Edge Vision against Varroa (EV2)

Informazioni sul progetto

• Nome progetto: EdgeVision against Varroa (EV2): Edge computing in defence of bees

• Soggetti finanziatori: MUR
  Ammesso al finanziamento con decreto Bando PRIN 2022, Decreto Direttoriale n. 104 del 02-02-2022.
  https://www.mur.gov.it/it/atti-e-normativa/decreti-di-ammissione-al-finanziamento-bando-prin-2022-decreto-direttoriale-n-104 

• Costo Ammesso: € 253.736,00 (cofinanziamento € 54.454,00)

• Durata: 24 mesi (28/09/2023-28/09/2025)

• Codice progetto: 202277WMAE

• Settore ERC: PE6

• CUP Master: B53D2301282 0006

  • Project Investigator (PI) / Research Unit 1: Roberto Beraldi (Dipartimento di Ingegneria Informatica, Automatica e Gestionale "Antonio Ruberti" - Sapienza Università di Roma), beraldi@diag.uniroma1.it. CUP B53D2301282 0006.

  • Research Unit 2: Ignazio Floris (Dipartimento di Agraria, Università degli Studi di Sassari), ifloris@uniss.it. CUP: J53D23007010 006.

  • Research Unit 3: Floriano De Rango (Dipartimento di Ingegneria Informatica, Modellistica, Elettronica e Sistemistica - Università della Calabria), derango@dimes.unical.it. CUP: H53D2300345 0006.

• Status: In essere

Descrizione del progetto

The Varroa destructor ectoparasite mite is the most serious threat to bees in Italy and the world [30,31]. Mortality rates of honey bees vary from 5% to almost 30% depending on the country and year, being the Varroa mite some of the main causes [41].

This pathology must be continuously contained on levels of infestation compatible with the survival of bee colonies through periodic drug treatments, on pain of the death of families in one or two seasons of bees. Current diagnostic techniques for this parasite are based on manual visual inspection of some characteristic regions in the body of bees or other time-consuming laboratory methods. Late diagnosis of this ectoparasite causes several harmful physical, physiological, and pathological effects on bees at the individual and colony levels.

The project aims at the design and prototyping of a visual detection system of the varroa mite, which exploits energy-aware resource management and application algorithms based on the cloud-to-IoT continuum paradigm, tailored and extended to run on cooperating energy-constrained devices. This innovative automatic detection system of the level of infestation uses visual deep learning algorithms that dramatically reduce the state-of-the-art detection time of the varroa infestation level.

Research Objectives and KPI

• RO1 Identification of new infestation markers suitable to design visual recognition algorithms based on deep learning (T2.1,T2.2).

• RO2 Generation of a labelled dataset of varroa-infested bees (T3.1,T3.2,T3.3,T3.4).

• RO3 Design of energy-aware management software for a network of edge devices (T4.1, T4.2)

• RO4 Design of energy-efficient deep learning algorithms for detecting infestation markers using the dataset (T4.3,T4.4)

• RO5 Experimental analysis and validation of the relationship between the true level of infestation and the markers (T2.3, T2.4,T5.1, T5.2,T5.3)

• DO1 Publish research results in top-tier scientific journals and conferences (T5.3)

• DO2 Dissemination activities through seminars, trade associations, and media (T5.3)

• K1a: Number of static detectable Varroa infestation markers in photographic shots ≥ 2

• K1b: Number of dynamic detectable Varroa infestation markers in video recording >= 2

• K2a: Number of labelled static markers ≥ 5000

• K2b: Number of labelled dynamic markers ≥ 5000

• K3: Number of manageable smart hives ≥ 50

• K4a: DL processing capacity on TPU-equipped devices ≥ 256x256 pixels

• K4b: Energy consumption required by image processing ≤ 3J/frame (compatible with an installable solar panel on hives)

• K4c: mAP of DL algorithms ≥ 65%

• K5: Number of validating experiments ≥ 1

• K6 (DO1): Number of papers in top-tier scientific journals and conferences ≥ 7

• K7a (DO2): Number of dissemination materials through social channels ≥ 2

• K7b (DO2): Number of dissemination events through trade associations ≥ 1

• K7c (DO2): Number of scientific seminars ≥ 1

Deliverables 

• D2.1 Tech report on the state of the art (28/03/2024) [Link]

• D2.2 Smart beehive and system architecture (28/05/2024) [Link]

• D3.1 Raw Dataset containing shots and videos from T3.1, T3.2, T3.3 (28/03/2024) [10.5281/zenodo.10875856]

• D3.2 Labelled Dataset (26/09/2024) [10.5281/zenodo.13771384]

• D4.1 Edge resource management algorithms (25/02/2025)

• D4.2 Edge orchestration and cloud integration algorithms (25/02/2025)

• D4.3 Visual algorithms for Varroa detection (25/02/2025)

• D4.4 Energy system model (25/02/2025)

• D5.1 System release documentation (28/05/2025)

• D5.2 Experiment results (26/09/2025)