MobileNet-SSD for surface detection

• Main Author: Wang, Weiyi
• Other Authors: Zheng Jianmin
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/165968
• Institution: Nanyang Technological University

In this Final Year Project (FYP), we aim to develop a Mobilenet-SSD network to detect defects in surface images. In particular, the SSD detection network locates objects in the feature map using a collection of default boxes of various aspect ratios and sizes. To handle different object sizes, the SSD uses feature maps from different resolutions and integrates all computing into a single network. However, the large number of parameters makes the SSD too slow for edge devices. MobileNets, on the other hand, have a streamlined architecture using depth-wise separable convolutions to build lightweight deep neural networks. Therefore, we integrate MobileNet with SSD to create a Mobilenet-SSD network for surface identification that is both fast and accurate. Experimental results on the PASCAL VOC and NEU-DET datasets have validated the effectiveness of our Mobilenet-SSD network.