@article{XIA2026, 
author = {Yu XIA},
title = {Experimental design of object classification based on multimedium fusion},
year = {2026},
journal = {Experimental Technology and Management},
volume = {43},
number = {6},
pages = {235-240},
keywords = {object classification, multi-medium fusion, experimental teaching design},
url = {https://www.sciopen.com/article/10.16791/j.cnki.sjg.2026.06.030},
doi = {10.16791/j.cnki.sjg.2026.06.030},
abstract = {ObjectiveObject classification is a widely studied topic and an important teaching component in computer vision, primarily involving two key issues: feature extraction and representation, and the definition of classification criteria. In practical complex scenarios, small sample sizes, low signal-to-noise ratios, and small interclass differences are the main factors limiting classification performance, posing technical challenges in both teaching and research. To address these issues, we enhance the descriptive capacity of features to accurately characterize the target objects, thereby improving classification accuracy. Designing project-based experimental cases is essential to help students verify theories, understand algorithms, and innovate applications, cultivating their computer vision algorithm design capabilities throughout the process.MethodsThe experimental teaching design employs multimedia information fusion for target classification by acquiring image data from radar, visible light, and infrared sources. Three features are then extracted: target area, invariant moments, and Fourier power spectrum. An adaptive weight fusion strategy generates fused features, and a target classification benchmark library is constructed. The K-nearest neighbor (KNN) method is used for target recognition and classification; during this process, classification images are learned, and the feature benchmark library is updated in real time. To verify the proposed algorithm, three sets of experiments were conducted using relevant video sequences to validate the accuracy of feature-level fusion-based target classification and recognition. The results demonstrate that multimedium fusion markedly improves target classification accuracy and yields superior classification performance.ResultsFeature-level image fusion leverages multiple imaging sources to generate new representation features. Information from different media is more complementary, with higher credibility, less ambiguity, and better interpretability, enabling classification closer to human visual perception. Area intuitively expresses the size, dimensions, and shape of an object. Invariant moments are invariant to translation, rotation, and scaling. The Fourier power spectrum effectively describes texture and reflects detailed object information. Adaptive fusion of multiple features enhances matching ability in target classification, and highly discriminative template features markedly improve classification accuracy and robustness. Feature-level fusion, conducted at the intermediate level of computer vision, retains sufficient key information while compressing data to enable real-time processing.ConclusionsThe experimental program focuses on capability cultivation, emphasizing students’ abilities in knowledge application, analytical comprehension, development and design, and comprehensive professional qualities. Analysis of implementation outcomes across 162 students confirmed that the scheme is effective and advanced. It enhances students’ ability to recognize feature descriptions across different media, broadens the technical approach to image fusion, strengthens practical engineering skills in target classification, and improves innovative design capabilities in computer vision learning.}
}