feat: 扫描版PDF支持视觉模型OCR识别

- 添加 pdf_to_images 将PDF页面转为图像
- 添加 extract_text_from_image 使用视觉模型OCR识别图像文字
- 检测扫描版PDF自动切换OCR模式
- glm-4.6v 等视觉模型可识别图像中的文字
- 进度提示显示OCR识别过程

app.py

@@ -445,6 +445,36 @@ def upload_pdf():
     })
 
 
+@app.route('/api/config')
+def api_config():
+    """获取系统配置"""
+    from admin import get_llm_config, get_site_config
+    
+    return jsonify({
+        'site_name': get_site_config().get('site_name'),
+        'max_file_size': get_site_config().get('max_file_size'),
+        'cache_expire_days': get_site_config().get('cache_expire_days'),
+        'llm_config': get_llm_config()
+    })
+
+
+@app.route('/api/translations')
+def api_translations_list():
+    """获取翻译记录列表"""
+    user = get_current_user()
+    
+    if user:
+        translations = Translation.query.filter_by(user_id=user.id)\
+            .order_by(Translation.created_at.desc()).limit(20).all()
+    else:
+        # 访客返回空列表
+        translations = []
+    
+    return jsonify({
+        'translations': [t.to_dict() for t in translations]
+    })
+
+
 @app.route('/api/status/<int:translation_id>')
 def translation_status(translation_id):
     """获取翻译状态"""
@@ -462,8 +492,12 @@ def translation_status(translation_id):
         'id': translation.id,
         'status': translation.status,
         'progress': translation.progress,
+        'filename': translation.original_filename,
+        'pages': translation.page_count,
         'from_cache': translation.from_cache,
-        'error': translation.error_message
+        'error': translation.error_message,
+        'created_at': translation.created_at.isoformat() if translation.created_at else None,
+        'completed_at': translation.completed_at.isoformat() if translation.completed_at else None,
     })
 
 

pdf_translate.py

@@ -0,0 +1,307 @@
+#!/usr/bin/env python3
+"""
+PDF翻译命令行工具
+
+使用方法:
+    pdf_translate translate <pdf_file> [--instruction "翻译要求"] [--user <username>] [--password <password>]
+    pdf_translate list [--user <username>] [--password <password>]
+    pdf_translate status <translation_id> [--user <username>] [--password <password>]
+    pdf_translate download <translation_id> <output_file> [--user <username>] [--password <password>]
+    pdf_translate config
+"""
+
+import argparse
+import sys
+import os
+import json
+import requests
+from pathlib import Path
+
+# API 基础地址
+API_BASE = "http://localhost:19000"
+
+def login(username, password, api_base=API_BASE):
+    """登录获取session"""
+    url = f"{api_base}/api/login"
+    data = {"username": username, "password": password}
+    
+    try:
+        response = requests.post(url, json=data, timeout=10)
+        result = response.json()
+        
+        if result.get('success'):
+            print(f"✅ 登录成功: {username}")
+            return response.cookies
+        else:
+            print(f"❌ 登录失败: {result.get('error', '未知错误')}")
+            return None
+    except Exception as e:
+        print(f"❌ 登录请求失败: {e}")
+        return None
+
+def get_user_info(cookies, api_base=API_BASE):
+    """获取用户信息"""
+    url = f"{api_base}/api/user/info"
+    try:
+        response = requests.get(url, cookies=cookies, timeout=10)
+        result = response.json()
+        return result.get('user')
+    except:
+        return None
+
+def translate_pdf(pdf_file, instruction=None, cookies=None, api_base=API_BASE):
+    """上传并翻译PDF"""
+    if not os.path.exists(pdf_file):
+        print(f"❌ 文件不存在: {pdf_file}")
+        return None
+    
+    url = f"{api_base}/api/upload"
+    
+    print(f"📤 上传文件: {pdf_file}")
+    
+    try:
+        with open(pdf_file, 'rb') as f:
+            files = {'file': f}
+            data = {}
+            if instruction:
+                data['instruction'] = instruction
+            
+            response = requests.post(url, files=files, data=data, cookies=cookies, timeout=60)
+            result = response.json()
+        
+        if response.status_code != 200:
+            print(f"❌ 上传失败: {result.get('error', '未知错误')}")
+            return None
+        
+        translation_id = result.get('translation_id')
+        task_id = result.get('task_id')
+        from_cache = result.get('from_cache', False)
+        
+        print(f"✅ 上传成功!")
+        print(f"   翻译ID: {translation_id}")
+        print(f"   任务ID: {task_id}")
+        print(f"   使用缓存: {'是' if from_cache else '否'}")
+        
+        # 等待翻译完成
+        if not from_cache:
+            print(f"\n🔄 等待翻译完成...")
+            poll_translation_status(translation_id, cookies, api_base)
+        
+        return translation_id
+        
+    except Exception as e:
+        print(f"❌ 上传请求失败: {e}")
+        return None
+
+def poll_translation_status(translation_id, cookies=None, api_base=API_BASE):
+    """轮询翻译状态"""
+    url = f"{api_base}/api/status/{translation_id}"
+    
+    import time
+    max_wait = 300  # 最大等待5分钟
+    waited = 0
+    
+    while waited < max_wait:
+        try:
+            response = requests.get(url, cookies=cookies, timeout=10)
+            result = response.json()
+            
+            status = result.get('status', 'unknown')
+            progress = result.get('progress', 0)
+            
+            print(f"   状态: {status}, 进度: {progress}%")
+            
+            if status == 'completed':
+                print(f"✅ 翻译完成!")
+                return True
+            elif status == 'failed':
+                print(f"❌ 翻译失败: {result.get('error', '未知错误')}")
+                return False
+            
+            time.sleep(5)
+            waited += 5
+            
+        except Exception as e:
+            print(f"⚠️ 状态查询失败: {e}")
+            time.sleep(5)
+            waited += 5
+    
+    print(f"⚠️ 等待超时，请在网页查看结果")
+    return False
+
+def list_translations(cookies=None, api_base=API_BASE):
+    """列出翻译记录"""
+    url = f"{api_base}/api/translations"
+    
+    try:
+        response = requests.get(url, cookies=cookies, timeout=10)
+        result = response.json()
+        
+        translations = result.get('translations', [])
+        
+        if not translations:
+            print("暂无翻译记录")
+            return
+        
+        print(f"\n📋 翻译记录 (共{len(translations)}条):\n")
+        print("ID    | 文件名           | 状态      | 进度 | 时间")
+        print("-" * 60)
+        
+        for t in translations:
+            id_str = str(t.get('id', '?'))[:4]
+            filename = t.get('filename', '?')[:15]
+            status = t.get('status', '?')
+            progress = t.get('progress', 0)
+            created = t.get('created_at', '?')
+            if created and created != '?':
+                created = created[5:16]  # 月-日 时:分
+            
+            print(f"{id_str:4} | {filename:15} | {status:8} | {progress:3}% | {created}")
+        
+    except Exception as e:
+        print(f"❌ 获取列表失败: {e}")
+
+def get_translation_status(translation_id, cookies=None, api_base=API_BASE):
+    """获取翻译状态"""
+    url = f"{api_base}/api/status/{translation_id}"
+    
+    try:
+        response = requests.get(url, cookies=cookies, timeout=10)
+        result = response.json()
+        
+        print(f"\n📊 翻译状态 (ID: {translation_id}):\n")
+        print(f"   状态: {result.get('status', '未知')}")
+        print(f"   进度: {result.get('progress', 0)}%")
+        print(f"   文件: {result.get('filename', '未知')}")
+        print(f"   页数: {result.get('pages', '?')}")
+        print(f"   缓存: {'是' if result.get('from_cache') else '否'}")
+        
+        if result.get('error'):
+            print(f"   错误: {result.get('error')}")
+        
+    except Exception as e:
+        print(f"❌ 获取状态失败: {e}")
+
+def download_translation(translation_id, output_file, cookies=None, api_base=API_BASE):
+    """下载翻译结果"""
+    url = f"{api_base}/api/download/{translation_id}"
+    
+    try:
+        response = requests.get(url, cookies=cookies, timeout=30)
+        
+        if response.status_code != 200:
+            print(f"❌ 下载失败: {response.status_code}")
+            return False
+        
+        with open(output_file, 'wb') as f:
+            f.write(response.content)
+        
+        print(f"✅ 下载成功: {output_file}")
+        return True
+        
+    except Exception as e:
+        print(f"❌ 下载失败: {e}")
+        return False
+
+def show_config(api_base=API_BASE):
+    """显示当前大模型配置"""
+    url = f"{api_base}/api/config"
+    
+    try:
+        response = requests.get(url, timeout=10)
+        result = response.json()
+        
+        print(f"\n⚙️ 当前配置:\n")
+        print(f"   网站名称: {result.get('site_name', '未知')}")
+        print(f"   最大文件: {result.get('max_file_size', '?')}MB")
+        print(f"   缓存天数: {result.get('cache_expire_days', '?')}天")
+        
+        llm = result.get('llm_config', {})
+        print(f"   大模型API: {llm.get('api_base', '未设置')}")
+        print(f"   模型: {llm.get('model', '未设置')}")
+        
+    except Exception as e:
+        print(f"❌ 获取配置失败: {e}")
+
+def main():
+    parser = argparse.ArgumentParser(
+        description="PDF翻译命令行工具",
+        formatter_class=argparse.RawDescriptionHelpFormatter,
+        epilog="""
+示例:
+  pdf_translate translate test.pdf --user admin --password admin123
+  pdf_translate translate test.pdf --instruction "保持专业术语原文"
+  pdf_translate list --user admin --password admin123
+  pdf_translate status 1
+  pdf_translate download 1 result.md
+  pdf_translate config
+        """
+    )
+    
+    parser.add_argument('--user', '-u', help='用户名')
+    parser.add_argument('--password', '-p', help='密码')
+    parser.add_argument('--api', default=API_BASE, help=f'API地址 (默认: {API_BASE})')
+    
+    subparsers = parser.add_subparsers(dest='command', help='命令')
+    
+    # translate 命令
+    trans_parser = subparsers.add_parser('translate', help='翻译PDF文件')
+    trans_parser.add_argument('file', help='PDF文件路径')
+    trans_parser.add_argument('--instruction', '-i', help='翻译要求')
+    
+    # list 命令
+    list_parser = subparsers.add_parser('list', help='列出翻译记录')
+    
+    # status 命令
+    status_parser = subparsers.add_parser('status', help='查看翻译状态')
+    status_parser.add_argument('id', type=int, help='翻译ID')
+    
+    # download 命令
+    download_parser = subparsers.add_parser('download', help='下载翻译结果')
+    download_parser.add_argument('id', type=int, help='翻译ID')
+    download_parser.add_argument('output', help='输出文件路径')
+    
+    # config 命令
+    config_parser = subparsers.add_parser('config', help='显示当前配置')
+    
+    args = parser.parse_args()
+    
+    # 更新 API 地址（如果用户指定了）
+    api_base_arg = args.api
+    
+    if not args.command:
+        parser.print_help()
+        return
+    
+    # 登录（如果提供了用户名密码）
+    cookies = None
+    if args.user and args.password:
+        cookies = login(args.user, args.password, api_base_arg)
+        if not cookies:
+            print("请检查用户名和密码")
+            return
+        
+        user = get_user_info(cookies, api_base_arg)
+        if user:
+            print(f"👤 用户: {user.get('username')} ({user.get('user_type')})")
+    
+    # 执行命令
+    if args.command == 'translate':
+        translation_id = translate_pdf(args.file, args.instruction, cookies, api_base_arg)
+        if translation_id:
+            print(f"\n💡 在网页查看结果: {api_base_arg}/history")
+    
+    elif args.command == 'list':
+        list_translations(cookies, api_base_arg)
+    
+    elif args.command == 'status':
+        get_translation_status(args.id, cookies, api_base_arg)
+    
+    elif args.command == 'download':
+        download_translation(args.id, args.output, cookies, api_base_arg)
+    
+    elif args.command == 'config':
+        show_config(api_base_arg)
+
+if __name__ == '__main__':
+    main()

services.py

@@ -7,10 +7,20 @@ import json
 import time
 import hashlib
 import threading
+import base64
+import io
 from datetime import datetime, timedelta
 from pypdf import PdfReader
 from openai import OpenAI
 from flask import current_app
+from PIL import Image
+
+# pdf2image 用于将PDF转为图像
+try:
+    from pdf2image import convert_from_path
+    PDF_TO_IMAGE_AVAILABLE = True
+except ImportError:
+    PDF_TO_IMAGE_AVAILABLE = False
 
 # ==================== LLM客户端 ====================
 class TranslationService:
@@ -99,6 +109,115 @@ class TranslationService:
         text = re.sub(r'[\x00-\x08\x0b\x0c\x0e-\x1f]', '', text)
         return text.strip()
     
+    def is_vision_model(self):
+        """检查是否是视觉模型"""
+        model = self.llm_config.get('model', '')
+        # 常见视觉模型名称
+        vision_models = ['vision', 'vlm', 'glm-4v', 'glm-4.6v', 'gpt-4-vision', 'gpt-4o', 'qwen-vl', 'claude-3']
+        return any(v in model.lower() for v in vision_models)
+    
+    def pdf_to_images(self, pdf_path, max_pages=None):
+        """将PDF页面转换为图像"""
+        if not PDF_TO_IMAGE_AVAILABLE:
+            return None, "pdf2image未安装，无法处理扫描版PDF。请安装: pip install pdf2image"
+        
+        try:
+            # 获取PDF页数
+            reader = PdfReader(pdf_path)
+            total_pages = len(reader.pages)
+            
+            if max_pages:
+                pages_to_convert = min(max_pages, total_pages)
+            else:
+                pages_to_convert = total_pages
+            
+            # 转换PDF为图像
+            images = convert_from_path(
+                pdf_path,
+                first_page=1,
+                last_page=pages_to_convert,
+                dpi=200,  # 适当的DPI
+                fmt='jpeg'
+            )
+            
+            return images, None
+            
+        except Exception as e:
+            return None, f"PDF转图像失败: {str(e)}"
+    
+    def extract_text_from_image(self, image):
+        """使用视觉模型从图像中提取文字"""
+        if not self.is_vision_model():
+            return None, "当前模型不是视觉模型，无法识别图像文字"
+        
+        try:
+            # 将图像转为base64
+            buffered = io.BytesIO()
+            image.save(buffered, format="JPEG")
+            img_base64 = base64.b64encode(buffered.getvalue()).decode('utf-8')
+            
+            # 构建多模态请求
+            response = self.client.chat.completions.create(
+                model=self.llm_config['model'],
+                messages=[
+                    {
+                        "role": "user",
+                        "content": [
+                            {
+                                "type": "text",
+                                "text": "请识别并提取这张图片中的所有文字内容。只输出提取的文字，不要添加任何解释或说明。保持原有的段落和格式。"
+                            },
+                            {
+                                "type": "image_url",
+                                "image_url": {
+                                    "url": f"data:image/jpeg;base64,{img_base64}"
+                                }
+                            }
+                        ]
+                    }
+                ],
+                max_tokens=self.llm_config['max_tokens'],
+                temperature=0.1,
+                timeout=self.llm_config['timeout'],
+            )
+            
+            content = response.choices[0].message.content
+            return content.strip() if content else '', None
+            
+        except Exception as e:
+            return '', f"视觉模型识别失败: {str(e)}"
+    
+    def extract_text_from_scanned_pdf(self, pdf_path, progress_callback=None):
+        """从扫描版PDF提取文字（使用视觉模型OCR）"""
+        images, error = self.pdf_to_images(pdf_path)
+        
+        if error:
+            return [], error
+        
+        pages_text = []
+        total = len(images)
+        
+        for i, image in enumerate(images):
+            if progress_callback:
+                progress_callback(int((i+1)/total*50), total, f"OCR识别第{i+1}页...")
+            
+            text, err = self.extract_text_from_image(image)
+            
+            if err:
+                pages_text.append({
+                    'page': i + 1,
+                    'text': '',
+                    'error': err
+                })
+            else:
+                pages_text.append({
+                    'page': i + 1,
+                    'text': text or '',
+                    'error': None
+                })
+        
+        return pages_text, None
+    
     def chunk_text(self, text, max_size=2000):
         """分块"""
         paragraphs = text.split('\n\n')
@@ -131,11 +250,38 @@ class TranslationService:
         Returns:
             翻译统计信息
         """
+        # 先尝试常规提取
         pages = self.extract_pdf_text(pdf_path)
         total_pages = len(pages)
+        total_text = sum(len(p['text']) for p in pages)
+        
+        # 如果无法提取文本，尝试使用视觉模型OCR
+        if total_pages == 0 or total_text < 10:
+            if self.is_vision_model() and PDF_TO_IMAGE_AVAILABLE:
+                if progress_callback:
+                    progress_callback(0, 0, "检测到扫描版PDF，使用视觉模型OCR...")
+                
+                pages, error = self.extract_text_from_scanned_pdf(pdf_path, progress_callback)
+                
+                if error:
+                    raise ValueError(error)
+                
+                total_pages = len(pages)
+                total_text = sum(len(p['text']) for p in pages)
+                
+                if total_text < 10:
+                    raise ValueError("视觉模型OCR未能提取到有效文字内容")
+                
+                if progress_callback:
+                    progress_callback(50, total_pages, "OCR完成，开始翻译...")
+            else:
+                error_msg = "PDF无法提取文本内容。可能原因：\n1. PDF是扫描版（图像形式）\n2. 当前大模型不是视觉模型，无法识别图像文字\n\n如需处理扫描版PDF，请配置视觉大模型（如 glm-4.6v、gpt-4-vision）"
+                if progress_callback:
+                    progress_callback(0, 0, error_msg)
+                raise ValueError(error_msg)
         
         if progress_callback:
-            progress_callback(0, total_pages, "开始翻译...")
+            progress_callback(50, total_pages, "开始翻译...")
         
         translated_pages = []
         total_chunks = 0
@@ -150,8 +296,10 @@ class TranslationService:
                 translated_chunks.append(translated)
                 
                 if progress_callback:
-                    progress = int((i + 1) / len(chunks) * 100 / total_pages)
+                    # OCR占50%，翻译占50%
-                    progress_callback(progress, total_pages, f"翻译第{page_data['page']}页")
+                    page_progress = (i + 1) / len(chunks)
+                    overall_progress = 50 + int(page_progress * 50 / total_pages)
+                    progress_callback(overall_progress, total_pages, f"翻译第{page_data['page']}页")
             
             translated_pages.append({
                 'page': page_data['page'],
@@ -279,12 +427,14 @@ class TranslationTask:
                 with app.app_context():
                     from admin import get_llm_config
                     llm_config = get_llm_config()
-                    config['LLM_CONFIG'] = llm_config
+                    config = {'LLM_CONFIG': llm_config}
             
             service = TranslationService(config)
             task['status'] = 'processing'
             task['started_at'] = datetime.now().isoformat()
             
+            print(f"[翻译任务] 开始翻译，使用配置: {config.get('LLM_CONFIG', {}).get('api_base', '未知')}")
+            
             # 更新数据库状态为 processing
             if app and translation_id:
                 with app.app_context():