spike: 准备 S1 冒烟构建 + voxel 离线验证

- tools/validate_voxel.py: 两交叉剖面 IDW+maxDist 成体素, 切片可视化; 实证 105k 体素仅 15.9% 有约束(可信体需≥3线/3D网格)
- CMakeLists: VTK find_package 指定 COMPONENTS(VTK9 必需, 否则 VTK_LIBRARIES 空)
- vcpkg.json: 依赖按层递增, 当前仅 gtest(免 S1 被 GDAL/PROJ 拖慢)

CMakeLists.txt

@@ -25,7 +25,14 @@ endif()
 # =====================================================================
 
 find_package(Qt6 REQUIRED COMPONENTS Core Gui Widgets Network Sql Concurrent)
-find_package(VTK REQUIRED)            # 来自 VTK_DIR（预编 install 前缀）；含 GUISupportQt
+# VTK 9 必须指定 COMPONENTS，否则 VTK_LIBRARIES 为空、链接不到任何 VTK。
+# 来自 VTK_DIR（external/vtk-install）。随渲染层增补模块（Volume/Filters 等）。
+find_package(VTK REQUIRED COMPONENTS
+  GUISupportQt
+  RenderingOpenGL2
+  InteractionStyle
+  FiltersSources
+)
 
 # 非 Qt 依赖（vcpkg），随分层逐步启用：
 # find_package(GDAL CONFIG REQUIRED)

tools/validate_voxel.py

@@ -0,0 +1,121 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""
+dd_voxel 离线验证 / 渲染基准：把两条交叉剖面(剖面原数据1+2)三维空间插值(IDW + maxDist 裁剪)
+成体素,切片可视化。用于:
+  1) 提前验证 IDW + 包络裁剪算法(与 Phase 1 core/algo/IdwInterpolator 同逻辑);
+  2) 把"可信体"问题看实——两条 1D 测线只在其所在竖直面附近有约束,体内大部分靠插值/留空;
+  3) 产出 VTK 体绘制/切片的地面真值对照图。
+
+每个测点三维坐标 = (projX, projY, z=ylist[高程]),值 = vlist。
+运行: python tools/validate_voxel.py   输出: docs/_validate/voxel_*.png
+依赖: numpy, scipy, matplotlib
+"""
+import json
+import os
+import sys
+
+import numpy as np
+from scipy.spatial import cKDTree
+import matplotlib
+matplotlib.use("Agg")
+import matplotlib.pyplot as plt
+
+DATA = os.path.join(os.path.dirname(__file__), "..", "docs", "剖面网格数据的色阶数据2等文件")
+OUT = os.path.join(os.path.dirname(__file__), "..", "docs", "_validate")
+
+POWER = 2.0          # IDW 幂
+K = 8                # 取最近 K 个点
+MAX_DIST = 4.0       # 超过此距离(米)→ 留空(NaN),约束插值域(设计 §10)
+DXY = 1.0            # 水平网格步长(米)
+DZ = 0.5             # 垂向步长(米)
+
+
+def load_points(name):
+    d = json.load(open(os.path.join(DATA, name), encoding="utf-8"))["data"]
+    px = np.asarray(d["projectXList"], float)
+    py = np.asarray(d["projectYList"], float)
+    z = np.asarray(d["ylist"], float)          # 高程
+    v = np.asarray(d["vlist"], float)
+    return np.column_stack([px, py, z]), v
+
+
+def idw_voxel(pts, vals, spec):
+    ox, oy, oz, nx, ny, nz, dx, dy, dz = spec
+    tree = cKDTree(pts)
+    # 网格中心点
+    gx = ox + (np.arange(nx) + 0.5) * dx
+    gy = oy + (np.arange(ny) + 0.5) * dy
+    gz = oz + (np.arange(nz) + 0.5) * dz
+    GX, GY, GZ = np.meshgrid(gx, gy, gz, indexing="ij")
+    Q = np.column_stack([GX.ravel(), GY.ravel(), GZ.ravel()])
+    dist, idx = tree.query(Q, k=K)
+    dist = np.maximum(dist, 1e-9)
+    w = 1.0 / dist ** POWER
+    vol = (w * vals[idx]).sum(1) / w.sum(1)
+    vol[dist[:, 0] > MAX_DIST] = np.nan          # 最近点都超距 → 留空
+    return vol.reshape(nx, ny, nz), (gx, gy, gz)
+
+
+def main():
+    os.makedirs(OUT, exist_ok=True)
+    p1, v1 = load_points("剖面原数据1.txt")
+    p2, v2 = load_points("剖面原数据2.txt")
+    pts = np.vstack([p1, p2])
+    vals = np.concatenate([v1, v2])
+
+    # 局部化（减最小，规避大坐标；同 LocalFrame 思路）
+    origin = pts.min(0)
+    ptsL = pts - origin
+    ext = ptsL.max(0)
+    nx = int(ext[0] / DXY) + 1
+    ny = int(ext[1] / DXY) + 1
+    nz = int(ext[2] / DZ) + 1
+    spec = (0, 0, 0, nx, ny, nz, DXY, DXY, DZ)
+    print(f"points={len(vals)} grid={nx}x{ny}x{nz}={nx*ny*nz} cells "
+          f"horiz_extent={ext[0]:.1f}x{ext[1]:.1f}m depth={ext[2]:.1f}m")
+
+    vol, (gx, gy, gz) = idw_voxel(ptsL, vals, spec)
+    filled = np.isfinite(vol).sum()
+    print(f"filled cells={filled} ({100*filled/vol.size:.1f}%) "
+          f"-> 其余为留空(两测线包络外无约束) | v[{np.nanmin(vol):.0f},{np.nanmax(vol):.0f}]")
+
+    vmin, vmax = np.nanpercentile(vol, [2, 98])
+
+    # 1) 中深度水平切片：应显示两条交叉的带(十字支撑)
+    kz = nz // 2
+    plt.figure(figsize=(6, 5))
+    plt.imshow(vol[:, :, kz].T, origin="lower", extent=[0, ext[0], 0, ext[1]],
+               cmap="turbo", vmin=vmin, vmax=vmax)
+    plt.scatter(ptsL[:, 0], ptsL[:, 1], s=1, c="k", alpha=0.15)
+    plt.title(f"voxel 水平切片 z={gz[kz]:.1f} (应见交叉十字支撑)")
+    plt.xlabel("E(m)"); plt.ylabel("N(m)"); plt.colorbar(shrink=.8)
+    plt.tight_layout(); plt.savefig(f"{OUT}/voxel_hslice.png", dpi=90); plt.close()
+
+    # 2) 沿 E 的垂直切片（取 profile1 附近 N）：应重现剖面竖直结构
+    jy = ny // 2
+    plt.figure(figsize=(8, 3))
+    plt.imshow(vol[:, jy, :].T, origin="lower", extent=[0, ext[0], 0, ext[2]],
+               cmap="turbo", vmin=vmin, vmax=vmax, aspect="auto")
+    plt.title(f"voxel 垂直切片 N={gy[jy]:.1f}")
+    plt.xlabel("E(m)"); plt.ylabel("elev(m)"); plt.colorbar(shrink=.8)
+    plt.tight_layout(); plt.savefig(f"{OUT}/voxel_vslice.png", dpi=90); plt.close()
+
+    # 3) 非留空区的三维散点（直观看"十字片"支撑）
+    xi, yi, zi = np.where(np.isfinite(vol))
+    sub = slice(None, None, max(1, len(xi) // 6000))
+    fig = plt.figure(figsize=(7, 5)); ax = fig.add_subplot(111, projection="3d")
+    sc = ax.scatter(gx[xi[sub]], gy[yi[sub]], gz[zi[sub]],
+                    c=vol[xi[sub], yi[sub], zi[sub]], cmap="turbo",
+                    vmin=vmin, vmax=vmax, s=3)
+    ax.set_title("voxel 非留空区(两测线竖直面附近=十字片)")
+    ax.set_xlabel("E"); ax.set_ylabel("N"); ax.set_zlabel("elev")
+    fig.colorbar(sc, shrink=.6)
+    plt.tight_layout(); plt.savefig(f"{OUT}/voxel_support.png", dpi=90); plt.close()
+
+    print("written:", sorted(f for f in os.listdir(OUT) if f.startswith("voxel")))
+
+
+if __name__ == "__main__":
+    sys.stdout.reconfigure(encoding="utf-8")
+    main()

vcpkg.json

@@ -1,15 +1,8 @@
 {
   "name": "geopro-desktop",
   "version": "0.1.0",
-  "description": "Geopro 3.0 desktop client (Qt6 + VTK9) - M1. 方案②-修订: 仅非 Qt 依赖走 vcpkg; Qt/VTK/ADS/QtKeychain 对接官方 MSVC Qt(见 docs/ENV_SETUP_Windows.md)",
+  "description": "Geopro 3.0 desktop client (Qt6 + VTK9) - M1. 方案②-修订: Qt/VTK/ADS/QtKeychain 对接官方 MSVC Qt(不走 vcpkg); 仅非 Qt 依赖走 vcpkg, 按层递增。",
   "dependencies": [
-    "gdal",
-    "proj",
-    "eigen3",
-    "spdlog",
-    "fmt",
-    "nlohmann-json",
-    "openssl",
     "gtest"
   ]
 }