23 April 2024 in IEEE Transactions on Neural Networks and Learning Systems
Recently, the multiscale problem in computer vision has gradually attracted people’s attention. This article focuses on multiscale representation for object detection and recognition, comprehensively introduces the development of multiscale deep learning, and constructs an easy-to-understand, but powerful knowledge structure. First, we give the definition of scale, explain the multiscale mechanism of human vision, and then lead to the multiscale problem discussed in computer vision. Second, advanced multiscale representation methods are introduced, including pyramid representation, scale-space representation, and multiscale geometric representation. Third, the theory of multiscale deep learning is presented, which mainly discusses the multiscale modeling in convolutional neural networks (CNNs) and Vision Transformers (ViTs). Fourth, we compare the performance of multiple multiscale methods on different tasks, illustrating the effectiveness of different multiscale structural designs. Finally, based on the in-depth understanding of the existing methods, we point out several open issues and future directions for multiscale deep learning.
19 January 2026 in IEEE Transactions on Geoscience and Remote Sensing
Hyperspectral anomaly detection (HAD) is one of the research hotspots within the domain of remote sensing image interpretation. The key is how to effectively distinguish the anomaly from the background. Existing methods usually rely on the assumption that the background of hyperspectral images (HSIs) conforms to multivariate Gaussian distributions. However, for real data, this assumption does not necessarily fit. Building upon the study of background distributions in real hyperspectral data, this article leverages optimal transport theory to model the background reconstruction task. The optimal transport of reconstruction and background is mathematically proven to be converted into that of reconstruction and the original image. To facilitate the reconstruction process, the Transformer architecture is employed due to its inherent ability to compute similarity weights. Based on the above, the optimal transport-guided Transformer anomaly detector (OT-AD) is designed by applying the Wasserstein distance constraints derived from optimal transport theory to the deep network. The experimental validation conducted on multiple public datasets confirms the effectiveness and reliability of the proposed method. The code for this work is publicly available at https://github.com/jiojio0505/OT-AD