Using Arti­fi­cial Intel­li­gence in the field

Jadwiga and Marek Czerwczak’s 300-ha organic farm – Farma na górce – in Kiełpino, 10km south of Gorzów Wielkopolski, Poland, has imple­mented an AI-powered moni­toring system – Smart­Mon­i­toring – to improve herd obser­va­tions. With a herd of 80 Simmental cows for dual use purposes grazed in blocks over an exten­sive 180ha of peat meadows and pasture, the AI tech­nology was intended to help improve labour and obser­va­tion effi­cien­cies by replacing the phys­ical task with tech­nology.

AI recog­nizes cows using cameras

So how has it worked? Devel­oped by tech­nology company TaxusIT and the Insti­tute of Manage­ment at Warsaw Univer­sity of Life Sciences (funded by the EU Rural Devel­op­ment programme), the system uses infrared cameras and convo­lu­tional neural networks (CNNs); complex computer models, to auto­mat­i­cally recog­nise cows and improve herd obser­va­tions. “Neural networks are complex computer models inspired by the struc­ture and oper­a­tion of the human brain, consisting of inter­con­nected layers of neurons that process data,” says project manager Sławomir Łoś “They enable auto­matic learning of patterns from data to perform tasks like recog­ni­tion.

By analysing the streaming from the cameras using dedi­cated soft­ware, it was possible to track the herd in the pasture.

Sławomir Łoś

“Convo­lu­tional networks were chosen because of their unique ability to analyse and iden­tify images by detecting patterns like shapes and textures, which is crucial for recog­nising indi­vidual animals.” The farm used no other moni­toring devices when assessing the technology’s capa­bil­i­ties to avoid addi­tional factors influ­encing their behav­iour, making it easier to iden­tify indi­vidual activity patterns. “Masts with long-range PTZ cameras and neural networks were used and by analysing the streaming from the cameras using dedi­cated soft­ware, it was possible to track the herd in the pasture,” adds Sławomir.

For a better pasture manage­ment

The project had mixed results, and while the AI moni­toring system failed to iden­tify indi­vidual cattle, it did succeed in deliv­ering other highly bene­fi­cial obser­va­tions. Detecting heat, calving, and unusual behav­iours like cows isolated from the herd, the system has the poten­tial to not only improve fertility and calving proto­cols at grass, but also to help iden­tify sick cows, as well as possible human and predator (wolves) inter­fer­ence.

With organic status there is little room for the farm to make pasture amend­ments, like reseeding and fertiliser appli­ca­tion, which has made managing the exten­sive and non-uniform grazing plat­form a chal­lenge. Inter­est­ingly, the tech­nology can be used to help the farm over­come this chal­lenge by deter­mining when vege­ta­tion has dimin­ished and cows need to be moved to fresh pasture, all based on cow behav­iour.

Jadwiga and Marek plan to continue the project and are consid­ering synchro­nising it with other systems they operate inside the cow sheds. They would also like to intro­duce robotic milking, which would be a major invest­ment for the farm, as part of a fully auto­mated system. They have already had prelim­i­nary discus­sions with a company that offers this type of solu­tion and are also looking to involve milk quality special­ists in the project to assess how the switch would affect overall milk produc­tion and quality.