Automated procedure for early prediction of apple yield in orchardse | Automatizirani postupak ranog predviđanja prinosa jabuka u voćnjacima

Inglés. Procedure for early forecasting of apple yields in Slovenian orchards is presented with the help of a modern application that works in the Android environment and uses a fast 5G transmission network. To recognize fruits in images of ‘Golden delicious’ trees, we used the YOLO pre-learning network, which is based on convolutional neural networks and regression techniques to determine the position of apples in the image. To model the fruit yield, specific cultivar-adjusted growth curve is used, which makes it possible to predict fruit mass from the current and expected fruit diameter and the ratio between diameter and mass. The procedure was tested on a series of 20 images captured from six-years-old orchard revealing on average 71.84% accuracy in fruits counting, 108.27% accuracy in diameter calculating and 97.90% accuracy in yield forecasting. With the first estimation, we have shown that the automated method for yield forecasting is an excellent tool for accurately estimating the yield of an individual plot so we will continue to upgrade it in the future.

Croata. Prikazan je postupak ranog predviđanja prinosa jabuka u slovenskim voćnjacima uz pomoć moderne aplikacije koja radi u Android okruženju i koristi brzu 5G prijenosnu mrežu. Za prepoznavanje plodova na slikama sorte ‚Golden delicious‘ koristili smo YOLO mrežu prethodnog učenja koja se temelji na konvolucijskim neuronskim mrežama i regresijskim tehnikama za određivanje položaja jabuka na slici. Za modeliranje prinosa ploda koristi se specifična krivulja rasta prilagođena sorti. koja omogućuje predviđanje mase ploda iz trenutnog i očekivanog promjera ploda te omjera promjera i mase. Postupak je testiran na nizu slika snimljenih u šest godina starom voćnjaku ‚Golden delicious‘ otkrivajući 71,64 % točnost u brojanju plodova, 91,73 % točnost u izračunavanju promjera i 72,65 % točnost u predviđanju prinosa. Već prvim testovima pokazali smo da je automatizirana metoda za prognozu prinosa izvrstan alat za točnu procjenu prinosa pojedine parcele pa ćemo je i u budućnosti nadograđivati.