class AviationFatigueMonitoring:
    """
    航空驾驶舱疲劳监测技术分类
    
    基于 Frontiers 2026 综述论文，对比分析
    汽车和航空领域的技术差异和共通点。
    """
    
    # 监测技术分类
    TECHNOLOGIES = {
        "EEG": {
            "description": "脑电图",
            "fatigue_indicators": [
                "frontal theta power increase",    # 前额叶θ波增加
                "parietal alpha power increase",   # 顶叶α波增加
                "theta/alpha ratio",              # θ/α 比值
                "beta power decrease"             # β波降低
            ],
            "automotive_use": "研究阶段，未量产",
            "aviation_use": "实验室验证，飞行模拟器",
            "challenges": ["佩戴舒适度", "信号干扰", "多电极"],
            "accuracy": "高（金标准）",
            "latency": "30-60s 累积"
        },
        "eye_tracking": {
            "description": "眼动追踪",
            "fatigue_indicators": [
                "PERCLOS",                    # 眼睑闭合百分比
                "blink duration increase",    # 眨眼时长增加
                "saccade velocity decrease",  # 扫视速度降低
                "gaze entropy decrease",     # 注视熵降低（视觉固定）
                "pupil diameter change"      # 瞳孔直径变化
            ],
            "automotive_use": "量产标配（Smart Eye, Seeing Machines）",
            "aviation_use": "军用验证，民用探索",
            "challenges": ["玻璃舱反光", "太阳镜", "强光变化"],
            "accuracy": "中-高",
            "latency": "1-10s"
        },
        "HRV": {
            "description": "心率变异性",
            "fatigue_indicators": [
                "LF/HF ratio decrease",       # 低频/高频比降低
                "RMSSD decrease",             # 相邻RR间期差值均方根降低
                "SDNN decrease",             # RR间期标准差降低
                "mean HR decrease"           # 平均心率降低
            ],
            "automotive_use": "方向盘电极、座椅传感器",
            "aviation_use": "飞行模拟器验证，真实飞行测试",
            "challenges": ["运动伪影", "电磁干扰", "佩戴"],
            "accuracy": "中",
            "latency": "60-300s"
        },
        "EDA": {
            "description": "皮肤电活动",
            "fatigue_indicators": [
                "skin conductance level decrease",  # 皮肤电导水平降低
                "non-specific SCR frequency"        # 非特异性皮肤电导反应频率
            ],
            "automotive_use": "方向盘接触",
            "aviation_use": "实验阶段",
            "challenges": ["个体差异大", "温度敏感"],
            "accuracy": "低-中",
            "latency": "60-120s"
        }
    }

class DMS2AMSTransfer:
    """
    汽车DMS技术向航空AMS迁移分析
    
    关键发现（来自论文）：
    1. 眼动追踪是迁移可行性最高的技术
    2. PERCLOS 在航空场景需要适配（飞行员不需要持续看路）
    3. 多模态融合在两个领域都优于单模态
    4. 航空场景的光照挑战比汽车更极端
    """
    
    # 迁移可行性评估
    TRANSFER_FEASIBILITY = {
        "eye_tracking_percclos": {
            "automotive_maturity": "量产",
            "aviation_feasibility": "中等",
            "adaptation_needed": "PERCLOS 基线需要重新标定（飞行员注视仪表而非道路）",
            "key_challenge": "玻璃驾驶舱反光、强光直射",
        },
        "blink_duration": {
            "automotive_maturity": "量产",
            "aviation_feasibility": "高",
            "adaptation_needed": "阈值调整（飞行员眨眼模式不同）",
            "key_challenge": "太阳镜遮挡",
        },
        "gaze_entropy": {
            "automotive_maturity": "研究",
            "aviation_feasibility": "高",
            "adaptation_needed": "仪表扫描模式基线建立",
            "key_challenge": "多显示器注视点分散",
        },
        "HRV": {
            "automotive_maturity": "有限量产",
            "aviation_feasibility": "中等",
            "adaptation_needed": "飞行服集成电极",
            "key_challenge": "振动干扰",
        },
        "EEG": {
            "automotive_maturity": "研究",
            "aviation_feasibility": "低",
            "adaptation_needed": "飞行员头盔集成",
            "key_challenge": "佩戴舒适度、信号稳定性",
        },
    }
    
    # Seeing Machines 跨域案例
    SEEING_MACHINES_CASE = {
        "automotive": "DriverGuardian Gen3 量产DMS",
        "aviation": "正在将DMS技术迁移到航空",
        "timeline": "2025-2027 预计航空验证",
        "link": "https://aerospaceamerica.aiaa.org/departments/tracking-pilots-eyes/",
    }

import torch
import torch.nn as nn
import numpy as np

class AviationFatigueDetector(nn.Module):
    """
    航空驾驶舱疲劳检测器
    
    架构：多模态融合（眼动 + HRV + 行为）
    基于论文综述的发现设计
    
    硬件配置：
    | 组件 | 型号 | 参数 | 安装位置 |
    |------|------|------|----------|
    | 眼动摄像头 | Smart Eye AX5 | 60fps, 红外 | 仪表板上方 |
    | RGB 摄像头 | 1080p, IR | 30fps | 仪表板上方 |
    | HRV 传感器 | 胸带/座椅电极 | 1Hz | 飞行服/座椅 |
    | 处理器 | 机载计算单元 | - | 电子舱 |
    """
    
    def __init__(self):
        super().__init__()
        
        # 眼动分支
        self.eye_branch = nn.Sequential(
            nn.Linear(8, 64),  # [PERCLOS, blink_dur, saccade_vel, gaze_entropy_x, gaze_entropy_y, pupil_dia, blink_freq, fixation_dur]
            nn.ReLU(),
            nn.Linear(64, 32),
            nn.ReLU(),
            nn.Linear(32, 16)
        )
        
        # HRV 分支
        self.hrv_branch = nn.Sequential(
            nn.Linear(6, 32),  # [LF_HF, RMSSD, SDNN, mean_HR, HR_trend, RR_var]
            nn.ReLU(),
            nn.Linear(32, 16)
        )
        
        # 行为分支（操纵杆/油门操作模式）
        self.behavior_branch = nn.Sequential(
            nn.Linear(5, 32),  # [stick_input_freq, throttle_var, response_time, input_entropy, auto_pilot_engage_freq]
            nn.ReLU(),
            nn.Linear(32, 16)
        )
        
        # 融合层
        self.fusion = nn.Sequential(
            nn.Linear(48, 128),
            nn.ReLU(),
            nn.Dropout(0.3),
            nn.Linear(128, 64),
            nn.ReLU(),
            nn.Dropout(0.2),
            nn.Linear(64, 3)  # [alert, mild_fatigue, severe_fatigue]
        )
        
        # 航空特定权重（眼动权重低于汽车DMS）
        self.modality_weights = nn.Parameter(torch.tensor([0.5, 0.3, 0.2]))  # 眼动50%, HRV30%, 行为20%
    
    def forward(self, eye_features, hrv_features, behavior_features):
        eye_out = self.eye_branch(eye_features)
        hrv_out = self.hrv_branch(hrv_features)
        behav_out = self.behavior_branch(behavior_features)
        
        # 加权融合
        w = torch.softmax(self.modality_weights, dim=0)
        fused = torch.cat([
            w[0] * eye_out,
            w[1] * hrv_out,
            w[2] * behav_out
        ], dim=-1)
        
        logits = self.fusion(fused)
        return logits


# 测试
if __name__ == "__main__":
    model = AviationFatigueDetector()
    
    # 模拟输入
    batch = 4
    eye = torch.randn(batch, 8)
    hrv = torch.randn(batch, 6)
    behav = torch.randn(batch, 5)
    
    with torch.no_grad():
        out = model(eye, hrv, behav)
    
    print("=== 航空疲劳检测器测试 ===")
    print(f"输出: {out.shape}")
    classes = ["alert", "mild_fatigue", "severe_fatigue"]
    for i, cls in enumerate(classes):
        print(f"  {cls}: {torch.softmax(out, dim=-1)[0, i]:.4f}")