- Conference Article
- 10.1109/mlcipr68329.2025.11407339
Research on Target Recognition Method in Complex Scenes Based on Improved Convolutional Neural Network for Multi-Source Data Fusion
- Dec 19, 2025
- Yupeng Liu + 3 more +3
Under the background of informatized and intelligentized application scenarios, the requirements for real-time performance, accuracy, and robustness of target recognition in complex scenes have surged. Traditional single-sensor methods are difficult to meet the demands due to one-sided information and vulnerability to interference. This paper proposes a multi-source data fusion recognition method based on an improved Convolutional Neural Network (CNN): a multi-sensor collaborative system integrating LiDAR, infrared, and millimeter-wave sensors is constructed. Deep learning is used to process multi-source heterogeneous data, mining deep complementary information to generate composite representations. The core innovation lies in the improved CNN model, which automatically extracts deep features of composite images, avoids defects of manual feature design, and overcomes recognition challenges such as target posture, illumination, and occlusion. To validate the effectiveness of the proposed method, a simulation dataset was constructed and preliminary experiments were conducted. The results demonstrate that significant improvements over traditional methods, single-sensor methods, and simple fusion schemes are achieved by the proposed method on key metrics such as mean Average Precision (mAP) and F1-Score. Ablation studies further confirm the effectiveness of each proposed module. A technical path for high-precision target recognition in complex Scenes, integrating theoretical innovation with preliminary empirical support, is provided by this research.
Read more