George Arvanitakis

Recent advancements in computer vision, particularly in detection, segmentation, and classification, have significantly impacted various domains. However, these advancements are still strongly tied to RGB-based systems, which are insufficient for applications in industries such as waste sorting, pharmaceuticals, and defence, where material characterization beyond shape or visible colour is necessary. Hyperspectral (HS) imaging captures spatial and spectral information for each pixel and therefore offers a promising route for material-level classification. This study evaluates the potential of combining HS imaging with deep learning for plastic material classification. The work includes: (i) the design of an experimental setup with a HS line-scan camera, conveyor, and controlled illumination; (ii) the construction of an object-disjoint dataset of HDPE, PET, PP, and PS samples with semi-automated mask generation and Raman spectroscopy-based labelling; and (iii) the development of P1CH, a lightweight pixel-wise 1D convolutional hyperspectral classifier. On object-disjoint test images, P1CH achieved 97.44% all-pixel accuracy. A boundary sensitivity analysis, reported separately because semi-automated labels are uncertain at material/background interfaces, yielded 99.94% accuracy after excluding a pre-defined two-pixel border band. Additional ablation, baseline, and robustness analyses show that the proposed pixel-wise spectral approach is effective for small fragments, visually similar plastics, and overlapping materials, while black or very dark plastics remain challenging under the present camera and illumination configuration.

Mosquito-borne diseases have been spreading across Europe over the past two decades, with climate change contributing to this spread. Temperature and precipitation are key factors in a mosquito’s life cycle, and are greatly affected by climate change. Using a machine learning framework, Earth Observation data, and future climate projections of temperature and precipitation, this work studies three different cases (Veneto region in Italy, Upper Rhine Valley in Germany and Pancevo, Serbia) and focuses on (i) evaluating the impact of climate factors on mosquito abundance and (ii) long-term forecasting of mosquito abundance based on EURO-CORDEX future climate projections under different Representative Concentration Pathways (RCPs) scenarios. The study shows that increases in precipitation and temperature are directly linked to increased mosquito abundance, with temperature being the main driving factor. Additionally, as the climatic conditions become more extreme, meaning higher variance, the mosquito abundance increases. Moreover, we show that in the upcoming decades mosquito abundance is expected to increase. In the worst-case scenario (RCP8.5) Serbia will face a 10% increase, Italy around a 40% increase, and Germany will reach almost a 200% increase by 2100, relative to the decade 2010–2020. However, in terms of absolute numbers both in Italy and Germany, the expected increase is similar. An interesting finding is that either strong (RCP2.6) or moderate mitigation actions (RCP4.5) against greenhouse gas concentration lead to similar levels of future mosquito abundance, as opposed to no mitigation action at all (RCP8.5), which is projected to lead to high mosquito abundance for all cases studied.

An Early Warning System for mosquito abundance is a valuable tool that can alert authorities for potential outbreaks of mosquito populations in a given area for the upcoming period. This information is used to take mitigation actions in order to avoid spread of vector borne diseases such as West-Nile Virus, Malaria, Zika etc. A promising direction of those systems today aims to predict the upcoming mosquito population by following a data driven approach and taking advantage of machine learning (ML) algorithms. The ML algorithms are trained on a limited set of point level data that include the environmental, geomorphological, climatic information and historical in-situ measurements of mosquito population for specific latitude and longitude coordinates. Goal of the ML algorithms is to learn the patents that connect the characteristics (features) of a given area (temperature, humidity, NDVI, rainfall, latitude, longitude, etc) with the upcoming mosquito population. Once the in-situ entomological data are expensive to be collected and limited, one of the key challenge of the aforementioned approach is to understand where those models can generalize with an acceptable accuracy in order to be re-used in areas that prior entomological information do not exist or in other words to understand the area of applicability of those models. In this study we analyze the performance of ML algorithms that have been trained in specific areas and applied to “unseen” areas. Our analysis aims to understand the characteristics of the cases where the algorithms manage to generalize compared with the ones where the performance is poor. Our scope is to establish a systematic approach for determining the area of applicability of the models, thus, to obtain a prior knowledge regarding the areas that we expect models to generalize properly and the areas the predictions of the models are not trustworthy. Our work relied on historical data of Culex pipiens mosquitoes (West Nile virus) collected in the Veneto region of Italy for the decade 2011-2021 and satellite Earth Observation data. For ML regressor we used a feedforward Neural Network with typical mean square error cost function. Initially we conclude that the typical euclidian distance between the coordinates of the trained area and the unseen data is not an informative metric about the model’s area of applicability. Instead, we propose a metric that calculates the distance between the known and the unknown points in the feature space (environmental, geomorphological etc.) and also takes into account the feature importance of trained Neural Network using the SHAP values. The results showed that our proposed metric is informative regarding where the model is expected to have more accurate predictions and manage to capture the cases where the generalization will be poor. This information is useful both to judge if the predictions of a model are trustworthy and also to understand for which areas our prior information is not sufficient and to take actions in future network planning.

Mosquito-borne diseases (MBDs) have been spreading across many countries including Europe over the past two decades, causing thousands of deaths annually. They are transmitted through the bites of infectious mosquitoes. Environmental, meteorological and other spatio-temporally variables affect the mosquito abundance (MA), and thus affect the circulation of the MBDs in the community. So an early warning system of MA based on these parameters could serve as a warning for the upcoming MBDs incidence. We propose Deep-MAMOTH, a data driven, generic and accurate early warning system for predicting MA in the upcoming period, based on earth observational (EO) environmental data and optionally in-situ entomological data. Deep-MAMOTH can be easily replicated and applied to multiple areas of interest without any special parametrization.The Deep-MAMOTH pipeline collects EO information from various data sources (temperature, rainfall, vegetation, distance from coast, elevation, etc.) and in-situ entomological data for each area of interest. Then, there is a feature extraction phase that combines the previous collected information to more complex features, and finally this data is fed into a Deep Neural Network responsible to capture the relationship between the above mentioned features and the MA, delivering a MA risk class ordered from 0 to 9 for the upcoming period (e.g. 15 days). The pipeline provides a standardized way to predict MA without depending on the area of interest or the mosquito genus and can be modified to predict the actual MA instead of a risk class. However, risk classes help to better propagate the severity of the situation.Two versions of Deep-MAMOTH were implemented, the first one is using recently collected entomological information in order to produce predictions (i.e. mosquitoes collected 1 week ago). The other version works when there is no recently collected entomological information for the area of interest. The latter version is expected to perform worse than the first one, but gives us the capability to produce predictions anywhere on earth without the need of recently collected entomological data. We applied Deep-MAMOTH in Veneto (Italy), in Upper Rhine region (Germany), and the Vojvodina region (Serbia) regarding the Culex spp. genus mosquito. The results are promising as Deep-MAMOTH in Italy achieves a mean absolute error (MAE) of 1.27 classes with the percentage of predictions that deviate at most 3 classes (e3) from the actual one reaching up to 95%. In Serbia MAE is 1.77 classes, with e3 equal to 88% and finally for Germany MAE is 0.92 classes and e3 equal to 94%.  It’s worth mentioning that prediction performance in the version of Deep-MAMOTH without using entomological information remains promising. MAE in Italy was increased only by 0.02 and in Germany by 0.1, with e3 remaining at the same level in both cases, while in Serbia MAE increased by 0.2 with e3 decreasing by 8%. We conclude that the prediction of MA from EO data can be accurate with or without recently collected entomological data.Acknowledgment:This research has been co-financed by the ERD Fund of the EU and Greek national funds through the Operational Program Competitiveness, Entrepreneurship and Innovation, RESEARCH-CREATE-INNOVATE(project code:T2EDK-02070)