LightIndust-Net: A Lightweight Object Detection and Instance Segmentation Algorithm for Robotic Arm Assembly in Industrial Scenarios

TY - GEN

T1 - LightIndust-Net

T2 - Chinese Intelligent Automation Conference, CIAC 2025

AU - Xiao, Weiye

AU - Hu, Lingxi

AU - Li, Zhiwei

AU - Wang, Li

AU - Zhang, Wei

AU - Jiang, Linhua

AU - Long, Wei

N1 - Publisher Copyright: © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2026.

PY - 2026

Y1 - 2026

N2 - This research proposes LightIndust-Net, a lightweight object detection and instance segmentation algorithm specifically designed for robotic arm operations in industrial assembly scenarios. The algorithm effectively addresses the challenge of balancing accuracy and computational efficiency in current robotic operation systems through an innovative StarNet backbone network, Cross-Stage Star Bottleneck (CSSB) modules, and a Shared Convolutional Separator Segmentation Head (Seg_SCSS). Experimental evaluation shows that LightIndust-Net achieves a 22.1% improvement in computational efficiency and a 37.7% reduction in model parameters relative to the baseline approach, while maintaining or even improving detection and instance segmentation accuracy, particularly excelling in processing complex industrial component edge details and shape features. This efficient and precise visual perception solution provides reliable support for real-time robotic arm assembly tasks in resource-constrained environments.

AB - This research proposes LightIndust-Net, a lightweight object detection and instance segmentation algorithm specifically designed for robotic arm operations in industrial assembly scenarios. The algorithm effectively addresses the challenge of balancing accuracy and computational efficiency in current robotic operation systems through an innovative StarNet backbone network, Cross-Stage Star Bottleneck (CSSB) modules, and a Shared Convolutional Separator Segmentation Head (Seg_SCSS). Experimental evaluation shows that LightIndust-Net achieves a 22.1% improvement in computational efficiency and a 37.7% reduction in model parameters relative to the baseline approach, while maintaining or even improving detection and instance segmentation accuracy, particularly excelling in processing complex industrial component edge details and shape features. This efficient and precise visual perception solution provides reliable support for real-time robotic arm assembly tasks in resource-constrained environments.

KW - Edge Computing

KW - Industrial Assembly

KW - Instance Segmentation

KW - Lightweight Model

KW - Object Detection

UR - https://www.scopus.com/pages/publications/105023429984

U2 - 10.1007/978-981-95-4045-7_48

DO - 10.1007/978-981-95-4045-7_48

M3 - Conference contribution

AN - SCOPUS:105023429984

SN - 9789819540440

T3 - Lecture Notes in Electrical Engineering

SP - 491

EP - 504

BT - Proceedings of 2025 Chinese Intelligent Automation Conference - Volume I

A2 - Liu, Huaping

A2 - Guo, Di

PB - Springer Science and Business Media Deutschland GmbH

Y2 - 4 July 2025 through 6 July 2025

ER -