The opposite of “open” isn’t “closed”. The opposite of “open” is “broken”. — John Wilbanks
Modules:
Context
The company did conduct core research on problems that cut across the data sciences and engineering. The general research scope focused on formal and mathematical models for data processing, as well as on issues concerning the engineering of large-scale data processing systems.The greatest technical goal at this project have been the application of data-driven approaches to extract information from satellite. An important part of my work consisted of the devising novel methods to statistically learn about patterns of life of any defined and observed area, and how it changes over time. Generally speaking, the tasks consisted in the revision of methods for analysis geolocating and longitudinal data, in particular on anomaly detection and data fusion techniques. This partially includes trajectory reconstruction and spatio-temproral normalcy recognition for addressing today societal challenges. Deep learning algorithms were found well fitted into our analysis. I also found useful the DBSCAN algorithm in order to cluster different regions using instead, maritime data as AIS dataset. When task were intented to deal with commercial inputs (e.g. heatmap data visualisation demo), I’ve used eolearn, javascrpit and google maps API to portrait data into imagery layer.
Lessons learned
I personally learnt a vast amount of code sources (includes also data project management skills such tools and documentation work breakdown structure) and met people involved in such complex challenges from all around the world (European Space Agency - Copernicus Programme). Also, worth to mention the tasks performed for the SDG-14 to halt illegal maritime activities.
Related projects
Bayesian simulation of EO Data Mission:
Mimic of social network has been addressed using Monte Carlo simulation in order to estimate the probability of detection for the satellite. I performed t-test statistics for testing hypothesis about sample size and means groups and also gave advice in the MC Simulation. I also had experience with NoSql and json file format (MongoDB) in order to retrieve data.
Machine Learning (ML) and Deep Learning Pipeline in Remote Sensing:
A Recurrent Neural Network (RNN) approach architecture for classification of objects from Satellite Imagery is being considering within my team. The proposed RNN architecture is expected to offer a viable operational solution for pipeline integration into multitude of potential applications in the areas of precision and monitoring risk & damage assessment.
The under construction “geo-located data” pipeline (e.g. formalization and algorithms not complete yet) significantly reduces the complexity of creating models by removing the need for handcrafted filtering, and making it a cost-effective option for bringing neural network models to the market.
Big Data EO open-source Software:
Significant efforts were concentrated on the development of software based on multi-temporal, multi-sensor datasets over large geographical regions. Software Engineer, System Engineer and I myself were defining the requirements for such software, IT infrastructure solution and computing capabilities for the four-stage pipeline (1) stored, (2) processed and (3) ready for analysis. Commercial and collaborative approach solutions were taking into account for the sake of data latency and many other resources under management requirements.