THE USE OF VHR REMOTE SENSING IMAGERY FOR THE IDENTIFICATION OF ROOFS POTENTIALLY SUITABLE FOR THE INSTALLATION OF PHOTOVOLTAIC PANELS ORFEO PLEIADES December 8 th, 2010
OBJECTIVES Develop a method to automatically identify roofs suitable for the installation of photovoltaic panels. Assess the possibilities of VHR image processing. Assess the analytical abilities of OTB-Monteverdi software. 1/24
MATERIALS Images QuickBird images (Toulouse, France) Image processing software OTB Monteverdi v.1.0 IDRISI Andes v.15.0 GIS software MapInfo v.8.0 2/24
IMAGE GENERATION To generate suitable images Fusion made with OTB-Monteverdi + = 1 band R = 0.61 m 4 bands 4 bands R = 2.44 m R = 0.61 m 3/24
IMAGE GENERATION Images suitable for processing Very long generation time High-performance computer required 4/24
IMAGE CLASSIFICATION To extract roofs from the image 3 main roof classes: Red = tiles White = steel Grey = asphalt or fibrocement 5/24
IMAGE CLASSIFICATION Many other classes Roads Bare soil Vegetation Shadows Water OTB = 2 classification methods Object-based Per-pixel 6/24
IMAGE CLASSIFICATION : Global KIA Object and pixel = equivalent results Quality depends on heterogeneity 7/24
IMAGE CLASSIFICATION : Specific KIA Red and white roofs often well classified 7/24
IMAGE CLASSIFICATION : Specific KIA Red and white roofs often well classified Grey roofs often misclassified 7/24
IMAGE CLASSIFICATION Confusions Grey roofs/roads same coating 8/24
IMAGE CLASSIFICATION Confusions Grey roofs/roads same coating Grey roofs/roads/shady red roofs 8/24
IMAGE CLASSIFICATION Confusions Grey roofs/roads same coating Grey roofs/roads/shady red roofs Red roofs/bare soil 8/24
IMAGE CLASSIFICATION with indexes RI, PSI and Length Equivalent to per-pixel approach 9/24
IMAGE CLASSIFICATION Problems 2 approaches = equivalent results Roof s form and surface accurate enough Indexes adding significant improvement In these conditions, how to answer the question? 10/24
IMAGE CLASSIFICATION Adding information High Resolution DEM Second VHR scene Masking other classes Cadastral plan Topographic database 11/24
IMAGE CLASSIFICATION Adding information High Resolution DEM Second VHR scene Expensive Masking other classes Cadastral plan digital version not always available Topographic database : BD TOPO IGN suitable characteristics, cheap 11/24
MASKING BD TOPO vector raster (IDRISI) Band math + concatenate (OTB- Monteverdi) 12/24
MASKING : Global KIA Benefits to heterogeneous areas 13/24
MASKING : Specific KIA Red and white roofs = equivalent results 13/24
MASKING : Specific KIA Red and white roofs = equivalent results Grey roofs = improvement in each area 13/24
MASKING Titre diapo Titre diapo Titre diapo 14/24
GIS treatment DATA CORRECTION Vectorize classification Class extraction for treatment Assign a unique class 1 roof = often many classes B V R Class 1 Class 2 Class 3 15/24 Labell BD TOPO Polygons from classification = not suitable for treatment
GIS treatment DATA CORRECTION Superposition: classified shape/image Success rate > 90% Succes rate with grey class dissociation ~80% Class Detected Well qualified Rate of correct qualification Red 147 147 100% White 4 1 25% Grey 12 7 58% Total 163 155 95% Downtown Class Detected Well qualified Rate of correct qualification Red 110 104 95% White 13 13 100% Grey 81 68 84% Total 204 185 91% Industrial park 16/24
GIS treatment CONTIGUOUS BUILT DETECTION Based on 3 criteria Class (red or grey) Surface (>250 m²) Polygon smoothness (red >15 faces; grey > 30 faces) Complex detection Strongly depending on thresholds Omission / Over-selection risk Unlikely contiguous Likely contiguous 17/24
GIS treatment DISTANCE TO PROTECTED SITES Based on a buffer Restriction = 500 m Protected site layers available Easy detection Restricted area Restricted polygons Not restricted polygons 18/24
GIS treatment SLOPE ASSIGNEMENT Linked to the class Source: Regulatory documents Slaters Roof ridges not detectable: Roofs with many slopes? Red = 25 to 35% (Toulouse ~ 33%) White Steep grey = 5 to 8% Flat grey =0% 19/24
GIS treatment SURFACE AREA CALCULATION Corrected by slope Real area = calculated area / cos (slope ) Roof s ridge not detectable Postulate: available surface = ½ surface Only for ideal roofs total 20/24
GIS treatment ORIENTATION CALCULATION Based on each face s azimuth Interesting azimuths L(interesting faces)/perimeter Ratio > 0.5 Roof s ridges not detectable Postulate: the ridge follows the roof s length Only for ideal roofs Interesting Not interesting 21/24
CRITERIA Area = 15 km² and 4 439 polygons Red and white roof s surface area > 1400 m² Grey roofs surface area > 700 m² Contiguous built = unlikely Checked by a technician from a firm specialised in installation of solar panels 22/24
FINAL RESULTS Rate of commission = 48.6% Method\Truth Suitable roofs Unsuitable roofs Total Commission error Suitable roofs 111 105 216 0.486 Unsuitable roofs 3 4220 4223 0.001 Total 114 4325 4439 Omission error 0.026 0.024 0.024 23/24
FINAL RESULTS Rate of commission = 48.6% But Rate of omission = 2.6% Method\Truth Suitable roofs Unsuitable roofs Total Commission error Suitable roofs 111 105 216 0.486 Unsuitable roofs 3 4220 4223 0.001 Total 114 4325 4439 Omission error 0.026 0.024 0.024 23/24
FINAL RESULTS Rate of commission = 48.6% But Rate of omission = 2.6% Rate of correct qualification = 97.6% Method\Truth Suitable roofs Unsuitable roofs Total Commission error Suitable roofs 111 105 216 0.486 Unsuitable roofs 3 4220 4223 0.001 Total 114 4325 4439 Omission error 0.026 0.024 0.024 23/24
METHOD Reduces the amount of work Quite good precision Powerful computer required OTB-MONTEVERDI Many indispensable tools Good quality and fast «pixel» classification «Object» does not work with mask 24/24
VHRS IMAGES Large area covered Sufficient resolution Cannot efficiently identify roofs without complementary information THANK YOU 24/24