TY  - JOUR
T1  - Implementation of a Fruit Stem Detection System Using Anchor-Free-Based 3D Object Detection
AU  - Lee, Yeong-wook 
AU  - Noh, Dong-hee 
AU  - Lee, Hea-min 
JO  - The Journal of Korean Institute of Communications and Information Sciences
PY  - 2024
DA  - 2024/1/1
DO  - 10.7840/kics.2024.49.2.301
KW  - Deep learning
KW  - Digital agriculture
KW  - 3D Object detection
KW  - Fruit stem detection
AB  - This paper proposes a fruit stem detection system using Anchor-Free-based 3D object detection model for fruit stem removal. The fruit stem is an important part that affects the hygiene, food safety, quality, and freshness of fruits. It can save manpower and time by automating existing manual-dependent tap removal operations, and increase efficiency in related agricultural fields. The FCAF3D model is a 3D object detection algorithm that predicts the skin and stem of the fruit, respectively, showing high detection performance even in the small size of the fruit stem. The network structure of the model consists of ResNet, GSDN, and FCOS networks, which handle various object scales through the FPN structure. In this paper, model training was performed using apples as an example, and the learned model showed high accuracy in the apple dataset, and the bounding box coordinate value of the test results can be used in the fruit stem removal system.
Experimental results showed that the FCAF3D model showed high performance in detecting fruit stem.