agademer/yolo-remote-sensing-photovoltaic

YOLO Remote Sensing Photovoltaic Detection

📌 Overview

This model detects buildings and rooftop photovoltaic panels from remote sensing imagery using YOLO.

• Model type: YOLOv8 / YOLO26

• Task: Object Detection / Segmentation

• Classes:

  • building
  • photovoltaic panel

---

Dataset

• training-dataset-2 : {'train': 461, 'test': 66, 'val': 131}
• training-dataset-seg : {'train': 466, 'test': 71, 'val': 122}

IGN's Orthophotos (960x960 pixels aka 192x192 meters) from South-West of France (Castelnaudary to Perpignan), Montpellier surroundings and South of Paris. Rooftop photovoltaic panels and buildings shapes identified with the help of OpenStreetMap.

🚀 Usage

Install dependencies

pip install ultralytics

---

Run inference

from ultralytics import YOLO

model = YOLO("yolo-remote-sensing-photovoltaic-v26l-solar-farms-and-cities-v20260402-segment-100_epochs.pt")

results = model("example_input.png", save=True)

---

🖼️ Example

Input image:

Output image:

---

📦 Files

yolo-remote-sensing-photovoltaic-v26l-solar-farms-and-cities-v20260403-segment-300_epochs.pt

+200 Epochs (from the previous model) - +5h of training on L40S - based on yolo-remote-sensing-photovoltaic-v26l-solar-farms-and-cities-v20260402-segment-100_epochs.pt - Segmentation task

training-dataset-seg / val

Class	Images	Instances	Box P	Box R	Box mAP50	Box mAP50-95	Mask P	Mask R	Mask mAP50	Mask mAP50-95
all	122	4792	0.574	0.428	0.424	0.23	0.558	0.384	0.378	0.173
building	108	4393	0.581	0.353	0.376	0.182	0.57	0.314	0.331	0.129
photovoltaic panel	121	399	0.568	0.504	0.473	0.278	0.547	0.454	0.426	0.217

training-dataset-seg / test (never seen before)

Class	Images	Instances	Box P	Box R	Box mAP50	Box mAP50-95	Mask P	Mask R	Mask mAP50	Mask mAP50-95
all	71	2526	0.652	0.481	0.508	0.26	0.621	0.435	0.425	0.172
building	68	2313	0.667	0.385	0.433	0.229	0.656	0.367	0.404	0.173
photovoltaic panel	71	213	0.637	0.577	0.582	0.291	0.586	0.502	0.447	0.171

yolo-remote-sensing-photovoltaic-v26l-solar-farms-and-cities-v20260402-segment-100_epochs.pt

100 Epochs - 3.5h of training on L40S - yolo26l - Segmentation task

training-dataset-seg / val

Class	Images	Instances	Box P	Box R	Box mAP50	Box mAP50-95	Mask P	Mask R	Mask mAP50	Mask mAP50-95
all	122	4792	0.552	0.413	0.402	0.213	0.519	0.383	0.352	0.163
building	108	4393	0.566	0.339	0.359	0.175	0.539	0.319	0.319	0.127
photovoltaic panel	121	399	0.538	0.486	0.444	0.251	0.498	0.446	0.385	0.199

training-dataset-seg / test (never seen before)

Class	Images	Instances	Box P	Box R	Box mAP50	Box mAP50-95	Mask P	Mask R	Mask mAP50	Mask mAP50-95
all	71	2526	0.581	0.462	0.461	0.236	0.528	0.412	0.399	0.17
building	68	2313	0.636	0.379	0.429	0.225	0.611	0.364	0.401	0.172
photovoltaic panel	71	213	0.526	0.545	0.494	0.246	0.444	0.46	0.397	0.168

yolo-remote-sensing-photovoltaic-v26l-solar-farms-and-cities-v20260401-segment-10_epochs.pt

10 Epochs - 20 minutes of training on L40S - yolo26l - Segmentation task

training-dataset-seg / val

Class	Images	Instances	Box P	Box R	Box mAP50	Box mAP50-95	Mask P	Mask R	Mask mAP50	Mask mAP50-95
all	122	4792	0.414	0.340	0.291	0.151	0.395	0.307	0.255	0.114
building	108	4393	0.503	0.252	0.266	0.119	0.478	0.225	0.230	0.084
photovoltaic panel	121	399	0.325	0.429	0.317	0.183	0.313	0.388	0.281	0.144

training-dataset-seg / test (never seen before)

Class	Images	Instances	Box P	Box R	Box mAP50	Box mAP50-95	Mask P	Mask R	Mask mAP50	Mask mAP50-95
all	71	2526	0.402	0.353	0.294	0.144	0.404	0.314	0.259	0.109
building	68	2313	0.505	0.307	0.33	0.16	0.52	0.288	0.31	0.123
photovoltaic panel	71	213	0.298	0.399	0.257	0.127	0.287	0.341	0.208	0.0953

yolo-remote-sensing-photovoltaic-v8l-solar-farms-and-cities-v20260331-detect-1000_epochs.pt

1000 Epochs - 3.6h of training on L40S - yolov8l - Detection task

training-dataset-2 / val

Class	Images	Instances	Box P	Box R	Box mAP50	Box mAP50-95
all	131	5101	0.655	0.382	0.412	0.235
building	117	4629	0.639	0.268	0.312	0.148
photovoltaic panel	131	472	0.671	0.496	0.512	0.322

training-dataset-2 / test (never seen before)

Class	Images	Instances	Box P	Box R	Box mAP50	Box mAP50-95
all	66	3050	0.529	0.375	0.358	0.174
building	64	2823	0.551	0.336	0.341	0.172
photovoltaic panel	66	227	0.507	0.414	0.374	0.176

yolo-remote-sensing-photovoltaic-v8l-solar-farms-and-cities-v20260329-detect-100_epochs.pt

100 Epochs - 22 minutes of training on L40S - yolov8l - Detection task

training-dataset-2 / val

Class	Images	Instances	Box P	Box R	Box mAP50	Box mAP50-95
all	131	5101	0.506	0.423	0.403	0.228
building	117	4629	0.566	0.336	0.353	0.168
photovoltaic panel	131	472	0.447	0.511	0.454	0.287

training-dataset-2 / test (never seen before)

Class	Images	Instances	Box P	Box R	Box mAP50	Box mAP50-95
all	66	3050	0.521	0.383	0.375	0.184
building	64	2823	0.574	0.353	0.381	0.197
photovoltaic panel	66	227	0.468	0.414	0.369	0.17
photovoltaic panel	66	227	0.507	0.414	0.374	0.176

---

⚠️ Notes

• Designed for aerial / high resolution satellite imagery

---

🤝 Credits

Trained using YOLO and tailored remote sensing datasets.

---