Research Projects

POLENET: Automatic pollen analysis using convolutional neural networks: application to the monofloral classification of honey

Start Date:     


End Date:    


Financing entity:  

Ministerio de Asuntos Económicos y Transformación Digital, MINECO.



Other entities participants: Asensio-Grau, Andrea; Escriche Roberto, Mª Isabel; Doménech Antich, Eva Mª

About the project

Quickly classify honey varieties, map pollen-related allergies, or even pinpoint the location of a crime. These are some of the applications of the artificial intelligence system for pollen counting that are being developed by researchers in the Computer Vision Area of ​​the ai2 Institute, in collaboration with the staff of the UPV’s Institute of Food Engineering for Development.

The project “POLENET, Automatic pollen analysis using convolutional neural networks: application to the monofloral classification of honey”, arose a year ago from the need of the Honey Laboratory of the Institute of Food Engineering for Development of the UPV to automate its honey qualification work. José Miguel Valiente, ai2 researcher responsible for the project, explains that the laboratory currently carries out melissopalynology work: “it receives samples from producers from all over Spain and classifies it to find out if the honey is single-flowered or multi-flowered, since on the market one is better valued than the other. The usual procedure to carry out this work is for an expert in pollen identification, working under a microscope, to count the pollen particles contained in each variety of honey”.

Neural networks

The system in which the researchers at the ai2 Institute are now working to automate this process would reduce the workload involved in this task by hours. Until now, the application of computer vision to natural products, which have great variability, involved certain problems that have been solved by the use of artificial intelligence through neural networks and the technique of deep learning. “Thanks to these techniques, the network learns and then infers,” says Valiente. However, until that is achieved, thousands of images of pollens are necessary, which must previously be classified by the staff of the Institute of Food Technology.

Currently, the project, which will run until 2022, is in that initial stage of capturing information. “We have automated the microscope to scan samples in a fixed way and take hundreds and hundreds of images; then, we pass these images through an application that we have developed and the expert labels them, so that we have the material to train the neural network”, comments Valiente.

The ultimate goal is to have an application that helps laboratory technicians to identify pollens, standardizing classification criteria, thus being able to carry out the analysis on many more samples and, therefore, obtaining faster and more objective results.

Allergies, forensic and fossil analysis

The system would not only reduce this work in hours, but would have applications beyond pollen counting for the honey production market, since the same technique is used in aerobiology and aeropalinology, sciences related to pollen counting to develop models of prediction that allow knowing the beginning and the pollen content of a certain place, thus alerting the population with possible allergies.

“There are other contexts, such as forensic palynology, where it could also be used, since the technique used for pollen counting is the same. In certain forensic studies, pollens are analyzed to try to deduce the specific geographical location in which an event occurred, for example”, explains Valiente. “The same technique is used to study certain fossils,” adds the expert.

The POLENET project is funded by the Ministry of Economic Affairs and Digital Transformation, MINECO.

In addition to Valiente, the researcher from the Institute of Food Technology María Isabel Escriche, as well as Eva María Domenech, Manuel Agustí, Vicente Luis Atienza, Fernando López and Mario Visquert participate in the project.