openEuler操作系统环境：目标检测软件开发实操与性能评估

# openEuler内置AI框架支持验证
$ yum search tensorflow
tensorflow.x86_64 : TensorFlow is an open source machine learning framework
$ yum search pytorch
pytorch.x86_64 : PyTorch is an open source machine learning framework

# 示例标注文件 bird_001.xml
<annotation>
  <folder>bird_images</folder>
  <filename>bird_001.jpg</filename>
  <size>
    <width>640</width>
    <height>480</height>
  </size>
  <object>
    <name>swallow</name>
    <bndbox>
      <xmin>120</xmin>
      <ymin>85</ymin>
      <xmax>356</xmax>
      <ymax>297</ymax>
    </bndbox>
  </object>
</annotation>

bird_dataset/
├── JPEGImages/      # 原始图片
├── Annotations/     # XML标注文件
└── ImageSets/
    └── Main/
        ├── train.txt
        └── val.txt

# train.py
from ultralytics import YOLO
# 加载预训练模型
model = YOLO('yolov8n.pt')  # nano版本适合快速原型开发
# 训练配置
results = model.train(
    data='bird_dataset.yaml',  # 数据集配置文件
    epochs=100,
    imgsz=640,
    batch=16,
    device=0,  # GPU加速
    project='bird_detection',
    name='exp'
)

path: /data/bird_dataset
train: images/train
val: images/val
nc: 5  # 类别数
names: ['swallow', 'sparrow', 'magpie', 'pigeon', 'eagle']

# inference_test.py
from ultralytics import YOLO
import time
import cv2
model = YOLO('runs/detect/exp/weights/best.pt')
img = cv2.imread('test_bird.jpg')
# 预热GPU
for _ in range(10):
    model(img)
# FPS测试
times = []
for _ in range(100):
    start = time.time()
    results = model(img)
    times.append(time.time() - start)
fps = 100 / sum(times)
print(f"平均推理速度: {fps:.2f} FPS")

# Dockerfile
FROM openeuler-22.03-lts-sp3:latest
RUN yum update -y && \
    yum install python3-pip -y && \
    pip install ultralytics==8.0.196
COPY best.pt /app/
COPY detect.py /app/
CMD ["python3", "/app/detect.py"]

# 构建与运行容器
$ podman build -t bird-detector:oe22 .
$ podman run --gpus all -v /data:/app/data bird-detector:oe22