Deep Neural Network Based Real-time Kiwi Fruit Flower Detection in an\n Orchard Environment

Abstract

In this paper, we present a novel approach to kiwi fruit flower detection\nusing Deep Neural Networks (DNNs) to build an accurate, fast, and robust\nautonomous pollination robot system. Recent work in deep neural networks has\nshown outstanding performance on object detection tasks in many areas. Inspired\nthis, we aim for exploiting DNNs for kiwi fruit flower detection and present\nintensive experiments and their analysis on two state-of-the-art object\ndetectors; Faster R-CNN and Single Shot Detector (SSD) Net, and feature\nextractors; Inception Net V2 and NAS Net with real-world orchard datasets. We\nalso compare those approaches to find an optimal model which is suitable for a\nreal-time agricultural pollination robot system in terms of accuracy and\nprocessing speed. We perform experiments with dataset collected from different\nseasons and locations (spatio-temporal consistency) in order to demonstrate the\nperformance of the generalized model. The proposed system demonstrates\npromising results of 0.919, 0.874, and 0.889 for precision, recall, and\nF1-score respectively on our real-world dataset, and the performance satisfies\nthe requirement for deploying the system onto an autonomous pollination\nrobotics system.\n