A dual-labeled mango leaf image dataset for varietal classification and flavor prediction

This dataset contains high-resolution mango leaf images collected to support non-destructive, pre-harvest computer vision applications in agriculture. It contains 1,873 original raw images with 1,759 preprocessed images of 23 different varieties of mango. The data has a unique combination of annotation: (1) specific botanical variety (multiclass); (2) internal fruit flavor profile (binary – sweet or sour). This allows machine learning models to be trained for varietal identification as well as for prediction of fruit quality prior to harvest, using leaf morphology and venation. Geographic Location of Data Collection: Mango Orchards, Rajshahi District, Bangladesh. Variables and Labels:The dataset covers 23 total Bangladeshi mango varieties. Categorical Scoring for Flavor (Binary):During data collection, varieties were categorized as "Sweet" or "Sour" based on consultations with experienced local gardeners and agricultural officers. Sour Varieties: Asshina.Sweet Varieties: Aamrupali, BananaMango, Barifor, Dudhkumor, Fazli, Guti, Haribhanga, Himsagar, Jalibandha, Kachamitha, Kathimon, Khirshapat, Misridomdom, Nagra, Patra, Shendri, Surmafajli, Totapuri.Standardized "Sweet" Varieties: Four varieties naturally possess a balanced "sweet-tangy" or "sweet-sour" profile: BaroMashi, BlackStone, Lokna, and Shurjodim. To prevent subjective ambiguity in binary classification tasks, these four varieties were standardized under the "Sweet" label.