PRA-only-demo/packages/rpa-core/src/capabilities/ocr-caps.ts

import path from "node:path";
import { promises as fs } from "node:fs";
import { register } from "../engine/registry";
import { RuntimeError } from "../engine/types";
import { defaultEngine } from "../engine/engine";

interface OcrWord { text: string; x: number; y: number; width: number; height: number; confidence: number }
interface RapidOcrLine { text: string; mean: number; box?: number[][] }

async function getOcr(): Promise<{ detect: (image: string) => Promise<{ texts: RapidOcrLine[] }> }> {
  if (defaultEngine.ocrInstance) return defaultEngine.ocrInstance as { detect: (image: string) => Promise<{ texts: RapidOcrLine[] }> };
  const mod = await import("@gutenye/ocr-node");
  const instance = await mod.default.create();
  defaultEngine.ocrInstance = instance;
  return instance;
}

function requireImage(args: Record<string, unknown>, nodeId: string): string {
  // uiSelector 是 OcrOffline_GetOcrTextNew 目录里唯一的图片字段（P0-1 修复）
  for (const key of ["image_path", "imagePath", "file_path", "filePath", "path", "image", "pictureSource", "uiSelector"]) {
    const v = String(args[key] ?? "").trim();
    if (v) return v;
  }
  throw new RuntimeError("validation_error", "图片路径为空", { nodeId, componentId: nodeId, recommendation: "填写本地图片路径（png/jpg/bmp）" });
}

function boxToRect(box: number[][] | undefined): { x: number; y: number; width: number; height: number } {
  const points = box ?? [];
  const xs = points.map(p => p[0]);
  const ys = points.map(p => p[1]);
  const minX = Math.min(...(xs.length ? xs : [0]));
  const minY = Math.min(...(ys.length ? ys : [0]));
  const maxX = Math.max(...(xs.length ? xs : [0]));
  const maxY = Math.max(...(ys.length ? ys : [0]));
  return { x: minX, y: minY, width: maxX - minX, height: maxY - minY };
}

register("OcrOffline_GetOcrTextNew", async (ctx, node, args) => {
  const imagePath = requireImage(args, node.id);
  const ocr = await getOcr();
  const result = await ocr.detect(imagePath);
  const text = result.texts.map(t => t.text).join("\n");
  ctx.log("OCR 识别完成：" + text.length + " 字符 / " + result.texts.length + " 行", "info", node.id);
  return text;
});

register("OcrOffline_GetOcrTextPosition", async (ctx, node, args) => {
  const imagePath = requireImage(args, node.id);
  const ocr = await getOcr();
  const result = await ocr.detect(imagePath);
  const words = result.texts.map(t => {
    const rect = boxToRect(t.box);
    return { text: t.text, x: rect.x, y: rect.y, width: rect.width, height: rect.height, confidence: t.mean };
  });
  ctx.log("OCR 定位完成：" + words.length + " 行", "info", node.id);
  return words;
});

/** 按垂直中心聚类成行（行高 60% 为阈值），行内按 x 排序，返回二维文本表格 */
function groupOcrRows(words: RapidOcrLine[]): string[][] {
  const prepared = words
    .filter((word) => String(word.text ?? "").trim())
    .map((word) => {
      const box = word.box ?? [];
      const xs = box.map((point) => point[0]);
      const ys = box.map((point) => point[1]);
      const x = xs.length ? Math.min(...xs) : 0;
      const y = ys.length ? Math.min(...ys) : 0;
      const height = ys.length ? Math.max(...ys) - y : 0;
      return { text: word.text, x, cy: y + height / 2, h: height };
    })
    .sort((a, b) => a.cy - b.cy || a.x - b.x);
  const rows: Array<{ cy: number; h: number; cells: typeof prepared }> = [];
  let current: (typeof rows)[number] | null = null;
  for (const item of prepared) {
    if (!current || Math.abs(item.cy - current.cy) > Math.max(current.h, item.h, 8) * 0.6) {
      current = { cy: item.cy, h: item.h, cells: [] };
      rows.push(current);
    }
    current.cells.push(item);
  }
  return rows.map((row) => row.cells.sort((a, b) => a.x - b.x).map((cell) => cell.text.trim()));
}

register("OcrOffline_GetOcrTable", async (ctx, node, args) => {
  const imagePath = requireImage(args, node.id);
  const ocr = await getOcr();
  const result = await ocr.detect(imagePath);
  const rows = groupOcrRows(result.texts);
  ctx.log("OCR 表格识别完成：" + rows.length + " 行", "info", node.id);

  // showTable=1：写入流程数据表（与 Bivartable 系列组件共享同一份数据簿）
  if (String(args.showTable ?? "1") === "1") {
    ctx.vars["__rpaDatabook"] = rows;
    ctx.log("已写入流程数据表", "info", node.id);
  }

  // isExcel=1：另存为 Excel 文件
  let excelFile = "";
  if (String(args.isExcel ?? "0") === "1") {
    const ExcelJS = (await import("exceljs")).default;
    const workbook = new ExcelJS.Workbook();
    const sheet = workbook.addWorksheet("Sheet1");
    sheet.addRows(rows);
    const rawFolder = String(args.excelPath ?? "").trim();
    // ${res} 流程资源目录占位与空值统一落到运行证据目录
    const folder = rawFolder && rawFolder !== "${res}" ? rawFolder : ctx.evidenceDir;
    const fileName = String(args.excelName ?? "表格数据.xlsx").replace(/\$\{randomString\}/g, Math.random().toString(36).slice(2, 8));
    excelFile = path.resolve(path.isAbsolute(fileName) ? fileName : path.join(folder, fileName));
    await fs.mkdir(path.dirname(excelFile), { recursive: true });
    await workbook.xlsx.writeFile(excelFile);
    ctx.log("已保存 Excel：" + excelFile, "info", node.id);
  }

  return excelFile ? { rows, excelFile } : rows;
});