A brain-inspired multibranch parallel interactive vision mechanism for advanced driver assistance systems. (17th November 2022)

• Record Type: Journal Article
• Authors: Ou, Wei; Jin, Zihan; Huang, Shiying; Du, Danlei; Ye, Jun; Han, Wenbao
• Is Part Of: International journal of sensor networks. Volume 40:Number 3(2022)
• Page Start: 203
• Page End: 216
• Publication Date: 2022-11-17
• Subjects: brain-inspired -- CNNs -- convolutional neural networks -- multibranch parallel interaction -- feature extraction -- shining attack; Sensor networks -- Periodicals; 681.2
• Languages: English
• ISSNs: 1748-1279

Abstract

Brain-inspired research promotes the intersection and integration of brain science, computers, and other disciplines, driving a new round of scientific and technological revolution. This paper discusses the research content, features, and current status of research on brain-inspired computing and proposes a brain-inspired multibranch parallel interaction model. The feature extraction of the proposed model consists of two parallel convolutional neural networks (CNNs): the main feature extractor connected to the classifier and the auxiliary feature extractor. The latter can interact with the main feature extractor for multistage feature interaction. Additionally, four parallel branches are introduced in our CNN to learn the information on the importance of different feature map positions. Given the existing attacks on self-driving cars (in the case of shining attacks), we use the proposed model for training and detection. The experimental results show that the proposed model effectively reduces the redundant information on the feature maps, improves image classification accuracy to a certain extent, and has high learning efficiency and accuracy.