基于YOLO26的玻璃缺陷检测系统（Python源码+数据集+Pyside6界面）

重复模块m (n次迭代):
┌─────────────────────────────────────────────────────────┐
│                                                         │
│  如果 attn=True:                                        │
│    Sequential(                                          │
│        Bottleneck(self.c, self.c),                     │  ←─ 先特征提取
│        PSABlock(self.c, attn_ratio=0.5, num_heads=...) │  ←─ 后注意力增强
│    )                                                    │
│                                                         │
│  否则如果 c3k=True:                                     │
│    C3k(self.c, self.c, 2)                             │  ←─ 同YOLOv11
│                                                         │
│  否则:                                                  │
│    Bottleneck(self.c, self.c)                         │  ←─ 同YOLOv11
│                                                         │
└─────────────────────────────────────────────────────────┘

class BboxLoss(nn.Module):
    """Criterion class for computing training losses for bounding boxes."""

    def __init__(self, reg_max: int = 16):
        """Initialize the BboxLoss module with regularization maximum and DFL settings."""
        super().__init__()
        self.dfl_loss = DFLoss(reg_max) if reg_max > 1 else None

# glass_defect_yolo26_pyside6.py
# 伪代码：基于 YOLO26 + PySide6 的玻璃缺陷检测系统
# 功能：数据集配置、模型训练、图片/文件夹检测、结果展示、缺陷统计、日志
# 依赖：pip install ultralytics pyside6 pyyaml opencv-python

import sys
from pathlib import Path
import yaml
from PySide6.QtCore import Qt, QThread, Signal
from PySide6.QtGui import QPixmap
from PySide6.QtWidgets import (
    QApplication, QMainWindow, QWidget, QLabel, QPushButton, QTextEdit,
    QFileDialog, QVBoxLayout, QHBoxLayout, QListWidget, QMessageBox
)
from ultralytics import YOLO

# -------------------- 配置 --------------------
DATASET_DIR = Path("glass_defect_dataset")
DATA_YAML = DATASET_DIR / "data.yaml"
MODEL_WEIGHT = "yolo26n.pt"   # 替换为实际 YOLO26 权重
PROJECT_DIR = "runs/glass_defect"

CLASS_NAMES = {
    0: "scratch",      # 划痕
    1: "bubble",       # 气泡
    2: "crack",        # 裂纹
    3: "inclusion",    # 夹杂
    4: "stain",        # 污渍
}

# -------------------- 数据集配置 --------------------
def ensure_data_yaml():
    """若 data.yaml 不存在则自动生成"""
    if DATA_YAML.exists():
        return
    data = {
        "path": str(DATASET_DIR.resolve()),
        "train": "images/train",
        "val": "images/val",
        "names": CLASS_NAMES,
    }
    DATA_YAML.parent.mkdir(parents=True, exist_ok=True)
    with open(DATA_YAML, "w", encoding="utf-8") as f:
        yaml.safe_dump(data, f, allow_unicode=True)

# -------------------- 训练 --------------------
def train_yolo():
    ensure_data_yaml()
    model = YOLO(MODEL_WEIGHT)
    model.train(
        data=str(DATA_YAML),
        epochs=100,
        imgsz=640,
        batch=16,
        project=PROJECT_DIR,
        name="train",
    )

# -------------------- 检测线程 --------------------
class DetectionWorker(QThread):
    log_signal = Signal(str)
    result_signal = Signal(str, dict)  # 结果图路径, 缺陷统计

    def __init__(self, source, conf=0.25, iou=0.5):
        super().__init__()
        self.source = source
        self.conf = conf
        self.iou = iou

    def run(self):
        try:
            self.log_signal.emit(f"加载模型: {MODEL_WEIGHT}")
            model = YOLO(MODEL_WEIGHT)
            self.log_signal.emit(f"开始检测: {self.source}")

            results = model.predict(
                source=self.source,
                conf=self.conf,
                iou=self.iou,
                save=True,
                project=PROJECT_DIR,
                name="predict",
            )

            counts = {name: 0 for name in CLASS_NAMES.values()}
            result_image = ""

            for r in results:
                for cls_id in r.boxes.cls:
                    name = model.names[int(cls_id)]
                    counts[name] = counts.get(name, 0) + 1
                # 伪代码：取结果图路径
                result_image = str(Path(r.save_dir) / Path(r.path).name)

            self.result_signal.emit(result_image, counts)
            self.log_signal.emit(f"检测完成，统计: {counts}")

        except Exception as e:
            self.log_signal.emit(f"检测失败: {e}")

# -------------------- 主界面 --------------------
class GlassDefectWindow(QMainWindow):
    def __init__(self):
        super().__init__()
        self.setWindowTitle("基于 YOLO26 的玻璃缺陷检测系统")
        self.resize(1200, 800)

        self.current_source = None
        self.worker = None

        self.init_ui()

    def init_ui(self):
        central = QWidget()
        self.setCentralWidget(central)
        main_layout = QHBoxLayout(central)

        # 左侧：按钮 + 统计 + 日志
        left_layout = QVBoxLayout()

        self.btn_select_image = QPushButton("选择图片")
        self.btn_select_folder = QPushButton("选择文件夹")
        self.btn_detect = QPushButton("开始检测")
        self.btn_train = QPushButton("训练模型")

        self.list_counts = QListWidget()
        self.text_log = QTextEdit()
        self.text_log.setReadOnly(True)

        left_layout.addWidget(self.btn_select_image)
        left_layout.addWidget(self.btn_select_folder)
        left_layout.addWidget(self.btn_detect)
        left_layout.addWidget(self.btn_train)
        left_layout.addWidget(QLabel("缺陷统计"))
        left_layout.addWidget(self.list_counts)
        left_layout.addWidget(QLabel("运行日志"))
        left_layout.addWidget(self.text_log)

        # 右侧：原图 + 结果图
        right_layout = QVBoxLayout()

        self.label_original = QLabel("原图")
        self.label_result = QLabel("检测结果")

        for lab in (self.label_original, self.label_result):
            lab.setAlignment(Qt.AlignCenter)
            lab.setMinimumSize(400, 300)
            lab.setStyleSheet("border: 1px solid gray;")

        right_layout.addWidget(self.label_original)
        right_layout.addWidget(self.label_result)

        main_layout.addLayout(left_layout, 1)
        main_layout.addLayout(right_layout, 2)

        # 信号连接
        self.btn_select_image.clicked.connect(self.select_image)
        self.btn_select_folder.clicked.connect(self.select_folder)
        self.btn_detect.clicked.connect(self.start_detect)
        self.btn_train.clicked.connect(self.start_train)

    def log(self, msg):
        self.text_log.append(msg)

    def select_image(self):
        path, _ = QFileDialog.getOpenFileName(
            self, "选择图片", "", "Images (*.jpg *.jpeg *.png *.bmp)"
        )
        if path:
            self.current_source = path
            self.show_image(path, self.label_original)
            self.log(f"选择图片: {path}")

    def select_folder(self):
        path = QFileDialog.getExistingDirectory(self, "选择文件夹")
        if path:
            self.current_source = path
            self.log(f"选择文件夹: {path}")

    def show_image(self, path, label):
        pix = QPixmap(path)
        if not pix.isNull():
            label.setPixmap(
                pix.scaled(label.size(), Qt.KeepAspectRatio, Qt.SmoothTransformation)
            )

    def start_detect(self):
        if not self.current_source:
            QMessageBox.warning(self, "提示", "请先选择图片或文件夹")
            return

        self.list_counts.clear()
        self.worker = DetectionWorker(self.current_source)
        self.worker.log_signal.connect(self.log)
        self.worker.result_signal.connect(self.on_result)
        self.worker.start()

    def on_result(self, image_path, counts):
        if image_path and Path(image_path).exists():
            self.show_image(image_path, self.label_result)

        for name, num in counts.items():
            self.list_counts.addItem(f"{name}: {num}")

    def start_train(self):
        # 伪代码：实际项目应放入线程，避免界面卡顿
        self.log("开始训练 YOLO26 模型...")
        train_yolo()
        self.log("训练完成")

# -------------------- 入口 --------------------
if __name__ == "__main__":
    app = QApplication(sys.argv)
    window = GlassDefectWindow()
    window.show()
    sys.exit(app.exec())

path: 'D:/YOLO26/data/glass-defect' # dataset root dir
train: images/train
val: images/val
test: images/test

# Classes
names:
  0: "DEFECT"

import warnings
warnings.filterwarnings('ignore')
from ultralytics import YOLO

if __name__ == '__main__':
    model = YOLO('ultralytics/cfg/models/26/yolo26.yaml')
    model.train(data='data/glass-defect.yaml',
                cache=False,
                imgsz=640,
                epochs=200,
                batch=8,
                close_mosaic=10,
                device='0',
                optimizer='SGD', # using SGD
                project='runs/train',
                name='exp',
                )

YOLO26 summary (fused): 122 layers, 2,375,031 parameters, 0 gradients, 5.3 GFLOPs
                 Class     Images  Instances      Box(P          R      mAP50  mAP50-95): 100% ━━━━━━━━━━━━ 4/4 2.8it/s 1.4s
                   all        122        138      0.963       0.95      0.982      0.778

pip install --upgrade pip
pip install pyside6 -i https://mirror.baidu.com/pypi/simple