Downloads · 30 days
33
28% of all-time downloads
sanqierere/yolo26m-UAV-fire-smoke-detection
yolo26m-UAV-fire-smoke-detection is a object detection model from sanqierere. Use it when you need objects located in an image. It is set up for ultralytics. The card lists the license as agpl-3.0.
无人机林地巡检烟火检测模型 · 基于 YOLOv26m 在 FASDD 数据集上微调
Downloads · 30 days
33
28% of all-time downloads
All-time downloads
120
Public
Repo size
44.9 MB
Likes
0
Public
Click a slice to open those files.
.pt44 MB · 98%
From the Hugging Face model README
无人机林地巡检烟火检测模型 · 基于 YOLOv26m 在 FASDD 数据集上微调

本模型用于无人机航拍林地场景下的烟火目标检测,可同时识别两类目标:
基于 COCO 预训练的 YOLOv26m,在 FASDD(Forest Aerial Smoke/Fire Detection)数据集及 自制无人机森林火灾数据集 上微调得到。适用于林地巡检、森林防火、火情早期预警等无人机遥感场景。
| 指标 Metric | 数值 Value |
|---|---|
| [email protected] | 0.843 |
| [email protected]:0.95 | 0.594 |
| Precision | 0.808 |
| Recall | 0.821 |
| 输入尺寸 Input Size | 640 × 640 |
| 参数 Params | YOLOv26m(约 40M) |
指标取自最优 epoch(156),使用 SGD + cosine LR 训练 200 epochs。

| 配置项 | 值 |
|---|---|
| 基础模型 Base | YOLOv26m(COCO 预训练) |
| 数据 Data | FASDD(fire / smoke 两类) |
| 优化器 Optimizer | SGD |
| 学习率策略 LR Schedule | cosine (lr0=0.005, lrf=0.001) |
| Epochs | 200(patience=30) |
| Batch Size | 16 |
| Image Size | 640 |
| 增强 Augmentation | degrees=30, flipud=0.5, fliplr=0.5, scale=0.5, hsv 等 |
| 其他 | label_smoothing=0.1, multi_scale=0.3, mosaic=0 |
from ultralytics import YOLO
model = YOLO("yolo26m-fasdd-fire-smoke-detection.pt") # 或你的本地 best.pt 路径
# 图片检测
results = model.predict("aerial_image.jpg", conf=0.25, imgsz=640)
# 视频检测(无人机巡检视频)
results = model.predict("drone_video.mp4", conf=0.25, imgsz=640)
# 验证指标
metrics = model.val()
from huggingface_hub import hf_hub_download
from ultralytics import YOLO
model_path = hf_hub_download(
"YOUR_HF_USERNAME/yolo26m-fasdd-fire-smoke-detection",
"weights/best.pt",
)
model = YOLO(model_path)
yolo predict model=weights/best.pt source=aerial_image.jpg imgsz=640 conf=0.25
yolo26m-fasdd-fire-smoke-detection/
├── README.md # 模型说明(本文件)
├── args.yaml # 完整训练参数
├── results.png # 训练曲线
├── BoxPR_curve.png # P-R 曲线
├── confusion_matrix.png # 混淆矩阵
├── val_batch0_pred.jpg # 验证集预测示例
└── weights/
└── best.pt # 最优权重(44 MB)
conf=0.25,可视场景密度调整(0.15–0.35)。max_det=300。如果你使用了本模型,欢迎引用:
@misc{outsideguy2026yolo26mfasdd,
title={YOLOv26m UAV Forest Patrol Fire and Smoke Detection},
author={OutsideGuy},
year={2026},
howpublished={\url{https://huggingface.co/YOUR_HF_USERNAME/yolo26m-fasdd-fire-smoke-detection}},
}
如果这个模型有帮助,别忘了 ⭐ Star / ❤️ Follow
</div>