Install
openclaw skills install @ruiduobao/geoskill-object-detection-yolo滑窗+HOG/阈值特征+NMS+地理编码,输出检测框GeoJSON(离线numpy等价实现)
openclaw skills install @ruiduobao/geoskill-object-detection-yoloAutomatically detects targets of interest in remote sensing imagery (bright/high-contrast features such as buildings, vehicles, and ships), and outputs detection-box GeoJSON with WGS-84 geographic coordinates plus a score raster.
This skill is an offline numpy-equivalent implementation of deep object detectors such as YOLO: without relying on torch/ultralytics, it fully reproduces the detection pipeline of "candidate generation → redundancy suppression → geographic output" via sliding-window scanning + objectness scoring (local brightness z-score or HOG gradient energy) + non-maximum suppression (NMS) + pixel-box geocoding, and every step can be verified by unit tests independently.
pip install numpy rasterio scipy scikit-learn geopandas shapely
python geoskill-object-detection-yolo.py --bbox 116.0 39.0 117.0 40.0 --synthetic --output-dir ./out
python geoskill-object-detection-yolo.py --input scene.tif --feature hog --score-thresh 2.0 --output-dir ./out
python geoskill-object-detection-yolo.py --bbox 121.4 31.1 121.6 31.3 --synthetic --win-size 12 --step 4 --iou-thresh 0.3 --output-dir ./out
| File | Format | Description |
|---|---|---|
detections.geojson | GeoJSON | Detection-box polygons + score/pixel_box attributes |
score_map.tif | GeoTIFF | Detection score raster (max within boxes) |
output-manifest.json | JSON | Run manifest (inputs/outputs/QA/exit code) |
Local single/multi-band GeoTIFF (first band used as intensity), or --synthetic scene (dark background + bright square targets).
--synthetic mode requires no network at all.MIT
在遥感影像上自动检测感兴趣目标(建筑、车辆、船只等明亮/高对比地物),输出带 WGS-84 地理坐标的检测框 GeoJSON 与得分栅格。
本 skill 是 YOLO 等深度目标检测器的离线 numpy 等价实现:不依赖 torch/ultralytics,用滑窗扫描 + 目标性打分(局部亮度 z-score 或 HOG 梯度能量)+ 非极大值抑制(NMS)+ 像素框地理编码,完整复现"候选生成 -> 冗余抑制 -> 地理输出"的检测流水线,每一步都可单独单元测试验证。
pip install numpy rasterio scipy scikit-learn geopandas shapely
python geoskill-object-detection-yolo.py --bbox 116.0 39.0 117.0 40.0 --synthetic --output-dir ./out
python geoskill-object-detection-yolo.py --input scene.tif --feature hog --score-thresh 2.0 --output-dir ./out
python geoskill-object-detection-yolo.py --bbox 121.4 31.1 121.6 31.3 --synthetic --win-size 12 --step 4 --iou-thresh 0.3 --output-dir ./out
| 文件 | 格式 | 说明 |
|---|---|---|
detections.geojson | GeoJSON | 检测框多边形 + score/pixel_box 属性 |
score_map.tif | GeoTIFF | 检测得分栅格(框内最大值) |
output-manifest.json | JSON | 运行清单(输入/输出/QA/退出码) |
本地单/多波段 GeoTIFF(取首波段作强度),或 --synthetic 合成场景(暗背景 + 明亮方形目标)。
--synthetic 模式完全无网络。MIT