A system based on Artificial Intelligence is proposed, specifically, applying Artificial Vision and Deep Learning algorithms for the detection, location and classification of elements in images.
As input data sources, on the one hand, we have the CAD plan of the plant, which collects each of the regions and trackers that compose them in a georeferenced way. On the other hand, the images collected by a drone at the plant are reconstructed in a single GEOTIFF map, corresponding to a large size and resolution image, also georeferenced.
The Artificial Vision pipeline (processing flow) begins by pre-processing the images by areas. Subsequently, thanks to the recognition models based on neural networks and trained ad-hoc, each component present in the image is segmented, finally identified, classified according to its type.
Subsequently, the combined analysis of the components foreseen according to the plan and those actually visible in the images is carried out. Its presence and position intersect between both origins thanks to the GPS coordinates.
Lastly, the data is available for uploading to a dashboard designed by Gonvarri Solar Steel's engineering team, so that its display is representative and navigable according to monitoring requirements.