图片转excel vlm 提取手写清单信息 Qwen/Qwen3-VL-235B-A22B-Instruct
Qwen/Qwen3-VL-235B-A22B-Instruct 提取图片数据很好用
# vlm_png_to_excel.py
import json
import os
from PIL import Image
import pandas as pd
from openpyxl import Workbook
from openpyxl.styles import Alignment, Font
from openpyxl.utils import get_column_letter
import tempfile
from flask import Flask, request, jsonify, send_file
from werkzeug.utils import secure_filename
# 添加路径设置
import sys
current_dir = os.path.dirname(os.path.abspath(__file__))
parent_dir = os.path.dirname(current_dir)
if parent_dir not in sys.path:sys.path.insert(0, parent_dir)# 修改导入语句
from llm_server.llm_server import VLMService# Flask 应用实例
app = Flask(__name__)# 创建 VLMService 实例
vlm_service = VLMService()def parse_vlm_response(response_data):"""解析 VLM 返回的结构化数据"""try:# 获取响应内容result = response_data.get("choices", [{}])[0].get("message", {}).get("content", "{}")# 去除可能的代码块标记if result.startswith("```json"):result = result[7:] # 去除开头的 ```jsonif result.startswith("```"):result = result[3:] # 去除可能的 ```if result.endswith("```"):result = result[:-3] # 去除结尾的 ```# 去除首尾空白字符result = result.strip()# 解析JSONparsed = json.loads(result)# 只保留 drawing_info 字段if "drawing_info" in parsed:return {"drawing_info": parsed["drawing_info"]}else:return {}except Exception as e:print(f"解析失败: {e}")return {}def save_data_to_excel(data, output_path="output.xlsx", image_filename=None):"""将结构化数据保存为Excel文件Args:data (dict): 结构化数据output_path (str): 输出Excel文件路径image_filename (str): 原始图像文件名，用于填入Excel前三行合并单元格"""# 创建工作簿和工作表wb = Workbook()ws = wb.active# 如果提供了图像文件名，在前三行合并单元格并填入文件名if image_filename:# 写入文件名到第一行ws["A1"] = image_filename# 合并前三行ws.merge_cells("A1:E3")# 设置居中对齐和字体ws["A1"].alignment = Alignment(horizontal="center", vertical="center")ws["A1"].font = Font(size=14, bold=True)# 从第4行开始写入数据标题start_row = 4else:start_row = 1# 准备数据rows = []# 处理图纸基本信息if "drawing_info" in data and data["drawing_info"]:rows.append(["图纸信息", "", "", "", ""])rows.append(["图号", "长度(mm)", "宽度(mm)", "数量", "备注"])for item in data["drawing_info"]:drawing_number = item.get("drawing_number", "")length = item.get("length", "")width = item.get("width", "")quantity = item.get("quantity", "")remark = item.get("remark", "")rows.append([drawing_number, length, width, quantity, remark])# 移除了 dimensions 和 text 的处理部分# 将数据写入工作表for i, row in enumerate(rows, start=start_row):for j, value in enumerate(row, start=1):ws.cell(row=i, column=j, value=value)# 自动调整列宽for column in ws.columns:max_length = 0column_letter = get_column_letter(column[0].column)for cell in column:try:if len(str(cell.value)) > max_length:max_length = len(str(cell.value))except:passadjusted_width = (max_length + 2)ws.column_dimensions[column_letter].width = adjusted_width# 保存文件wb.save(output_path)print(f"Excel 文件已保存至: {output_path}")return output_pathdef analyze_image_and_generate_excel(image_path):"""主函数：调用 VLM 分析图像，并生成 Excel 文件Args:image_path (str): 输入图像路径output_path (str): 输出Excel文件路径，默认为"dxf_output/data/extracted_data.xlsx""""output_path = r"excel_output/npg2excel/extracted_data.xlsx"# 确保输出目录存在os.makedirs(os.path.dirname(output_path), exist_ok=True)# Step 1: 构造请求消息messages = [{"role": "user","content": '''请提取这张图片的信息，尽可能提取出图号、数量（要算）等信息，图号类似F2N-数字并以 JSON 格式返回。
输出格式要求：
{"drawing_info": [{"drawing_number": "ABC123","length": "1200","width": "800","quantity": "5","remark": "不锈钢板"}]
}'''}]# Step 2: 直接调用 VLMService 而不是通过 HTTP 请求try:result = vlm_service.create_with_image(messages, image_path)print("VLM 返回:", result)# 解析响应extracted_data = parse_vlm_response(result)# 生成 Excelsave_data_to_excel(extracted_data, output_path, os.path.basename(image_path))except Exception as e:print("错误:", str(e))@app.route('/api/analyze-image', methods=['GET'])
def analyze_image_api():"""Flask 接口：通过URL参数指定图片路径，调用VLM分析并生成Excel文件参数:- image_path: 图片路径 (必需)"""try:# 从请求参数中获取图片路径image_path = request.args.get('image_path')if not image_path:return jsonify({"error": "缺少 image_path 参数"}), 400# 检查输入文件是否存在if not os.path.exists(image_path):return jsonify({"error": f"图像文件 '{image_path}' 不存在"}), 400# 创建临时文件保存输出的Excelwith tempfile.NamedTemporaryFile(delete=False, suffix='.xlsx') as tmp_excel:output_path = tmp_excel.name# 调用分析函数analyze_image_and_generate_excel(image_path)# 返回生成的Excel文件return send_file(output_path,as_attachment=True,mimetype='application/vnd.openxmlformats-officedocument.spreadsheetml.sheet')except Exception as e:return jsonify({"error": str(e)}), 500@app.route('/routes')
def show_routes():routes = []for rule in app.url_map.iter_rules():# 获取端点函数的docstringendpoint_func = app.view_functions.get(rule.endpoint)docstring = endpoint_func.__doc__.strip() if endpoint_func and endpoint_func.__doc__ else "无描述"routes.append({'endpoint': rule.endpoint,'methods': list(rule.methods),'rule': str(rule),'description': docstring})return {'routes': routes}if __name__ == '__main__':# 检查是否以脚本方式直接运行if len(sys.argv) == 1:# 使用 input() 获取图像路径print("请输入图像文件路径（输入 'quit' 退出）:")while True:image_path = input("图像路径: ").strip()if image_path.lower() == 'quit':print("程序退出")breakif not os.path.exists(image_path):print(f"错误: 图像文件 '{image_path}' 不存在，请重新输入")continuetry:analyze_image_and_generate_excel(image_path)print("处理完成！")breakexcept Exception as e:print(f"处理过程中出现错误: {e}")continueelse:# Flask 服务器模式import argparseparser = argparse.ArgumentParser()parser.add_argument('--port', type=int, default=5001, help='Port to run the server on')args = parser.parse_args()app.run(host='0.0.0.0', port=args.port, debug=True)