Investigation of the performance of various machine learning algorithms and vegetation indices in the detection of strawberry fields

Abstract

This study aimed to determine the optimal remote sensing method for identifying cultivated strawberry (Fragaria × ananassa) fields in the Köprübaşı district of Manisa province, a region characterized by intensive commercial strawberry production. On the Google Earth Engine (GEE) platform, Sentinel-2 satellite images were used to classify the land cover of the region into six categories: forest, bare land, strawberry fields, water surfaces, other agricultural areas, and urban areas. For the classification process, various machine learning algorithms such as Random Forest (RF), Support Vector Machines (SVM), Classification and Regression Trees (CART) and Naïve Bayes (NB) were utilized. In addition to traditional spectral bands, vegetation indices such as the Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Red Edge Normalized Difference Vegetation Index (ReNDVI), and Soil-Adjusted Vegetation Index (SAVI) were also incorporated into the classification process. The classification results were validated using high-resolution Google Earth Pro images, and for accuracy analysis, overall accuracy (OA), producer's accuracy (PA), user's accuracy (UA), kappa (κ) statistics, and F1-Score were calculated. According to the results, the SVM algorithm was identified as the most successful method, achieving an OA of 81.76% and a κ value of 0.78 with the RGB+NIR+NDVI band combination. In the detection of strawberry croplands, it achieved the best performance with 86.95% producer's accuracy (PA), 92.04% user's accuracy (UA), and an F1-Score of 89.42%. This study demonstrates that remote sensing and machine learning techniques can be effectively used for mapping strawberry fields.