DeepFruits: A Fruit Detection System Using Deep Neural Networks

• Published in: Sensors (Basel, Switzerland) Vol. 16; no. 8; p. 1222
• Main Authors: Sa, Inkyu, Ge, Zongyuan, Dayoub, Feras, Upcroft, Ben, Perez, Tristan, McCool, Chris
• Format: Journal Article
• Language: English
• Published: Switzerland MDPI AG 03.08.2016; MDPI
• Subjects: agricultural robotics; Algorithms; Annotations; Automation; Capsicum; Construction; deep convolutional neural network; Detectors; Fruit; Fruits; Harvesting; harvesting robots; horticulture; Humans; Image detection; Image Processing, Computer-Assisted - methods; multi-modal; Neural networks; Neural Networks (Computer); Pattern Recognition, Automated - methods; rapid training; real-time performance; Recall; Robotics; State of the art; visual fruit detection; real-time performance; harvesting robots; horticulture; rapid training; deep convolutional neural network; agricultural robotics; multi-modal; visual fruit detection
• ISSN: 1424-8220; 1424-8220
• DOI: 10.3390/s16081222

This paper presents a novel approach to fruit detection using deep convolutional neural networks. The aim is to build an accurate, fast and reliable fruit detection system, which is a vital element of an autonomous agricultural robotic platform; it is a key element for fruit yield estimation and automated harvesting. Recent work in deep neural networks has led to the development of a state-of-the-art object detector termed Faster Region-based CNN (Faster R-CNN). We adapt this model, through transfer learning, for the task of fruit detection using imagery obtained from two modalities: colour (RGB) and Near-Infrared (NIR). Early and late fusion methods are explored for combining the multi-modal (RGB and NIR) information. This leads to a novel multi-modal Faster R-CNN model, which achieves state-of-the-art results compared to prior work with the F1 score, which takes into account both precision and recall performances improving from 0 . 807 to 0 . 838 for the detection of sweet pepper. In addition to improved accuracy, this approach is also much quicker to deploy for new fruits, as it requires bounding box annotation rather than pixel-level annotation (annotating bounding boxes is approximately an order of magnitude quicker to perform). The model is retrained to perform the detection of seven fruits, with the entire process taking four hours to annotate and train the new model per fruit.