Environmental and land-management organizations increasingly capture high-resolution aerial imagery with consumer drones, yet converting that footage into actionable land-cover insight requires deep-learning expertise those organizations rarely possess, and public training data poorly represents forest-health-critical conditions such as stressed or dead canopy. This study applies a design-science methodology to address both problems jointly. First, we report a multi-country drone-imagery data collection conducted across three European countries under a standardised acquisition protocol (a controlled altitude range, strict nadir angle, controlled solar window), annotated through a human-in-the-loop pipeline against an eight-class forest-health schema. Second, we present an end-to-end information system, built on this data foundation, that serves a fine-tuned SegFormer-B4 model as an independent inference microservice behind a web decision-support interface returning interpretable, forestry-specific metrics rather than raw model output. Within the configurations tested, training-data domain and thematic alignment was associated with segmentation performance more strongly than training-set volume: replacing satellite-derived data with drone-native data raised mean intersection-over-union (mIoU) from 0.45 to 0.715, while the dead and stressed tree class remained the weakest across all public-data configurations, consistent with existing benchmarks depicting bare winter branches rather than the pale, stressed foliage seen in real deployments. Retraining on a purpose-built collection from three European countries raised overall mIoU further to 0.7759 and improved dead-tree IoU from 0.34 to 0.6459, moving it above the internal reference value used to flag weak classes during training. We report the acquisition protocol, training configuration, interpretable metric design, and limitations, including a disclosed frame-level validation split, a single-run comparison, and a partially varying augmentation pipeline across configurations, positioning the combined artifact as a documented example of an accessible applied-AI system in environmental and green-tourism land management.