Using Pléiades imagery combined with Disaitek’s AI algorithms!

Tonnes of illegal dumping of both industrial and domestic waste is regularly left on the roadside in our cities, in commercial areas or even in the middle of our beautiful regions. Illegal dumping of waste is becoming a serious environmental problem almost everywhere in the world and it is difficult for local governments to efficiently fight against it. Field investigations are becoming complicated and therefore resulting in high costs for public authorities.

This is where Pléiades satellites and Artificial Intelligence come to the rescue! Thanks to its true daily revisit capabilities, monitoring with Pléiades satellites is easy and allow to precisely and efficiently detect more dumping than the human eyes could do.

On top of fresh Pléiades imagery, Disaitek, a French company, performs their own AI algorithms, where the results are automatically integrated into a GIS system and delivered directly to users. They collaborate on workflows and assess environmental risks so that local governments can precisely locate waste and prioritize actions to quickly investigate the situation..

“During our R&D process, we have tried out different VHR optical sensors from satellite operators all over the world. Our conclusion was that Pléaides images were the best fit for our semantic segmentation deep learning model. The regular observations provided by the heliosynchronicity of the satellites combined to the expert preprocessing pipeline - including orthorectification, pansharpening, resampling - yield products with a steady quality over time, which is a key feature for applying our models. Indeed, the ground resolution allows the artificial intelligence to base its decisions, among other things, on the shape and the texture of the candidate, allowing it to filter out false positives such as rocks and gravel roads. Our technologies combined with Pléaides imagery allow us to detect waste piles as small as 5m2” said Anthony Graveline, President - Founder of Disaitek.