Taizhou Urban Solid-Waste Optical Benchmark, Field-Verified Inventory and Analysis-Ready Data
Description
This dataset supports a field-informed framework for urban solid-waste-site identification, spatial interpretation, and management prioritisation in Taizhou, China. It contains three complementary data products. First, a harmonized optical benchmark comprising 6,000 processed 800 × 800 images and 8,033 manually reviewed bounding-box annotations for three waste classes: domestic/mixed waste, construction waste, and industrial solid waste. The benchmark was assembled from four previously published waste-image collections and was further screened, cropped where applicable, standardised, and re-annotated under a unified taxonomy. Second, the release includes a field-verified inventory of 278 retained solid-waste sites established through candidate-directed reconnaissance, including site type and selected field attributes. Third, analysis-ready tables are provided for candidate verification, conditional type refinement, spatial modelling, GeoShapley attribution, network-data evaluation, and uncertainty-aware intervention prioritisation. The optical benchmark retains source-level provenance and attribution information for the original image collections. Source-specific reuse conditions are documented in the accompanying licensing and attribution file. Newly created annotations, harmonized class labels, split definitions, field records, metadata, and derived analytical outputs were produced for this study. The released data are intended to support reproducible research on solid-waste detection, field-informed inventory construction, spatial analysis, and evidence-based waste-management planning.
Files
Steps to reproduce
The optical benchmark is provided with fixed training, validation, and test partitions together with YOLO- and COCO-format annotations. Users can directly train object-detection models using the released image–annotation pairs and the provided three-class taxonomy. The field-verification inventory can be used independently for site-level analysis or linked with the analysis-ready tables through the released identifiers. The analysis-ready files provide the processed inputs and derived outputs used for candidate verification, waste-type refinement, spatial modelling, GeoShapley attribution, network-data evaluation, and intervention prioritisation. File definitions, variable descriptions, dataset structure, source attribution, and reuse conditions are documented in the accompanying README, data dictionaries, and licensing files. The released tables correspond to the frozen analysis used in the associated manuscript; no additional preprocessing is required for reproducing the reported summary analyses.
Institutions
- Zhejiang UniversityZhejiang, Hangzhou