Photovoltaic module detection and mapping<\/h1><\/span><\/div><\/div><\/div>a Supper & Supper Use Case<\/p>\n<\/div><\/div><\/div><\/div><\/div><\/div><\/div><\/div>
![](\"https:\/\/supperundsupper.com\/wp-content\/uploads\/2019\/09\/Photovoltaik.jpg\" "\\"Photovoltaic\\"")
<\/span><\/div><\/i><\/i><\/span>Project Goal<\/span><\/a><\/h4><\/div>\nPlanning for new renewable energy infrastructure is only possible if there is no information gap on the distribution of the already implemented infrastructure. To bridge this information gap, we integrated deep learning and GIS to detect and map photovoltaic (PV) panels in North Rhine-Westphalia (NRW) using remote sensing imagery.<\/p>\n
The objective of this project was to train a neural network that automatically detects the PV Panels and then Visualise them in a GIS system.<\/p>\n<\/div><\/div><\/div>
<\/i><\/i><\/span>Dataset used<\/span><\/a><\/h4><\/div>\nThe drone imagery was obtained from OpenGeoData portal of NRW as digital Orthophotos. The images (~47,818) had 4 spectral bands, a Spatial resolution of 0.1 meters and a spectral reference of EPSG 25832.<\/p>\n<\/div><\/div><\/div>
<\/i><\/i><\/span>Challenges & solutions<\/span><\/a><\/h4><\/div>\nThe undiscriminating color properties of the PV panels made the labelling process tedious since all the color proper-ties needed to be represent in the training sample. Secondly, the PV panels with smaller surface area generated no visual features to train and lastly there was optical deformation of the panels due to the roofs\u2019 slants. With digital mapping, optical distortions of the PV modules can occur depending on the orientation and the different roof pitches.<\/p>\n
In ArcGIS (ESRI), the training datasets were labelled (as Shapefiles) and Exported (as RCNN Masks) to TensorFlow Object Detection API (deep learning) for training. In TensorFlow the Mask RCNN was trained to detect the PV pan-els using the Training sample. The PV Panels were detected with bounding boxes then masks. Finally, the model was loaded in ArcGIS to detect the rest of the region.<\/p>\n<\/div><\/div><\/div>
a Supper & Supper Use Case<\/p>\n<\/div><\/div><\/div><\/div><\/div><\/div><\/div><\/div>
<\/span><\/div><\/i><\/i><\/span>Project Goal<\/span><\/a><\/h4><\/div>\nPlanning for new renewable energy infrastructure is only possible if there is no information gap on the distribution of the already implemented infrastructure. To bridge this information gap, we integrated deep learning and GIS to detect and map photovoltaic (PV) panels in North Rhine-Westphalia (NRW) using remote sensing imagery.<\/p>\n
The objective of this project was to train a neural network that automatically detects the PV Panels and then Visualise them in a GIS system.<\/p>\n<\/div><\/div><\/div>
<\/i><\/i><\/span>Dataset used<\/span><\/a><\/h4><\/div>\nThe drone imagery was obtained from OpenGeoData portal of NRW as digital Orthophotos. The images (~47,818) had 4 spectral bands, a Spatial resolution of 0.1 meters and a spectral reference of EPSG 25832.<\/p>\n<\/div><\/div><\/div>
<\/i><\/i><\/span>Challenges & solutions<\/span><\/a><\/h4><\/div>\nThe undiscriminating color properties of the PV panels made the labelling process tedious since all the color proper-ties needed to be represent in the training sample. Secondly, the PV panels with smaller surface area generated no visual features to train and lastly there was optical deformation of the panels due to the roofs\u2019 slants. With digital mapping, optical distortions of the PV modules can occur depending on the orientation and the different roof pitches.<\/p>\n
In ArcGIS (ESRI), the training datasets were labelled (as Shapefiles) and Exported (as RCNN Masks) to TensorFlow Object Detection API (deep learning) for training. In TensorFlow the Mask RCNN was trained to detect the PV pan-els using the Training sample. The PV Panels were detected with bounding boxes then masks. Finally, the model was loaded in ArcGIS to detect the rest of the region.<\/p>\n<\/div><\/div><\/div>
Planning for new renewable energy infrastructure is only possible if there is no information gap on the distribution of the already implemented infrastructure. To bridge this information gap, we integrated deep learning and GIS to detect and map photovoltaic (PV) panels in North Rhine-Westphalia (NRW) using remote sensing imagery.<\/p>\n
The objective of this project was to train a neural network that automatically detects the PV Panels and then Visualise them in a GIS system.<\/p>\n<\/div><\/div><\/div>
<\/i><\/i><\/span>Dataset used<\/span><\/a><\/h4><\/div>\nThe drone imagery was obtained from OpenGeoData portal of NRW as digital Orthophotos. The images (~47,818) had 4 spectral bands, a Spatial resolution of 0.1 meters and a spectral reference of EPSG 25832.<\/p>\n<\/div><\/div><\/div>
<\/i><\/i><\/span>Challenges & solutions<\/span><\/a><\/h4><\/div>\nThe undiscriminating color properties of the PV panels made the labelling process tedious since all the color proper-ties needed to be represent in the training sample. Secondly, the PV panels with smaller surface area generated no visual features to train and lastly there was optical deformation of the panels due to the roofs\u2019 slants. With digital mapping, optical distortions of the PV modules can occur depending on the orientation and the different roof pitches.<\/p>\n
In ArcGIS (ESRI), the training datasets were labelled (as Shapefiles) and Exported (as RCNN Masks) to TensorFlow Object Detection API (deep learning) for training. In TensorFlow the Mask RCNN was trained to detect the PV pan-els using the Training sample. The PV Panels were detected with bounding boxes then masks. Finally, the model was loaded in ArcGIS to detect the rest of the region.<\/p>\n<\/div><\/div><\/div>
The drone imagery was obtained from OpenGeoData portal of NRW as digital Orthophotos. The images (~47,818) had 4 spectral bands, a Spatial resolution of 0.1 meters and a spectral reference of EPSG 25832.<\/p>\n<\/div><\/div><\/div>
<\/i><\/i><\/span>Challenges & solutions<\/span><\/a><\/h4><\/div>\nThe undiscriminating color properties of the PV panels made the labelling process tedious since all the color proper-ties needed to be represent in the training sample. Secondly, the PV panels with smaller surface area generated no visual features to train and lastly there was optical deformation of the panels due to the roofs\u2019 slants. With digital mapping, optical distortions of the PV modules can occur depending on the orientation and the different roof pitches.<\/p>\n
In ArcGIS (ESRI), the training datasets were labelled (as Shapefiles) and Exported (as RCNN Masks) to TensorFlow Object Detection API (deep learning) for training. In TensorFlow the Mask RCNN was trained to detect the PV pan-els using the Training sample. The PV Panels were detected with bounding boxes then masks. Finally, the model was loaded in ArcGIS to detect the rest of the region.<\/p>\n<\/div><\/div><\/div>
The undiscriminating color properties of the PV panels made the labelling process tedious since all the color proper-ties needed to be represent in the training sample. Secondly, the PV panels with smaller surface area generated no visual features to train and lastly there was optical deformation of the panels due to the roofs\u2019 slants. With digital mapping, optical distortions of the PV modules can occur depending on the orientation and the different roof pitches.<\/p>\n
In ArcGIS (ESRI), the training datasets were labelled (as Shapefiles) and Exported (as RCNN Masks) to TensorFlow Object Detection API (deep learning) for training. In TensorFlow the Mask RCNN was trained to detect the PV pan-els using the Training sample. The PV Panels were detected with bounding boxes then masks. Finally, the model was loaded in ArcGIS to detect the rest of the region.<\/p>\n<\/div><\/div><\/div>
![](\"https:\/\/supperundsupper.com\/wp-content\/uploads\/2019\/09\/Pv_web1-1-1024x351.png\")
<\/a><\/span><\/div>\n<\/div><\/div><\/div><\/div><\/div>