实战项目二：工业缺陷检测

📂 所属阶段：第六阶段 — 顶级综合项目实战
🔗 相关章节：实战项目一：智能人脸考勤系统 · 实战项目三：自动驾驶感知

1. 系统架构

输入：产品图像
↓
预处理（标准化、增强）
↓
特征提取（CNN）
↓
异常检测（Isolation Forest / 深度学习）
↓
输出：合格/不合格

2. 异常检测方法

import torch
import torch.nn as nn
from sklearn.ensemble import IsolationForest

# 方法一：Isolation Forest
class DefectDetector:
    def __init__(self):
        self.model = IsolationForest(contamination=0.1)
    
    def train(self, normal_images):
        # 提取特征
        features = self.extract_features(normal_images)
        self.model.fit(features)
    
    def detect(self, image):
        features = self.extract_features([image])
        prediction = self.model.predict(features)
        return prediction[0] == -1  # -1 表示异常
    
    def extract_features(self, images):
        # 使用预训练 CNN 提取特征
        model = torch.hub.load('pytorch/vision:v0.10.0', 'resnet50', pretrained=True)
        features = []
        for img in images:
            with torch.no_grad():
                feat = model(img.unsqueeze(0))
            features.append(feat.numpy())
        return np.concatenate(features)

3. 深度学习异常检测

import torch
import torch.nn as nn

class AutoEncoder(nn.Module):
    def __init__(self):
        super().__init__()
        self.encoder = nn.Sequential(
            nn.Conv2d(3, 32, 3, padding=1),
            nn.ReLU(),
            nn.MaxPool2d(2),
            nn.Conv2d(32, 64, 3, padding=1),
            nn.ReLU(),
            nn.MaxPool2d(2),
        )
        self.decoder = nn.Sequential(
            nn.ConvTranspose2d(64, 32, 4, stride=2, padding=1),
            nn.ReLU(),
            nn.ConvTranspose2d(32, 3, 4, stride=2, padding=1),
            nn.Sigmoid(),
        )
    
    def forward(self, x):
        encoded = self.encoder(x)
        decoded = self.decoder(encoded)
        return decoded

# 训练：只用正常产品图像
model = AutoEncoder()
optimizer = torch.optim.Adam(model.parameters())
criterion = nn.MSELoss()

for epoch in range(100):
    for normal_image in normal_images:
        output = model(normal_image)
        loss = criterion(output, normal_image)
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

# 检测：重建误差大 = 异常
def detect_defect(image):
    output = model(image)
    error = torch.mean((output - image) ** 2)
    return error > threshold

4. 完整系统

class IndustrialDefectDetection:
    def __init__(self):
        self.model = AutoEncoder()
        self.threshold = 0.05
    
    def process_image(self, image_path):
        # 读取图像
        image = cv2.imread(image_path)
        image = cv2.resize(image, (224, 224))
        image = torch.from_numpy(image).permute(2, 0, 1).float() / 255.0
        
        # 检测
        output = self.model(image.unsqueeze(0))
        error = torch.mean((output - image.unsqueeze(0)) ** 2)
        
        if error > self.threshold:
            return "不合格", error.item()
        else:
            return "合格", error.item()
    
    def process_video(self, video_path):
        cap = cv2.VideoCapture(video_path)
        results = []
        
        while cap.isOpened():
            ret, frame = cap.read()
            if not ret:
                break
            
            status, error = self.process_image(frame)
            results.append((status, error))
        
        cap.release()
        return results

5. 小结

工业缺陷检测关键技术：

1. 异常检测：Isolation Forest 或 AutoEncoder
2. 特征提取：预训练 CNN
3. 阈值设置：根据业务需求调整
4. 实时处理：视频流处理

性能指标：
- 检测速度：30+ FPS
- 准确率：95%+
- 误检率：< 5%

2026 年趋势：
- 多模态检测（RGB + 红外）
- 实时 3D 检测
- 边缘计算部署

💡 记住：异常检测是工业 AI 的核心应用。掌握它，你就掌握了工业 4.0 的关键技术。

🔗 扩展阅读