150 lines
5.8 KiB
Python
150 lines
5.8 KiB
Python
|
import numpy as np
|
|||
|
|
from PIL import Image
|
|||
|
|
import os
|
|||
|
|
import asyncio
|
|||
|
|
import hashlib
|
|||
|
|
import json
|
|||
|
|
import aiohttp
|
|||
|
|
import logging
|
|||
|
|
import base64
|
|||
|
|
from openai import OpenAI, BadRequestError
|
|||
|
|
from Config.Config import (
|
|||
|
|
ALY_LLM_API_KEY, VL_MODEL_NAME, VL_MODEL_NAME_AD
|
|||
|
|
)
|
|||
|
|
from Apps.AiTeJiYiChong.Config.Setting import (
|
|||
|
|
SAFE_EXCLUDE_RATIO, FALLBACK_WIDTH, FALLBACK_HEIGHT,
|
|||
|
|
BOTTOM_SAFE_EXCLUDE_RATIO
|
|||
|
|
)
|
|||
|
|
|
|||
|
|
logger = logging.getLogger(__name__)
|
|||
|
|
|
|||
|
|
|
|||
|
|
class ReadImageKit:
|
|||
|
|
_client = OpenAI(
|
|||
|
|
api_key=ALY_LLM_API_KEY,
|
|||
|
|
base_url="https://dashscope.aliyuncs.com/compatible-mode/v1"
|
|||
|
|
)
|
|||
|
|
|
|||
|
|
# 通用回退设备信息，仅在无法动态获取设备信息时使用
|
|||
|
|
_FALLBACK_DEVICE_INFO = {
|
|||
|
|
"displayWidth": FALLBACK_WIDTH,
|
|||
|
|
"displayHeight": FALLBACK_HEIGHT,
|
|||
|
|
"productName": "generic"
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
_prompt = (
|
|||
|
|
"仅输出JSON数组（不含任何说明文字），按从左到右、从上到下的顺序识别图片中由【绿色方框】标识的充电站区域。识别规则如下：\n"
|
|||
|
|
"1. 必须是图中用绿色实线方框圈出的区域。\n"
|
|||
|
|
"2. 每一个卡片区域必须同时具备以下所有要素，否则严禁识别：\n"
|
|||
|
|
" - 场站名称 (station_name)；\n"
|
|||
|
|
" - 距离信息 (distance, 例如 '2.82km' 或 '90m')，通常位于卡片右侧蓝色胶囊区域内；\n"
|
|||
|
|
" - 金额/电费 (price，例如 '1.2500')，通常以红色字体显示；\n"
|
|||
|
|
" - 充电枪信息 (piles，包含'快'或'慢'的类型、总枪数和空闲枪数，例如 '快 4/4')。\n"
|
|||
|
|
"3. 如果绿色方框内缺少上述任何一项要素，说明它不是真正的场站卡片，请直接跳过。\n"
|
|||
|
|
"\n"
|
|||
|
|
"JSON对象字段要求：\n"
|
|||
|
|
"1. b_use: 状态标识（1或0）。如果场站名称为灰色或带有“暂停使用”等标签，则为0，否则为1。\n"
|
|||
|
|
"2. station_name: 场站名称；\n"
|
|||
|
|
"3. price: 一度电的价格（数字，如 1.2500）；\n"
|
|||
|
|
"4. piles: 充电枪列表 [{type: '快', free: 4, total: 4}]；\n"
|
|||
|
|
"5. parking: 停车费用描述（通常在蓝色'P'图标后，例如 '免费停车三小时'）；\n"
|
|||
|
|
"6. distance: 距离信息字符串（例如 '2.82km' 或 '90m'）；\n"
|
|||
|
|
"7. bounds: {x1,y1,x2,y2} 区域像素坐标（0-1000）；\n"
|
|||
|
|
"8. bounds_norm: {left,top,right,bottom} 归一化坐标(0-1)；\n"
|
|||
|
|
"9. station_name_bounds: 场站名称文字区域坐标 {x1,y1,x2,y2}（0-1000）；\n"
|
|||
|
|
"10. station_name_bounds_norm: 场站名称文字归一化坐标(0-1)。\n"
|
|||
|
|
"\n"
|
|||
|
|
"重要约束：\n"
|
|||
|
|
"A. 严禁识别未被绿色方框圈出的区域。如顶部的“长春市”选择、顶部的搜索框、以及中间的“推荐站点”等标签。\n"
|
|||
|
|
"B. 真正的场站卡片在绿色方框内包含：场站名称、金额（红色文字）、距离（右侧蓝色背景内）、充电枪状态（绿色或蓝色徽章，格式为 闲x/x 或 x/x）。\n"
|
|||
|
|
"C. 严禁将顶部的功能图标（如我的订单、收藏站点等）误认为场站卡片。\n"
|
|||
|
|
"\n"
|
|||
|
|
"严格返回纯JSON格式。"
|
|||
|
|
)
|
|||
|
|
|
|||
|
|
@staticmethod
|
|||
|
|
def _extract_json(text: str) -> str:
|
|||
|
|
if not text:
|
|||
|
|
return "[]"
|
|||
|
|
|
|||
|
|
cleaned = text.strip()
|
|||
|
|
if "```" in cleaned:
|
|||
|
|
lines = []
|
|||
|
|
for line in cleaned.splitlines():
|
|||
|
|
if line.strip().startswith("```"):
|
|||
|
|
continue
|
|||
|
|
lines.append(line)
|
|||
|
|
cleaned = "\n".join(lines).strip()
|
|||
|
|
|
|||
|
|
decoder = json.JSONDecoder()
|
|||
|
|
|
|||
|
|
pos = 0
|
|||
|
|
while pos < len(cleaned):
|
|||
|
|
idx_dict = cleaned.find("{", pos)
|
|||
|
|
idx_list = cleaned.find("[", pos)
|
|||
|
|
|
|||
|
|
candidates = [i for i in (idx_dict, idx_list) if i != -1]
|
|||
|
|
if not candidates:
|
|||
|
|
break
|
|||
|
|
|
|||
|
|
start = min(candidates)
|
|||
|
|
snippet = cleaned[start:]
|
|||
|
|
try:
|
|||
|
|
_, end = decoder.raw_decode(snippet)
|
|||
|
|
return snippet[:end]
|
|||
|
|
except json.JSONDecodeError:
|
|||
|
|
pos = start + 1
|
|||
|
|
continue
|
|||
|
|
|
|||
|
|
return "[]"
|
|||
|
|
|
|||
|
|
@classmethod
|
|||
|
|
async def get_stations_from_image(cls, image_path: str, device_info=None):
|
|||
|
|
"""
|
|||
|
|
使用 Qwen-VL 模型从截图中识别充电站列表
|
|||
|
|
"""
|
|||
|
|
if device_info is None:
|
|||
|
|
device_info = cls._FALLBACK_DEVICE_INFO
|
|||
|
|
|
|||
|
|
if not os.path.exists(image_path):
|
|||
|
|
logger.error(f"Image not found: {image_path}")
|
|||
|
|
return []
|
|||
|
|
|
|||
|
|
# 将图片转换为 Base64
|
|||
|
|
with open(image_path, "rb") as image_file:
|
|||
|
|
encoded_image = base64.b64encode(image_file.read()).decode("utf-8")
|
|||
|
|
|
|||
|
|
try:
|
|||
|
|
response = await asyncio.to_thread(
|
|||
|
|
cls._client.chat.completions.create,
|
|||
|
|
model=VL_MODEL_NAME,
|
|||
|
|
messages=[
|
|||
|
|
{
|
|||
|
|
"role": "user",
|
|||
|
|
"content": [
|
|||
|
|
{"type": "text", "text": cls._prompt},
|
|||
|
|
{
|
|||
|
|
"type": "image_url",
|
|||
|
|
"image_url": {"url": f"data:image/jpeg;base64,{encoded_image}"},
|
|||
|
|
},
|
|||
|
|
],
|
|||
|
|
}
|
|||
|
|
],
|
|||
|
|
max_tokens=2000,
|
|||
|
|
temperature=0.01
|
|||
|
|
)
|
|||
|
|
|
|||
|
|
content = response.choices[0].message.content
|
|||
|
|
logger.info(f"VL Model Response: {content}")
|
|||
|
|
|
|||
|
|
json_str = cls._extract_json(content)
|
|||
|
|
stations = json.loads(json_str)
|
|||
|
|
|
|||
|
|
# 后处理：如果 bounds 是归一化的，则转换为像素坐标（如果需要）
|
|||
|
|
# 或者如果 bounds 是 0-1000 的，则保持原样或按需转换
|
|||
|
|
return stations
|
|||
|
|
|
|||
|
|
except Exception as e:
|
|||
|
|
logger.error(f"Error calling VL model: {e}")
|
|||
|
|
return []
|