METHOD OF COUNTING OBJECTS IN AN ORCHARD IN REAL TIME

Abstract

The paper proposes a novel method for counting apples in an orchard using unmanned aerial vehicles (UAVs), also known as drones. The authors acknowledge that traditional methods of counting apples are time-consuming and labor-intensive and therefore propose a more efficient and accurate solution using a group of drones equipped with high-resolution cameras. The proposed method involves using a convolutional neural network (CNN) called YOLOv5-v1 to detect and count the apples in the images captured by the drones. YOLOv5-v1 was trained using a dataset of annotated images of apples in different lighting conditions and orientations and was found to be highly accurate in identifying and counting apples in test images. The proposed method was tested on two different orchards and was highly accurate. Specifically, the results of the experimental studies show that the model accuracy was: i) in sunny weather – 92.11%, ii) in cloudy weather – 90.76%, and iii) in sunny weather, but with high shading – 82.69%. Such results indicate that under sunny and overcast weather conditions with a low level of shade, the proposed approach demonstrates high accuracy and reliability in real-time operation. At the same time, it should be noted that due to the presence of significant visual noise in the orchard, such as the covering of fruits by leaves and branches, the efficiency of the UAV group and the automated system, in general, cannot be 100% fulfilled in natural conditions, which can serve as a promising task for further of research. Overall, the proposed method of counting structural objects provides an efficient and accurate solution for counting apples in orchards, which could save time and resources for farmers. The use of UAVs and CNNs in agriculture is a promising area of research, and this paper presents a practical application of these technologies in fruit counting.