Estimation of apricot harvest in Malatya with computer vision techniques

In our country the products such as apricot, hazelnut, grape and fig has an important production potential all over the world. For example, approximately 74% of the world's dried apricot production is made in our country, and 85% of this production takes place only in the province of Malatya. It is very important in many respects that these products we exported can be estimated correctly before harvesting the annual production quantities (harvest). Accurate forecasting of revenues will result in the elimination of manipulations on the product market, the creation of a supply-demand balance and the associated logistical support to be carried out in a healthy manner. However, in the Apricot Research Report made by Euphrates Development Agency, the main factors threatening the apricot production are the uncertainty of price and the fact that the yield estimations cannot be done with the desired accuracy. Existing methods of estimating yields include examining the trees in the apricot gardens in the specific area by the observer, estimating the total amount of apricots, and finally making general estimations. Misconceptions arising from the human factor are unacceptably different between the yield estimates made. This negative situation applies to other agricultural products other than apricots. For example, there is a difference of about 25% (183 thousand tons) among the hazelnut harvest estimates made by five different institutions in 2014. In this thesis study, three different segmentation methods were used for apricot tree segmentation, namely Matting, edge-based and zone-based methods, and the method with the best performance was selected. In this study apricot tree images were tested on an apricot garden. The estimated results obtained about 90% success rate. This thesis study was supported by TUBITAK as a project. In addition, one of the broadcast IDAPs was presented at the international conference.