, R), (B, G, NIR), (B, R, NIR), (G, R, NIR), and (B, G, R, NIR). The visible spectrum combination (B, G, R) demonstrated superior performance with an overall Accuracy of 0.8669 and, for the Coffee Crops class, an F1-score of 0.8682 and an IoU of 0.7671, outperforming all NIR-inclusive configurations. Visible bands’ sensitivity to pigmentation variations proved more effective in heterogeneous environments, while NIR increased spectral confusion near native vegetation and crop edges. The model overestimated cultivated area by 18.3% due to mixed pixels from 4 m resolution and mountainous terrain. These findings confirm that visible-spectrum bands offer a cost-effective alternative for coffee segmentation, though higher spatial resolution is needed for improved boundary delineation.