Deep Learning-based Multimodal Fusion for Improved Object Recognition Accuracy
Qi Wang, Zijun Gao, Taiyuan Mei, Xiaohan Cheng, Wenjun Gu, Haohao Xia
Abstract
This paper explores the application of deep learning and multimodal information fusion to enhance image recognition capabilities. Multimodal fusion, integrating information from different sources like audio and images, offers a deeper understanding of objects compared to unimodal approaches. Deep learning techniques are employed to achieve this fusion, specifically through an enhanced Inception network. This research investigates the impact of multimodal information on recognition accuracy by comparing purely image-based, purely sound-based, and multimodal fusion scenarios. The results demonstrate a significant increase in average recognition accuracy for moving objects through multimodal fusion. This advancement holds considerable value in various fields, including autonomous driving. The findings contribute to the development of more accurate object recognition systems across diverse applications.