198 lines
7.6 KiB
Python
198 lines
7.6 KiB
Python
from fastapi import FastAPI, File, UploadFile, Form
|
||
from typing import Literal
|
||
from fastapi.responses import JSONResponse
|
||
from fastapi.exceptions import HTTPException
|
||
from PIL import Image
|
||
from paddleocr import PaddleOCR
|
||
import numpy as np
|
||
import uvicorn
|
||
import re
|
||
import io
|
||
from pdf2image import convert_from_bytes
|
||
import uuid
|
||
import os
|
||
import requests
|
||
import json
|
||
import paddle
|
||
import sys
|
||
from uvicorn.config import LOGGING_CONFIG
|
||
|
||
CUR_DIR = os.path.dirname(os.path.abspath(__file__))
|
||
sys.path.insert(0, CUR_DIR)
|
||
import conf
|
||
|
||
LOGGING_CONFIG["formatters"]["default"]["fmt"] = "%(asctime)s - %(levelname)s - %(message)s"
|
||
LOGGING_CONFIG["handlers"]["file"] = {
|
||
"class": "logging.handlers.RotatingFileHandler",
|
||
"filename": os.path.join(CUR_DIR, "app.log"),
|
||
"maxBytes": 5 * 1024 * 1024, # 5 MB
|
||
"backupCount": 5,
|
||
"formatter": "default",
|
||
"encoding": "utf-8"
|
||
}
|
||
LOGGING_CONFIG["loggers"]["uvicorn.error"]["handlers"]=["file"]
|
||
|
||
|
||
|
||
gpu_available = paddle.device.is_compiled_with_cuda()
|
||
print("GPU available:", gpu_available)
|
||
|
||
|
||
ocr = PaddleOCR(use_angle_cls=True, lang="ch", use_gpu=conf.GPU_ENABLE)
|
||
app = FastAPI()
|
||
|
||
ALLOWED_FILE_TYPES = {"application/pdf", "image/jpeg", "image/png", "image/jpg"}
|
||
|
||
def save_imgs(images):
|
||
"""
|
||
保存图像到本地
|
||
"""
|
||
if not os.path.exists('save_imgs'):
|
||
os.makedirs('save_imgs')
|
||
paths = []
|
||
for i, image in enumerate(images):
|
||
image_path = os.path.join('saved_imgs', f'{uuid.uuid4()}.png')
|
||
image.save(image_path)
|
||
paths.append(image_path)
|
||
return paths
|
||
|
||
def check_file_format(file: UploadFile):
|
||
if file.content_type not in ALLOWED_FILE_TYPES:
|
||
raise HTTPException(400, "File format not supported.")
|
||
|
||
def perform_ocr(ocr, images):
|
||
"""
|
||
使用 OCR 提取图像文件中的文字.
|
||
"""
|
||
all_text = []
|
||
for image in images:
|
||
# Convert PIL Image to numpy array
|
||
image_np = np.array(image)
|
||
result = ocr.ocr(image_np, cls=True)
|
||
all_text.extend(result)
|
||
return all_text
|
||
|
||
def extract_standard_info(text, patterns):
|
||
"""
|
||
从标准文件文本中提取信息.
|
||
"""
|
||
info = {}
|
||
for label, pattern in patterns:
|
||
if label in ("归口单位", "中文名称"):
|
||
matches = re.findall(pattern, text, re.MULTILINE)
|
||
else:
|
||
matches = re.findall(pattern, text, re.DOTALL)
|
||
if matches:
|
||
if label in ["发布日期", "实施日期"]:
|
||
info[label] = re.sub(r'－', '-', matches[0].replace(' ', ''))
|
||
elif label in ["起草单位", "起草人"]:
|
||
info[label] = matches[0].strip()
|
||
elif label == "提出单位":
|
||
info[label] = matches[-1].strip()
|
||
elif label == "标准号":
|
||
info[label] = matches[0].strip()
|
||
elif label == "发布部门":
|
||
if matches[0][1]:
|
||
info[label] = matches[0][0] + '、' + matches[0][1]
|
||
else:
|
||
info[label] = matches[0][0] + matches[0][1]
|
||
info[label] = re.sub(r'\n', '', info[label])
|
||
else:
|
||
info[label] = matches[-1] if matches else None
|
||
|
||
return info
|
||
|
||
|
||
def extract_patent_info(text, patterns):
|
||
"""
|
||
从专利文件文本中提取信息.
|
||
"""
|
||
info = {}
|
||
|
||
for label, pattern in patterns:
|
||
matches = re.findall(pattern, text, re.DOTALL)
|
||
info[label] = matches[-1] if matches else None
|
||
|
||
return info
|
||
|
||
def clean_info(info: dict):
|
||
for key in info:
|
||
if info[key]:
|
||
info[key] = info[key].replace('\n', '').replace('\r', '').replace('\t', '').strip()
|
||
|
||
# 定义标准文件用的正则表达式模式
|
||
standard_patterns = [
|
||
("标准号", r"[A-Z][A-Z/\ 0-9]{1,}[0-9\.]+[\-—－]\d{3,4}"), # 仅提取了第一个标准号(若有更多标准号可能错误)
|
||
("中文名称", r"(^[\u4e00-\u9fa5][\w:：、√—\-\ \nn]+)\n" # 定位英文名称前的中文名称
|
||
r"(?:[^\u4e00-\u9fa5]+)\n" # 定位发布日期前的英文名称部分
|
||
r"(?:\d{4}(?:-|－)\d{2}(?:-|－)\d{2}) *发布"), # 定位发布日期
|
||
("英文名称", r"\n([^\u4e00-\u9fa5]+)\n(?:\d{4}(?:-|－)\d{2}(?:-|－)\d{2}) *发布"), # 匹配发布日期前面的非中文部分
|
||
("发布部门", r"([\u4e00-\u9fa5]+)\n发布\n([\u4e00-\u9fa5\n]*)"), # 发布部门是通过匹配发布的上下行的中文
|
||
("发布日期", r"(\d{4}(?:-|－)\d{2}(?:-|－)\d{2}) *发布"),
|
||
("实施日期", r"(\d{4}(?:-|－)\d{2}(?:-|－)\d{2}) *实施"),
|
||
("提出单位", r"由(.+?)\s?提出。"),
|
||
("归口单位", r"由(.+?)\s?归口。"),
|
||
("起草单位", r"起草单位[：:](.+?)(?:。|起草人)"),
|
||
("起草人", r"起草人[：:](.+?)(?:。)")
|
||
]
|
||
|
||
# 定义专利文件用的正则表达式模式
|
||
patent_patterns = [
|
||
("专利名称", r"名称[：:](.+?)\n"),
|
||
("发明人", r"发[\s]*明[\s]*人[\s]*[：:](.+?)\n"),
|
||
("专利号", r"专[\s]*利[\s]*号[\s]*[：:](.+?)\n"),
|
||
("专利申请日", r"专利申请日[：:](.+?)\n"),
|
||
("专利权人", r"专[\s]*利[\s]*权[\s]*人[\s]*[：:](.+?)\n"),
|
||
("地址", r"地[\s]*址[\s]*[：:](.+?)\n"),
|
||
("授权公告日", r"授权公告日：(.+?)\n"),
|
||
("授权公告号", r"授权公告号：(.+?)\n")
|
||
]
|
||
|
||
@app.get("/error")
|
||
async def create_error():
|
||
raise ValueError("This is a test error!")
|
||
|
||
|
||
@app.post("/extract")
|
||
async def extract_info(
|
||
file_type: Literal["standard", "patent"] = Form(..., description="Specify the type of file to extract."),
|
||
extract_method: Literal["re", "chat"] = Form(..., description="Specify how to extract. 正则或对话模型"),
|
||
file: UploadFile = File(..., description="Upload a PDF, JPEG, or PNG file.")
|
||
):
|
||
|
||
check_file_format(file)
|
||
content = await file.read() # 读取文件内容
|
||
if file.filename.lower().endswith('.pdf'):
|
||
images = convert_from_bytes(content, first_page=1, last_page=6, dpi=300)
|
||
else:
|
||
images = [Image.open(io.BytesIO(content))]
|
||
result = perform_ocr(ocr, images)
|
||
ocr_text = "\n".join([line[1][0] for res in result if res is not None for line in res])
|
||
|
||
# 提取信息
|
||
if extract_method == "re":
|
||
if file_type == "patent":
|
||
info = extract_patent_info(ocr_text, patent_patterns)
|
||
clean_info(info)
|
||
elif file_type == "standard":
|
||
info = extract_standard_info(ocr_text, standard_patterns)
|
||
clean_info(info)
|
||
else:
|
||
raise HTTPException(400, detail="Invalid file type. Please choose 'standard' or 'patent'.")
|
||
else:
|
||
if file_type == 'patent':
|
||
prompt = f'我有以下文本,是一个专利的内容。请按专利名称，发明人，专利号，专利申请日， 专利权人，授权公告号，授权公告日为key的json格式返回数据,注意只返回json数据。文本如下：{ocr_text}'
|
||
elif file_type == "standard":
|
||
prompt = f'我有以下文本,是一个标准的内容。请按标准号，中文名称，英文名称，发布部门，发布日期，实施日期，提出单位，归口单位，起草单位，起草人为key的json格式返回数据,注意只返回json数据。文本如下：{ocr_text}'
|
||
else:
|
||
raise HTTPException(400, detail="Invalid file type. Please choose 'standard' or 'patent'.")
|
||
r = requests.post(conf.CHAT_API, json={
|
||
"model": conf.CHAT_MODEL,
|
||
"prompt": prompt,
|
||
"stream": False
|
||
})
|
||
info = json.loads((r.json()['response']))
|
||
return JSONResponse(content=info)
|
||
|
||
if __name__ == "__main__":
|
||
uvicorn.run(app="main:app", host="0.0.0.0", port=2260, reload=True, log_config=LOGGING_CONFIG, debug=False) |