feat(server): add workbench API, DSH planner worker and RAG pipeline

2026-09-16 19:06:20 +08:00
parent 72fc5b47c7
commit d4b5b3f43a
122 changed files with 31906 additions and 0 deletions
@@ -0,0 +1,116 @@
+"""导图图片 OCR：PowerShell WinRT Windows.Media.Ocr（zh-Hans-CN），坐标还原阅读顺序"""
+import re
+import subprocess
+import tempfile
+from pathlib import Path
+
+from .config import OCR_DIR
+
+# WinRT StorageFile 无法处理含中文的路径 → 中间文件一律用纯 ASCII 临时目录
+ASCII_TMP = Path(tempfile.gettempdir()) / "yawen_ocr"
+
+PS_TEMPLATE = r"""$ErrorActionPreference = 'Stop'
+Add-Type -AssemblyName System.Runtime.WindowsRuntime
+$null = [Windows.Storage.StorageFile, Windows.Storage, ContentType = WindowsRuntime]
+$null = [Windows.Media.Ocr.OcrEngine, Windows.Foundation, ContentType = WindowsRuntime]
+$null = [Windows.Foundation.IAsyncOperation`1, Windows.Foundation, ContentType = WindowsRuntime]
+$null = [Windows.Graphics.Imaging.BitmapDecoder, Windows.Foundation, ContentType = WindowsRuntime]
+$asTaskGeneric = ([System.WindowsRuntimeSystemExtensions].GetMethods() | Where-Object { $_.Name -eq 'AsTask' -and $_.GetParameters().Count -eq 1 -and $_.GetParameters()[0].ParameterType.Name -eq 'IAsyncOperation`1' })[0]
+function Await($WinRtTask, $ResultType) {
+    $asTask = $asTaskGeneric.MakeGenericMethod($ResultType)
+    $netTask = $asTask.Invoke($null, @($WinRtTask))
+    try { $netTask.GetAwaiter().GetResult() }
+    catch {
+        $ex = $_.Exception
+        while ($ex.InnerException) { $ex = $ex.InnerException }
+        Write-Output "AWAIT-ERR: $($ex.GetType().FullName): $($ex.Message)"
+        throw
+    }
+}
+$imgPath = '{{IMG}}'
+$outFile = '{{OUT}}'
+$file = Await ([Windows.Storage.StorageFile]::GetFileFromPathAsync($imgPath)) ([Windows.Storage.StorageFile])
+$stream = Await ($file.OpenAsync([Windows.Storage.FileAccessMode]::Read)) ([Windows.Storage.Streams.IRandomAccessStream])
+$decoder = Await ([Windows.Graphics.Imaging.BitmapDecoder]::CreateAsync($stream)) ([Windows.Graphics.Imaging.BitmapDecoder])
+$bitmap = Await ($decoder.GetSoftwareBitmapAsync()) ([Windows.Graphics.Imaging.SoftwareBitmap])
+$engine = [Windows.Media.Ocr.OcrEngine]::TryCreateFromUserProfileLanguages()
+if ($null -eq $engine) {
+    $engine = [Windows.Media.Ocr.OcrEngine]::TryCreateFromLanguage([Windows.Globalization.Language]::new('zh-Hans-CN'))
+}
+$result = Await ($engine.RecognizeAsync($bitmap)) ([Windows.Media.Ocr.OcrResult])
+$sb = New-Object System.Text.StringBuilder
+foreach ($line in $result.Lines) {
+    $firstWord = $line.Words | Select-Object -First 1
+    $x = [int]$firstWord.BoundingRect.X
+    $y = [int]$firstWord.BoundingRect.Y
+    $text = $line.Text
+    $null = $sb.AppendLine("[$x,$y] $text")
+}
+[System.IO.File]::WriteAllText($outFile, $sb.ToString(), [System.Text.Encoding]::UTF8)
+Write-Host "DONE lines=$($result.Lines.Count)"
+"""
+
+def _reorder(lines: list[tuple[int, int, str]]) -> str:
+    """按坐标还原导图阅读顺序：行优先（y 容差 60px），行内按 x"""
+    if not lines:
+        return ""
+    lines.sort(key=lambda t: t[1])
+    rows: list[list[tuple[int, int, str]]] = []
+    cur_y, cur_row = None, []
+    for x, y, text in lines:
+        if cur_y is None or abs(y - cur_y) <= 60:
+            cur_row.append((x, y, text))
+            cur_y = y if cur_y is None else min(cur_y, y)
+        else:
+            rows.append(sorted(cur_row, key=lambda t: t[0]))
+            cur_row, cur_y = [(x, y, text)], y
+    if cur_row:
+        rows.append(sorted(cur_row, key=lambda t: t[0]))
+    return "\n".join(" ".join(t[2] for t in row) for row in rows)
+
+
+def _prepare(img_path: str) -> str:
+    """预处理：EXIF 转正 + 缩放到 OCR 引擎可处理尺寸（≤2000px），存纯 ASCII 路径"""
+    try:
+        from PIL import Image, ImageOps
+        im = Image.open(img_path)
+        im = ImageOps.exif_transpose(im)
+        w, h = im.size
+        if max(w, h) > 2000:
+            r = 2000 / max(w, h)
+            im = im.resize((int(w * r), int(h * r)), Image.LANCZOS)
+        ASCII_TMP.mkdir(parents=True, exist_ok=True)
+        small = ASCII_TMP / f"{Path(img_path).stem}_small.jpg"
+        im.convert("RGB").save(small, quality=92)
+        return str(small)
+    except Exception:
+        return img_path  # PIL 失败就用原图
+
+def ocr_image(img_path: str) -> dict:
+    """OCR 一张图片，返回排序后的文本"""
+    target = _prepare(img_path)
+    img = Path(target)
+    OCR_DIR.mkdir(parents=True, exist_ok=True)
+    out = OCR_DIR / f"{img.stem}_ocr.txt"
+    ps = OCR_DIR / f"{img.stem}_ocr.ps1"
+    ps.write_text(PS_TEMPLATE.replace("{{IMG}}", str(img))
+                                .replace("{{OUT}}", str(out)),
+                  encoding="utf-8-sig")  # BOM：PowerShell 5.1 按 UTF-8 解析脚本
+    try:
+        r = subprocess.run(["powershell", "-NoProfile", "-ExecutionPolicy", "Bypass",
+                            "-File", str(ps)], capture_output=True, text=True,
+                           encoding="gbk", errors="replace", timeout=180)
+    except Exception as e:
+        return {"ok": False, "error": f"OCR 进程失败: {e}"}
+    if not out.exists():
+        return {"ok": False, "error": (r.stderr or r.stdout or "OCR 无输出")[:300]}
+    raw = out.read_text(encoding="utf-8", errors="ignore")
+    lines = []
+    for ln in raw.splitlines():
+        m = re.match(r"\[(\d+),(\d+)\] (.*)", ln)
+        if m:
+            lines.append((int(m.group(1)), int(m.group(2)), m.group(3).strip()))
+    text = _reorder(lines)
+    if not text.strip():
+        return {"ok": False, "error": "OCR 未识别到文字（图片可能不清晰）"}
+    return {"ok": True, "text": text, "lines": len(lines)}