文献类型：期刊文献

英文题名：Hyperspectral Image Small Target Detection: A Taylor-Type Six-Point Adaptive Discrete Neural Dynamics Approach

作者：Feng, Jiayi[1,2]; Liang, Jieming[2]; Chen, Jiayan[3]; Lin, Cong[3]; Xue, Zhong[1]

机构：[1] Ministry of Agriculture and Rural Affairs, South Subtropical Crops Research Institute, Chinese Academy of Tropical Agricultural Sciences, Key Laboratory of Tropical Fruit Biology, Guangdong, Zhanjiang, 524900, China; [2] Guangdong Ocean University, School of Mathematics and Computer, Zhanjiang, 524088, China; [3] Guangdong Ocean University, School of Electronic and Information Engineering, Zhanjiang, 524088, China

年份：2026

外文期刊名：IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing

收录：EI(收录号：20264021599088)、Scopus(收录号：2-s2.0-105052222584)

语种：英文

外文关键词：Constrained optimization - Hyperspectral imaging - Labeled data

外文摘要：In complex background scenarios, hyperspectral image small target detection (HISTD) is crucial for fine-grained information extraction and resource localization. However, existing traditional detection algorithms are simple, resulting in a high miss rate for weak targets. In contrast, data-driven methods suffer from poor generalization due to annotation difficulty and data scarcity. In this context, constrained energy minimization (CEM), which requires no annotations and can suppress background interference while enhancing target responses, has become a mainstream method. However, its solution process involves large-scale high-dimensional matrix operations, leading to slow convergence and low accuracy. To address this issue, this paper proposes a HISTD method based on a Taylor-type six-point adaptive discrete neural dynamics CEM model (T6ADND-CEM). The method is built on the zeroing neural dynamics (ZND) framework. An adaptive coefficient is introduced to accelerate convergence, and a Taylor-type six-point discrete formula is used for implementation. In addition, an adaptive step size is incorporated to further improve solution accuracy. Experimental results show that T6ADND-CEM achieves good accuracy and fast convergence in HISTD tasks, with a steady-state error of 5.9281 × 10-14 and a convergence time of 0.2320 s. Compared with other methods, the proposed method shows better error control and convergence efficiency, while requiring no large-scale labeled training data. ? 2008-2012 IEEE.