Enhanced Real-Time Target Detection for Picking Robots Using Lightweight CenterNet in Complex Orchard Environments DOI Creative Commons
Pan Fan, Chusan Zheng,

Jin Sun

et al.

Agriculture, Journal Year: 2024, Volume and Issue: 14(7), P. 1059 - 1059

Published: June 30, 2024

The rapid development of artificial intelligence and remote sensing technologies is indispensable for modern agriculture. In orchard environments, challenges such as varying light conditions shading complicate the tasks intelligent picking robots. To enhance recognition accuracy efficiency apple-picking robots, this study aimed to achieve high detection in complex environments while reducing model computation time consumption. This utilized CenterNet neural network framework, introducing gray-centered RGB color space vertical decomposition maps employing grouped convolutions depth-separable design a lightweight feature extraction network, Light-Weight Net, comprising eight bottleneck structures. Based on results, 3D coordinates point were determined within camera coordinate system by using transformation relationship between image’s physical system, along with depth map distance information map. Experimental results obtained testbed an orchard-picking robot indicated that proposed achieved average precision (AP) 96.80% test set, real-time performance 18.91 frames per second (FPS) size only 17.56 MB. addition, root-mean-square error positioning was 4.405 mm, satisfying high-precision requirements vision environments.

Language: Английский

Enhanced Real-Time Target Detection for Picking Robots Using Lightweight CenterNet in Complex Orchard Environments DOI Creative Commons
Pan Fan, Chusan Zheng,

Jin Sun

et al.

Agriculture, Journal Year: 2024, Volume and Issue: 14(7), P. 1059 - 1059

Published: June 30, 2024

The rapid development of artificial intelligence and remote sensing technologies is indispensable for modern agriculture. In orchard environments, challenges such as varying light conditions shading complicate the tasks intelligent picking robots. To enhance recognition accuracy efficiency apple-picking robots, this study aimed to achieve high detection in complex environments while reducing model computation time consumption. This utilized CenterNet neural network framework, introducing gray-centered RGB color space vertical decomposition maps employing grouped convolutions depth-separable design a lightweight feature extraction network, Light-Weight Net, comprising eight bottleneck structures. Based on results, 3D coordinates point were determined within camera coordinate system by using transformation relationship between image’s physical system, along with depth map distance information map. Experimental results obtained testbed an orchard-picking robot indicated that proposed achieved average precision (AP) 96.80% test set, real-time performance 18.91 frames per second (FPS) size only 17.56 MB. addition, root-mean-square error positioning was 4.405 mm, satisfying high-precision requirements vision environments.

Language: Английский

Citations

1