AgroUAV-209K Weed Detection Dataset

The UAV Undesirable Plants Dataset is a large-scale field-level observational dataset designed to support multi-label detection of crop and weed species in precision agriculture environments. The dataset contains 209,600 geo-referenced observations collected from heterogeneous agricultural fields using UAV-mounted multispectral imaging platforms and complementary agronomic sensing systems. Each record represents a spatially localized crop–weed observation unit derived from UAV imagery and associated field telemetry. The dataset integrates spectral reflectance measurements, vegetation indices, texture descriptors, geometric crop structure indicators, environmental conditions, and farm management metadata, enabling comprehensive modeling of weed presence under agronomic variability. Data acquisition spans diverse crop types, soil conditions, irrigation regimes, and seasonal growth stages, thereby capturing the non-IID characteristics typical of operational precision agriculture deployments. Feature Categories 1. Field Metadata and Operational Context These variables describe agronomic management and UAV acquisition conditions. crop_type: Encodes the primary cultivated crop in the field (e.g., wheat, maize, rice, soybean, sugar beet). Crop context strongly influences weed ecology and spectral response. soil_type: Soil classification indicator (loam, clay, sandy, silt), affecting moisture retention and weed emergence patterns. irrigation_regime: Irrigation method (rainfed, drip, sprinkler), reflecting water distribution heterogeneity. herbicide_days: Number of days since last herbicide application, capturing management recency effects. sowing_doy: Day-of-year of crop sowing, representing crop growth stage variability. flight_altitude_m: UAV flight altitude during acquisition, influencing spatial resolution. gsd_cm_per_px: Ground sampling distance, representing image spatial detail. sun_elev_deg: Solar elevation angle at acquisition time, affecting illumination geometry. wind_speed_ms: Wind speed during flight, potentially impacting canopy motion and image sharpness. 2. Spectral Reflectance Features These features represent UAV-derived surface reflectance proxies. R_reflectance, G_reflectance, B_reflectance: Visible-band reflectance components capturing plant pigmentation differences. NIR_reflectance: Near-infrared response strongly associated with vegetation vigor. RedEdge_reflectance: Sensitive to chlorophyll concentration and early stress signals. These spectral measurements form the foundation for vegetation discrimination between crops and weed species. 3. Vegetation Indices Standard agronomic indices derived from multispectral bands. NDVI: Normalized Difference Vegetation Index, indicating overall vegetation vigor. GNDVI: Green NDVI, sensitive to chlorophyll concentration. SAVI: Soil-Adjusted Vegetation Index, compensating for soil background effects. EVI: Enhanced Vegetation Index, improving sensitivity in dense vegetation. VARI: Visible Atmospherically Resistant Index, useful under varying illumination. These indices enhance separability between crop canopy and weed infestations. 4. Texture and Morphological Descriptors Texture metrics extracted from UAV imagery characterize spatial canopy patterns. texture_contrast: Measures local intensity variation, highlighting heterogeneous weed patches. texture_homogeneity: Indicates uniform canopy regions. texture_entropy: Captures randomness in spatial structure. edge_density: Density of detected edges, often higher in mixed vegetation. canopy_fragmentation: Degree of canopy discontinuity, associated with weed intrusion. row_alignment_dev_deg: Deviation from expected crop row orientation, useful for detecting off-row weeds. These features are particularly informative for structural weed detection beyond pure spectral cues. 5. Geometric and Structural Indicators These variables describe plant morphology and spatial organization. plant_height_m: Relative canopy height proxy derived from UAV observations. canopy_volume: Estimated volumetric canopy measure. surface_roughness: Micro-structural variability of the canopy surface. dispersion_index: Spatial dispersion of vegetation clusters. Such features help differentiate crop stands from irregular weed growth. 6. Environmental Field Conditions Environmental measurements provide context for plant growth dynamics. soil_moisture: Surface moisture proxy influencing germination likelihood. temperature_c: Ambient temperature at acquisition. solar_radiation_wm2: Incident solar radiation intensity. These variables capture environmental drivers of weed emergence. 7. Derived Agronomic Indicators Higher-level indicators synthesized from primary measurements. vegetation_vigor: Composite vigor score derived from vegetation indices. stress_score: Proxy indicator of vegetation stress combining spectral and texture signals. These features provide compact representations of crop health status. Target Labels (Multi-Label) Each observation may contain multiple labels simultaneously. Crop: Presence of healthy crop canopy. Thistle: Detection of thistle weed species. Ragweed: Detection of ragweed. Crabgrass: Detection of crabgrass. Wild_Oat: Detection of wild oat. Broadleaf_Weed: General broadleaf weed presence. Mixed_Weeds: Co-occurrence of multiple weed types. Early_Growth_Weeds: Weeds in early phenological stage. Dense_Weed_Cluster: High-density weed infestation zone. The multi-label formulation reflects field conditions where multiple weed species and growth stages often coexist within the same UAV observation footprint.