PIER B | PIER Journals

2026-11-03 Latest Published

TMCFNet: Triple-Stage Mamba Cross-Modal Fusion Network for Visible-Infrared Remote Sensing Object Detection
Download Full Article View Full Article

By Cheng Zhang Xiao-Zhen Ren Xiaqiong Fan Zhipeng Guo Gang Liu

Progress In Electromagnetics Research B, Vol. 119, 19-32, 2026

DOI:10.2528/PIERB26072205 | Google Scholar

Abstract

Unimodal remote-sensing data is susceptible to interference such as illumination changes and texture loss. Visible and infrared images have respective limitations in object detection, which multimodal fusion can address. To address these challenges, this study proposes a Triple-stage Mamba Cross-modal Fusion Network (TMCFNet) for cross-modal remote-sensing object detection. The model performs cross-modal image fusion at the input stage to preserve each modality's inherent information while keeping computational cost low. It consists of three key components: the Adaptive Multi-scale Convolution Module (AMC), Cross-Modal Shared Mamba Interaction Module (SMIM), and Dual-Global Pooling Enhanced Multimodal Feature Fusion Module (G2MFF). They are responsible for unimodal feature enhancement, cross-modal information alignment, and in-depth feature fusion, which effectively improve the model's detection accuracy and generalization ability in complex scenarios. To comprehensively evaluate the model performance, we conducted experiments on multiple public real-world multimodal datasets and dedicated remote sensing datasets. Compared with state-of-the-art unimodal and multimodal detection methods, the proposed model achieved mean average precision mAP@0.5:0.95 values of 58.5% and 69.0% on the DroneVehicle and LLVIP datasets, respectively. Results demonstrate that deep multimodal feature fusion can remarkably boost object detection performance in complex environments.

2026-11-03

PIER B

Vol. 119, 19-32, 2026

DOI:10.2528/PIERB26072205

download: 4

TMCFNet: Triple-Stage Mamba Cross-Modal Fusion Network for Visible-Infrared Remote Sensing Object Detection

Cheng Zhang, Xiao-Zhen Ren, Xiaqiong Fan, Zhipeng Guo and Gang Liu

2026-10-18

PIER B

Vol. 119, 1-18, 2026

DOI:10.2528/PIERB26062701

download: 29

Quasi-Reflectionless Filtering Antenna for 5G Wireless Systems

Mohammed Yaqoub Yousif, Faris Mohammed Ali, Nasr Alkhafaji, Rand Muwafaq Hadi and Bahaa Hamzah Alkhuwaildi

This paper presents a single-layer quasi-reflectionless filtering antenna (filtenna) operating at 3.5\,GHz for compact 5G mid-band radio-frequency (RF) front-ends. The structure simultaneously provides radiation, bandpass filtering, and absorption of out-of-band reflections. Unlike conventional reflective filtennas, rejected energy is dissipated in a chip resistor rather than returned toward the source, which is beneficial when the antenna port drives or is driven by nonlinear stages such as power amplifiers or low-noise amplifiers. Theoretical analysis, full-wave simulation, and measurement are used as a proof of concept. A T-slot microstrip patch, a two-pole U-shaped bandpass filter (BPF), and a one-pole absorptive stub are co-integrated on flame-retardant type 4 (FR4) (ε_r = 4.6, thickness 1.6 mm). At the design frequency, the absorptive stub behaves nearly as an open circuit, so in-band power is delivered to the radiator; outside the passband, out-of-band signals are routed into the absorptive path. Quasi-reflectionless matching with S₁₁ < -10 dB is obtained across approximately 1–7 GHz, with a realized gain of about 1.99 dBi and filtering radiation characteristics. Measured results agree well with analytical and simulated predictions.