A time-resolved image-environment dataset for tatsoi growth monitoring and intelligent phytoprotection benchmarking

Image-based crop monitoring is now widely used in sustainable plant production, but public datasets that combine visual canopy appearance with synchronized local environmental measurements remain scarce. As a result, models are difficult to evaluate under field variation that must operate under field variation, including changes in plant growth, microclimate, soil conditions, pest pressure, illumination, and acquisition date. We present a time-resolved tatsoi ( Brassica campestris L. ssp. chinensis var. rosularis Tsen) image-environment dataset collected during winter cultivation from 12 November 2025 to 28 February 2026. The dataset contains 3456 plant-level images from 128 repeatedly imaged plants across 27 acquisition dates. Each image is linked to Labelme polygon annotations of the tatsoi canopy, derived bounding boxes and segmentation masks, plant identity, acquisition date, seven phenotype or condition labels, and synchronized field solar insecticidal lamp Internet of Things (SIL-IoT) records. The sensor-derived metadata used in the baseline analyses include air temperature and humidity, illumination, rain-status records, device-level pest-signal counts, soil moisture and temperature, wind speed, and wind direction. These metadata are provided as nearest sensor readings and as 1 h, 6 h, and 24 h pre-capture aggregates. Standardized random, temporal, and plant-holdout splits are provided to support reproducible evaluation. Baseline benchmarks showed that the resource can be reused for canopy detection, canopy segmentation, and image-environment trait recognition. Under the plant-holdout split, YOLOv8n reached a box mAP50–95 of 0.887, RetinaNet-R50 reached a box mAP50–95 of 0.826 among torchvision detectors, and Swin-T with environmental features reached a mean macro-F1 of 0.819 across plant-level traits. The dataset provides a reusable resource for multimodal crop phenotyping, field-condition visual benchmarking, and intelligent phytoprotection research.