An intelligent software system based on a combination of advanced actuarial methods, artificial intelligence and deep learning models, satellite imagery and fundamental forestry knowledge to improve accuracy.
The most advanced deep learning modules Keras, TensorFlow, PyTorch, and advanced models for processing Unet satellite images are used, allowing to perform image segmentation and precisely define forest boundaries.
Full automation of data collection.
Our software package operates 24/7, collecting data from satellites, using know-how algorithms to remove noise, cloud cover, and identify periods of abnormal phytomass reduction.
Developers - have many years of experience in developing nonlinear models, machine learning and deep learning models in Python, JavaScript, R, C++. Winners of hackathons in machine learning and big data analysis, Olympiads in financial modeling and business analysis, authors of scientific articles.
Working with big data using Dask technology and Hadoop clusters.
- Processing speed of one application (per 100 hectares of forest): from 3 to 10 days.
- Number of spectral layers of space images for analysis: from 6 to 15.
- Number of developed models, in particular, machine learning: from 3 to 8 (based on UNET, Catboost, LinearRegression, XGBOOST, LGBM, H20, KNN).
- Number of exogenous factors taken into account in the models: indices of vegetative density, meteorological, landscape characteristics, other: from 10 to 40.