RGB-T salient object detection via excavating and enhancing CNN features

• Published in: Applied intelligence (Dordrecht, Netherlands) Vol. 53; no. 21; pp. 25543 - 25561
• Main Authors: Bi, Hongbo, Zhang, Jiayuan, Wu, Ranwan, Tong, Yuyu, Fu, Xiaowei, Shao, Keyong
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.11.2023; Springer Nature B.V
• Subjects: Artificial Intelligence; Computer Science; Deep learning; Design; Machines; Manufacturing; Mechanical Engineering; Modules; Neural networks; Object recognition; Processes; Representations; Researchers; Salience; Semantics; Joint attention; Feature enhancement; RGB-T salient object detection
• ISSN: 0924-669X; 1573-7497
• DOI: 10.1007/s10489-023-04784-1

RGB-T salient object detection aims to identify the most attractive object(s) in a scene using RGB and thermal data. For this task, on the one hand, how to excavate salient clues is crucial to improve the detection performance of the model. On the other hand, strengthening the representation of salient features is still a huge challenge. In order to solve the above issues, this paper proposes excavating and enhancing CNN features to boost the performance of RGB-T salient object detection ( E 2 N e t ). Specifically, we first design a new joint attention module (JAM) that jointly considers both dimensions of channel and pixel location to fully explore effective salient features by encoding contextual information as local features and simultaneously digging for useful clues from inside the channel. Different from previous works of expanding the feeling field, to enhance the feature representation, we propose a feature enhancement module (FEM), which realizes the parallel and independent learning of the four branches based on the channel splitting strategy, and greatly consolidates the feature expression ability. Extensive experiments show that our proposed model outperforms the existing twenty-two excellent models on three challenging benchmark datasets. Codes and results are available at: https://github.com/RanwanWu/E2Net .