Article: Improved SSD-based visual sorting control for industrial robots Journal: International Journal of Information and Communication Technology (IJICT) 2024 Vol.25 No.8 pp.64 - 80 Abstract: Traditional vision sorting methods for industrial robots often rely on rule-based image processing algorithms or simple deep learning models, which limit the system's recognition accuracy and real-time performance in complex environments. With the development of industrial automation and smart manufacturing, there is an increasing demand for vision sorting systems that can handle complex backgrounds and diverse targets. To address these issues, this work proposes an improved SSD-based vision sorting control system for industrial robots. Firstly, the VGG-16 backbone network of the conventional SSD is replaced with a GhostNet network to significantly enhance instant response capability of target recognition. Second, the SENet is brought to improve the accuracy of the target detection network. In addition, GhostNet is improved to enhance the flexibility of feature extraction by introducing dynamic convolution technique. The SENet module is optimised with a more efficient activation function to further reduce the computational complexity. Finally, the model's perceptiveness in identifying lesser targets is enhanced by a multi-scale feature fusion method. The experimental results show that compared with the original SSD model, the improved SSD model improves the mAP of target detection by 7.4% and the FPS by 28.4%. Inderscience Publishers - linking academia, business and industry through research

Title: Improved SSD-based visual sorting control for industrial robots

Authors: Wenke Yang; Guiyin Ran; Chao Yang

Addresses: Chongqing Institute of Engineering, Chongqing 400056, China ' Chongqing Institute of Engineering, Chongqing 400056, China ' Chongqing Institute of Engineering, Chongqing 400056, China

Abstract: Traditional vision sorting methods for industrial robots often rely on rule-based image processing algorithms or simple deep learning models, which limit the system's recognition accuracy and real-time performance in complex environments. With the development of industrial automation and smart manufacturing, there is an increasing demand for vision sorting systems that can handle complex backgrounds and diverse targets. To address these issues, this work proposes an improved SSD-based vision sorting control system for industrial robots. Firstly, the VGG-16 backbone network of the conventional SSD is replaced with a GhostNet network to significantly enhance instant response capability of target recognition. Second, the SENet is brought to improve the accuracy of the target detection network. In addition, GhostNet is improved to enhance the flexibility of feature extraction by introducing dynamic convolution technique. The SENet module is optimised with a more efficient activation function to further reduce the computational complexity. Finally, the model's perceptiveness in identifying lesser targets is enhanced by a multi-scale feature fusion method. The experimental results show that compared with the original SSD model, the improved SSD model improves the mAP of target detection by 7.4% and the FPS by 28.4%.

Keywords: industrial robots; vision sorting; single shot multibox detector; SSD; GhostNet; SENet.

DOI: 10.1504/IJICT.2024.142295

International Journal of Information and Communication Technology, 2024 Vol.25 No.8, pp.64 - 80

Received: 27 Aug 2024
Accepted: 13 Sep 2024
Published online: 17 Oct 2024 *

Title: Improved SSD-based visual sorting control for industrial robots

Keep up-to-date