On-tree fruit detection system using Darknet-19 based SSD network

An accurate and reliable on-tree fruit detection system is crucial for automatic yield estimation, mapping, and harvesting. This paper presents a deployment of the state-of-the-art deep object detection model, the Single Shot Detector (SSD) for detecting mango fruits on a tree canopy in an open mango orchard. The proposed approach utilizes the deep learning model Darknet-19 as a feature extractor and SSD for detection and localization in the object detection framework. The network’s training, validation, and testing are performed on the publicly available annotated dataset ‘Mango YOLO’. The data augmentation applied to training images consists of the random and central extraction of image patches and corresponding horizontal reflection. Experimental results show that the proposed network achieved an average precision of 92.43% in detecting on-tree mango fruits in the test dataset.