Year: 2024
Type: conference-paper
Abstract: To build an in-field fruit harvesting robot, it is important to locate the fruit efficiently and accurately. Instance segmentation is incorporated to evaluate peach ripeness, which enables precise identification of the ripeness level for each peach instance, allowing robots to selectively harvest ripe peaches, thereby maximizing harvesting efficiency. We have proposed a peach-specific instance segmentation model comprising three components: a ResNet50 backbone, a Feature Pyramid Network (FPN), and a Transformer decoder. Demonstrating a mean average precision (mAP) of 66.401, our model outperforms other state-of-the-art models in accurately segmenting peach instances. Notably, it achieves impressive AP of 64.818 for unripe peaches, 62.640 for semi-ripe ones, and 71.745 for ripe peaches, highlighting its effectiveness across varying ripeness levels. The model maintains the rapidest inference time of 91 ms per iteration. This comprehensive summary underscores the model's efficacy in peach instance segmentation, promising significant advancements in automated fruit harvesting and agricultural productivity.
DOI: 10.18573/conf3.i
Institution: Cardiff University
Language: English
Topic: Smart Agriculture and AI
Subfield: Plant Science
Domain: Life Sciences
SDG: Zero hunger
Open Access status: gold
OpenAlex ID: W4404502491