"""
Aumovio/trinamiX 触指红外 BAC 检测原理复现

核心技术：近红外光谱(NIR)分析间质液酒精浓度
- 酒精分子在近红外波段有特征吸收峰
- 通过多波长照射手指，分析透射/反射光谱
- 机器学习模型反推 BAC 值

参考：trinamiX NIR spectroscopy + BASF 化学分析能力
"""

import numpy as np
from sklearn.ensemble import GradientBoostingRegressor
from sklearn.model_selection import cross_val_score
from typing import Tuple

# 酒精分子在近红外波段的主要吸收特征
# 实际数据来自 NIR 光谱仪，此处用模拟数据演示原理
ALCOHOL_NIR_PEAKS = {
    # O-H 伸缩振动的一阶倍频
    "OH_stretch_1st_overtone": 1450,  # nm
    # O-H 伸缩振动的二阶倍频  
    "OH_stretch_2nd_overtone": 1730,  # nm
    # C-H 伸缩振动组合频
    "CH_stretch_combination": 2300,  # nm
    # O-H 弯曲振动组合频
    "OH_bend_combination": 2350,  # nm
}

# 典型 NIR 光谱仪波长配置（trinamiX 类似配置）
NIR_WAVELENGTHS = np.array([
    1300, 1450, 1550, 1650, 1730, 1950, 
    2100, 2300, 2350, 2500
])  # nm, 10个关键波长


def simulate_finger_nir_signal(
    bac_level: float, 
    skin_thickness: float = 2.0,
    temperature: float = 32.0,
    moisture: float = 0.35
) -> np.ndarray:
    """
    模拟手指 NIR 光谱信号
    
    Args:
        bac_level: 血液酒精浓度 (g/dL), 0.0-0.3
        skin_thickness: 皮肤厚度 (mm), 1.5-3.0
        temperature: 手指表面温度 (°C), 25-37
        moisture: 皮肤水分含量 (0-1)
    
    Returns:
        absorbance: 各波长吸光度, shape=(10,)
    
    物理模型：
    - Beer-Lambert 定律: A = ε*c*l
    - 酒精在间质液浓度 ≈ BAC * 0.85（渗透系数）
    - 皮肤基底吸光度 + 酒精特征吸收 + 噪声
    """
    # 间质液酒精浓度
    interstitial_alcohol = bac_level * 0.85
    
    # 各波长的摩尔吸光系数（简化模型）
    # 在酒精特征峰处系数最大
    molar_absorptivity = np.zeros(len(NIR_WAVELENGTHS))
    for i, wl in enumerate(NIR_WAVELENGTHS):
        if wl == 1450:
            molar_absorptivity[i] = 0.85  # O-H 一阶倍频
        elif wl == 1730:
            molar_absorptivity[i] = 0.65  # O-H 二阶倍频
        elif wl == 2300:
            molar_absorptivity[i] = 0.45  # C-H 组合频
        elif wl == 2350:
            molar_absorceptivity[i] = 0.40  # O-H 弯曲组合
        else:
            molar_absorptivity[i] = 0.05 + np.random.normal(0, 0.02)
    
    # 皮肤基底吸光度（与水分、厚度相关）
    baseline = (0.3 + 0.1 * moisture + 0.05 * skin_thickness) * np.ones(len(NIR_WAVELENGTHS))
    
    # 温度修正项
    temp_correction = 0.001 * (temperature - 32.0) * np.ones(len(NIR_WAVELENGTHS))
    
    # 酒精特征吸收
    alcohol_absorption = molar_absorptivity * interstitial_alcohol * skin_thickness
    
    # 总吸光度 = 基线 + 酒精吸收 + 温度修正 + 噪声
    absorbance = baseline + alcohol_absorption + temp_correction
    absorbance += np.random.normal(0, 0.01, len(NIR_WAVELENGTHS))
    
    return absorbance


def build_bac_prediction_model():
    """
    构建 BAC 预测模型
    
    使用梯度提升回归器，输入10个波长吸光度
    输出 BAC 预测值
    
    Returns:
        model: 训练好的模型
        X_test, y_test: 测试集
    """
    np.random.seed(42)
    n_samples = 1000
    
    # 生成多样本：覆盖不同BAC、皮肤条件、温度
    X = []
    y = []
    for _ in range(n_samples):
        bac = np.random.uniform(0.0, 0.25)
        skin = np.random.uniform(1.5, 3.0)
        temp = np.random.uniform(25, 37)
        moisture = np.random.uniform(0.2, 0.5)
        
        signal = simulate_finger_nir_signal(bac, skin, temp, moisture)
        X.append(signal)
        y.append(bac)
    
    X = np.array(X)
    y = np.array(y)
    
    # 训练模型
    from sklearn.model_selection import train_test_split
    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
    
    model = GradientBoostingRegressor(
        n_estimators=200,
        max_depth=5,
        learning_rate=0.1,
        subsample=0.8,
        random_state=42
    )
    model.fit(X_train, y_train)
    
    # 交叉验证
    cv_scores = cross_val_score(model, X_train, y_train, cv=5, scoring='r2')
    
    # 测试集评估
    y_pred = model.predict(X_test)
    mae = np.mean(np.abs(y_pred - y_test))
    
    print(f"=== BAC 预测模型性能 ===")
    print(f"交叉验证 R²: {cv_scores.mean():.4f} ± {cv_scores.std():.4f}")
    print(f"测试集 MAE: {mae:.4f} g/dL")
    print(f"测试集 MAE: {mae*100:.1f} mg/dL")
    
    # 法规阈值检测性能
    threshold = 0.08  # g/dL, 法定醉驾阈值
    y_binary = (y_test >= threshold).astype(int)
    pred_binary = (y_pred >= threshold).astype(int)
    
    from sklearn.metrics import classification_report
    print(f"\n=== 阈值 {threshold} g/dL 检测性能 ===")
    print(classification_report(y_binary, pred_binary, target_names=['未醉驾', '醉驾']))
    
    return model, X_test, y_test


if __name__ == "__main__":
    model, X_test, y_test = build_bac_prediction_model()
    
    # 模拟单次检测
    print("\n=== 模拟单次检测 ===")
    test_bac = 0.09  # 轻微醉驾
    signal = simulate_finger_nir_signal(test_bac, skin_thickness=2.0, temperature=32, moisture=0.35)
    predicted_bac = model.predict(signal.reshape(1, -1))[0]
    
    print(f"真实 BAC: {test_bac:.3f} g/dL")
    print(f"预测 BAC: {predicted_bac:.3f} g/dL")
    print(f"误差: {abs(predicted_bac - test_bac):.4f} g/dL")
    print(f"判定: {'醉驾' if predicted_bac >= 0.08 else '正常'}")

# BAC 检测方法精度对比

import numpy as np

methods = {
    "抽血测试(金标准)": {"mae": 0.000, "bias": 0.000, "95%_CI": 0.005},
    "Aumovio NIR触指": {"mae": 0.008, "bias": 0.002, "95%_CI": 0.020},  # 声称
    "DADSS被动呼气": {"mae": 0.025, "bias": 0.005, "95%_CI": 0.035},  # 目标
    "手持呼吸仪": {"mae": 0.015, "bias": -0.003, "95%_CI": 0.025},
    "方向盘汗液": {"mae": 0.040, "bias": 0.010, "95%_CI": 0.060},
}

print(f"{'方法':<25} {'MAE(g/dL)':<12} {'偏差(g/dL)':<12} {'95% CI(g/dL)':<14} {'达标?'}")
print("-" * 75)
for name, m in methods.items():
    meets = "✓" if m["95%_CI"] <= 0.025 else "✗"
    print(f"{name:<25} {m['mae']:<12.3f} {m['bias']:<12.3f} {m['95%_CI']:<14.3f} {meets}")

"""
反欺骗检测逻辑示例

三层检测：活体 → 身份 → 行为
"""

class AntiSpoofingSystem:
    def __init__(self):
        self.liveness_threshold = 0.85
        self.identity_threshold = 0.80
        self.behavior_threshold = 0.70
    
    def check_liveness(self, dot_pattern: np.ndarray) -> Tuple[bool, float]:
        """
        L1: 基于Dot Projector的活体检测
        
        Args:
            dot_pattern: 投射点云数据, shape=(N, 3)
        
        Returns:
            is_alive, confidence
        """
        # 真人手指表面有微小起伏
        surface_roughness = np.std(dot_pattern[:, 2])
        # 假手指（硅胶/3D打印）表面过于平滑
        roughness_score = min(surface_roughness / 0.5, 1.0)
        
        # 温度特征（红外波段推断）
        # 真人手指温度 28-37°C
        confidence = roughness_score * 0.6 + 0.4  # 简化
        return confidence > self.liveness_threshold, confidence
    
    def check_identity(self, face_image: np.ndarray) -> Tuple[bool, float]:
        """
        L2: DMS摄像头身份验证
        确认触碰传感器的人=注册驾驶员
        """
        # 简化：实际使用人脸识别模型
        confidence = 0.88  # 模拟
        return confidence > self.identity_threshold, confidence
    
    def check_behavior(self, vehicle_data: dict) -> Tuple[bool, float]:
        """
        L3: 行为分析
        检测是否有人代触碰后换人驾驶
        """
        # 检测触碰后到启动的时间间隔
        time_since_touch = vehicle_data.get("time_since_touch", 0)
        door_opened = vehicle_data.get("door_opened", False)
        
        if time_since_touch > 30 or door_opened:
            return False, 0.3
        return True, 0.85
    
    def verify(self, dot_pattern, face_image, vehicle_data) -> dict:
        l1_ok, l1_conf = self.check_liveness(dot_pattern)
        l2_ok, l2_conf = self.check_identity(face_image)
        l3_ok, l3_conf = self.check_behavior(vehicle_data)
        
        all_pass = l1_ok and l2_ok and l3_ok
        return {
            "verified": all_pass,
            "liveness": {"pass": l1_ok, "confidence": l1_conf},
            "identity": {"pass": l2_ok, "confidence": l2_conf},
            "behavior": {"pass": l3_ok, "confidence": l3_conf}
        }

# 酒驾检测开发验证路线

development_roadmap = {
    "Phase 1 (0-3月)": {
        "任务": "NIR传感器评估与数据采集",
        "输入": "trinamiX开发套件 + 志愿者手指NIR数据",
        "输出": "NIR-BAC标定数据库(≥100人, ≥3环境条件)",
        "验收": "MAE ≤ 0.02 g/dL @ 0.08阈值"
    },
    "Phase 2 (3-6月)": {
        "任务": "ML模型训练与优化",
        "输入": "标定数据库",
        "输出": "量化INT8模型, ≤2MB, MCU可运行",
        "验收": "MCU推理 ≤500ms, MAE ≤ 0.015 g/dL"
    },
    "Phase 3 (6-9月)": {
        "任务": "反欺骗集成",
        "输入": "Dot projector + DMS摄像头",
        "输出": "三层防欺骗系统, 误拒率<1%, 误受率<0.1%",
        "验收": "1000次假手指测试, 0次通过"
    },
    "Phase 4 (9-12月)": {
        "任务": "车载环境验证",
        "输入": "实车安装, 4季测试",
        "输出": "车规级原型, AEC-Q100认证计划",
        "验收": "-40°C~+85°C, 湿度10-95%, 振动测试通过"
    }
}

for phase, details in development_roadmap.items():
    print(f"\n{phase}:")
    for k, v in details.items():
        print(f"  {k}: {v}")