Portuguese scientists test new intelligent vineyard monitoring systems

The study was published in the scientific journal Computers and Electronics in Agriculture.

A multidisciplinary team, led by researchers from the Institute of Systems and Robotics (ISR) of the Faculty of Sciences and Technology of the University of Coimbra (FCTUC), explored new technological approaches to vineyard management, opening doors to the development of non-invasive monitoring systems and efficient that allow immediate and localized action in case of diseases and pests, improving production and reducing the harmful impact on the environment.

The study, which had the participation of researchers from the Institute of Systems and Computer Engineering of Coimbra (INESC Coimbra) and the Escola Superior Agrária de Coimbra (ESAC), was carried out as part of the Al+Green project: Intelligent Automation in Precision Agriculture. , funded by MIT-Portugal and the Fundação para a Ciência e a Tecnologia (FCT), which aims to improve the accuracy and reliability of monitoring and detection of pests and diseases in vineyards.

For 12 months, three vineyards in the Centro region – Coimbra, Valdoeiro and Quinta de Baixo – managed according to conventional practices, but with different biophysical characteristics, were studied.

The approaches explored and tested by the scientists were based on Deep Learning systems (deep learning, artificial intelligence), using spatio-temporal information obtained through remote sensing (satellite) and drones.

«This work studied the most appropriate spectral bands and segmentation techniques for the identification of vineyard lines in aerial images (eg, captured by drones). It is important to differentiate pixels belonging to vines from pixels belonging to other elements (for example, vegetation between rows), to avoid data contamination», says researcher Tiago Barros.

“By avoiding pixels that do not belong to the vines, more reliable estimates are obtained in tasks such as harvest estimation or plant vigor assessment. To this end, we equipped a drone with a multispectral camera and a high definition RGB camera, which were used to collect spectral information from three vineyards in the Centro region», he explains.

The results of the study indicate, according to the ISR researcher, that segmentation models «based on Deep Learning perform better when compared to classical methods. Regarding the spectral bands, the Near-Infrared band is the band that contributes to the best performance».

In other words, concludes Tiago Barros, the study presents good arguments for the use of this type of dual-chamber approach for data acquisition, contributing to the advancement of precision agriculture, because «promoting more efficient agriculture is essential to improve the quality and food security without compromising environmental sustainability. This sector, although it has benefited, in a modest way, from technological advances in other sectors, such as industry, robotics, intelligent vehicles, etc., remains a predominantly manual and inefficient sector. Precision agriculture promotes the use of technology (software and hardware) in applications such as agricultural protection, monitoring and management».

The study was published in the scientific journal Computers and Electronics in Agriculture. The scientific article, entitled “Multispectral vineyard segmentation: A deep learning comparison study”, is available at: https://doi.org/10.1016/j.compag.2022.106782.

 

 



Comments

Ads