|
|
|
| Sensor-Adaptive Target Recognition Algorithm Based on Depthwise Separable Convolution and Feature Fusion |
| CONG Xiaoyu1， HUANG Xinhua2，3， YANG Jiayi2，3， ZUO Qian1 |
| 1. School of Information and Artificial Intelligence， Yangzhou University， Yangzhou 225127， Jiangsu， China；
2. Shanghai Electro-Mechanical Engineering Institute， Shanghai 201109， China；
3. Key Laboratory of Automatic Target Recognition （Shanghai）， Shanghai 201109， China |
|
|
|
|
Abstract This paper addresses the problem of variable sensor numbers in multi-sensor collaborative target recognition by proposing an adaptive collaborative recognition technique. The algorithm constructs a feature processing framework applicable to data from multiple homogeneous sensors, including radar, synthetic aperture radar (SAR), infrared, and visible light. Methodologically, the network employs depthwise separable convolution (DS-Conv) to achieve lightweight feature extraction. The core feature fusion module utilizes an improved squeeze-and-excitation (SE) attention mechanism for weighted feature fusion. The fused features are subsequently fed into a channel attention classifier for recognition. Experiments are validated on the MSTAR dataset. Under the worst-case scenario where the number of sensors is only two, the recognition accuracy reaches 92.75%. The results demonstrate that the proposed algorithm effectively resolves the feature fusion problem under varying sensor numbers, outperforming multiple comparative algorithms and exhibiting favorable feasibility and effectiveness.
|
|
Received: 16 February 2026
Published: 10 July 2026
|
|
|
|
|
|
| [1] |
XUE Yujia, ZHUANG Ruowang, XU Yi, RUAN Kaizhi, ZHONG Kai, HU Jinfeng. Intelligent Spectrum Sensing Technology in Unmanned Air Defense Systems[J]. Air & Space Defense, 2026, 9(3): 69-79. |
| [2] |
CAO Jinghao, ZHANG Junju, HUANG Wei, YAO Ruotong, ZHANG Ping. UAV Recognition and Detection Based on Multi-scale Feature Fusion[J]. Air & Space Defense, 2021, 4(1): 60-64. |
|
|
|
|
2026, Vol. 9 
