We released DBPFinder as an open-source web-based tool designed to perform strategic sensor placement in chlorinated water distribution networks, aiming to optimize the detection of dissinfection byproducts (DBPs), formed when chlorinated water interacts with natural occurring matter.
We received a Best Paper Award from ACS ES&T Water (IF: 4.3) for our research paper titled “Assessing the Health Impact of Disinfection Byproducts in Drinking Water”.
We finished successfully the H2OFORALL Horizon Europe project, leading the Work Package related to sensing DBPs in chlorinated drinking water.
We exploited commercially the natural risks for real estate properties, included as services in the GAEA country-scale geospatial tool, having local banks and insurance companies as clients.
We created the Meliferea chatbot, embedded into the GAEA tool, investigating how large language models can interact with geoanalytical services using natural human language.
We proposed a methodology leveraging multi-modal deep learning and data-imputation techniques to improve frog counting and occurrence prediction, creating high-confidence Species Distribution Models (SDM) for frogs.
We published DAFR2, a simple yet effective framework for unsupervised domain adaptation via feature refinementunder under distribution shift. The proposed algorithm outperforms prior methods in robustness to corruption.
As part of the European PHENOWEX project, we created a data story that describes how we leveraged LiDAR data obtained from drones to calculate wheat plant height and used a forecasting model to predict future wheat plant growth.
We have classified the key tree species of the island of Cyprus via high spatial resolution visible satellite imagery, training AI models and achievimg beyond state-of-art results, applying multimodal deep learning by adding auxiliary info such as NIR satellite imagery, elevation, slope and soil characteristics. A key scientific contribution was the fact that we introduced a sleek methodology to augment training datasets for AI models by means of pseudo-labels and a weakly supervised training technique.
We officially released GAEA, a country-scale geospatial tool containing 27 environmental services, which relate to geoanalytics about Cyprus.
We launched the European PHENOWEX project, to enhance the scientific excellence and innovation capacity of two widening institutions from the Mediterranean Region, namely EGE (Türkiye) and CYENS (Cyprus), in the field of High-Throughput Phenotyping (HTP) with the assistance of top-class leading partners from across Europe.
We reached state-of-the-art accuracy in predicting and modeling large solar flares by means of space satellite imagery and AI, through the C-SpaRC project.
We have deployed 10 camera traps for insect monitoring around the island of Cyprus, via the Insect-AI COST action, choosing those locations strategically based on a novel landscape similarity technique we developed.
We proposed and published a methodology for strategic sensor placement for the identification of disinfection byproducts (DBPs) in chlorinated drinking water, using as a case study the water distribution network of Coimbra, Portugal, through the H2OFORALL project.
Our SPYCE camera trap, specially designed to monitor the behaviour of rodents, assessing the impact of various biological rodenticide on them, was deployed in the real world via the MED4PEST PRIMA project.
We demonstrated that we can train AI models based on incomplete information about insect classes in existing datasets (i.e. info can be at the level of Kingdom, Class, Order, Family, Genus, and/or Species), employing a class-weighted hierarchical loss function technique, which allows an AI model to understand the relationships between taxonomic levels while maintaining a high classification accuracy.
2023
We have developed an AI model for the scalable retrieval of similar landscapes based on optical satellite imagery using unsupervised representation learning, by breaking up the landscape similarity task into individual concepts closely related to remote sensing.
The FORBES magazine of the Phileleftheros local newspaper has created a reportage covering our projects related to environmental monitoring and modelling.
We have received funding and kick-started the InsectAI COST Action, supporting insect monitoring and conservation at the national and continental scale in order to understand and counteract widespread insect declines. The action received a 50 out of 50 evaluation score and our research group co-leads Working Group 3, which is about algorithms and techniques for insect monitoring.
We have developed GAEA, which is a country-scale geospatial environmental modelling tool that aspires to become a digital twin for the real estate market of Cyprus. The tool has been presented at the EnviroInfo 2023 conference held in Munich, Germany.
We have harnessed GAEA to develop a data story which visualizes Natura 2000 areas in relation to land-use change (constructions) and road networks in Cyprus. Our story emphasized the undeniable strong human presence inside Natura 2000 sites in Cyprus. In addition, the concentration of new constructions within close proximity to Natura 2000 areas poses an additional threat to the resilience of these natural regions.
We have proposed a new algorithm for honey bees' population estimation based on consecutive image frames, using deep learning and pose estimation techniques. The algorithm was presented at the International Conference on Intelligent Systems (IntelliSys), held in Amsterdam, the Netherlands.
We have published open datasets to allow researchers to monitor and count honeybees in bee hives from consecutive frames, as part of the BE-HIVE project.
We have developed a model-agnostic approach for generating Saliency Maps to explain inferred decisions of deep learning models, helping to understand how deep learning models take decisions in detection/classification problems when images are used as input.
We have developed smart beehives, installed in diferent rural and urban locations around Cyprus, having access to near real-time analytics about the populations of honeybees, their productivity and performance, as well as alerts about threats they might face, e.g. from wasps or varroa mite.
We have exploited Digital Surface Models for Inferring Super-Resolution for remotely sensed images, with better than state-of-the-art accuracy.
We have published the SuPerWorld Geo-API, which is an online API that consists of various geospatial and geo-analytics services which offer rich contextual information related to real estate properties, including environmental risks and risks related to climate change.
We have improved the operational efficiency of electric vehicle ridepooling fleets by predictive exploitation of idle times, exploiting these periods to harvest energy.
We have detected illegal dumping sites with high accuracy, by using high resolution aerial photography and the deep learning technique.
2021
We have developed and released Mindgrate, a mobile app which allows legal migrants, refugees and asylum seekers to better integrate themselves to the Cyprus society, assisting them to locate a job. Mindgrate addresses numerous barriers which hindered migrants during their integration to the country in the past.
We have published open-source code for agtech, allowing farmers and developers to use various popular path planning algorithms to find the best solution for a swarm of agents (drones and/or ground robots) to cover an agricultural field for a range of operations (e.g. pruning, spraying, monitoring, crop collection, etc.).
We have created a computer vision model, based on Deep Siamese Neural Networks, for detecting land use change from satellite imagery, using a weakly supervised learning technique, for accelerating the training process without significant effort in data annotation/labelling.
We have identified, listed and visualized 63 important key performance indicators used by 193 countries around the world to measure climate change. To define goals for cities and countries in regards to mitigating climatic change, we first need to understand which the important KPIs are, how they can be measured and which values they take. Then, each country can calculate its performance based on these KPIs, setting realistic goals for better performance in the near future.
We have trained a computer vision model to take advantage of the shadow map of a remotely sensed monocular image to calculate its heightmap with high precision, creating digital surface models without the need of expensive equipment and large costs.
CovTracer-EN, which constitutes the official national mobile phone application for COVID-19 contact tracing in Cyprus based on the GAEN API, developed by our group in a joint effort with other MRG groups at CYENS, has been recently launched around the country, counting 25,000 users after one month from official release.
We have published open-source code for fire authorities around the world, including mobile apps for citizens and web apps for admin personnel, which can be used to model the propagation of wildfires in real-time, assisting users to safely evacuate the area using our mobile apps.
We have examined whether animal manure constitutes an effective strategy to increase soil organic carbon stocks in the Mediterranean as a mitigation climate change action.
We have proposed an effective solution to the contamination of soils and water due to animal manure, by suggesting to transfer this manure from livestock farms to crop fields, to be used as fertilizer. We used Catalonia, Spain as a case study, employing a nature-inspired algorithm based on ants' foraging behaviour to solve the manure distribution problem.
We have developed a contact tracing app to fight the COVID-19 pandemic, which became the official app used by the Government of Cyprus.
We have considered various algorithms to investigate how swarms of ground robots and/or unmanned aerial vehicles (drones) could collaborate together for solving various tasks in agricultural fields efficiently.
We have predicted parking occupancy in short-term (i.e. next 60 minutes) and in real-time at the central parking station of Arnhem, the Netherlands. Our results have beaten the state-of-art predictions for parking occupancy.
We have proposed the use of synthetic data for training deep learning models, in cases where real-world datasets are inexistent or difficult to prepare/create. We have applied this concept in aerial photography for identifying disasters (i.e. fire, smoke, collapsed buildings) and to count houses and buildings.