Adaptive Semantic Perception Model for Deep Learning-Based Image Processing and Pattern Recognition

Abstract

With the rapid increase in the volume and complexity of image data, traditional image processing and pattern recognition techniques face growing challenges in accuracy, adaptability, and computational efficiency. To address these issues, this paper proposes an Adaptive Semantic Perception Model (ASPM), which integrates three core components: A Semantic-aware Convolutional Module (SSCM), a Hierarchical Semantic Fusion Unit (HSFU), and an Adaptive Domain Adjustment Module (ADAM). These components work synergistically to extract, integrate, and adapt multi-level semantic information from images. The ASPM model is evaluated on three representative datasets: MNIST, CIFAR-10, and chest Xray images. Quantitatively, ASPM achieves 99.8% accuracy and over 99.5% F1 scores across all digit classes in MNIST; 95.5% accuracy and an average F1 score improvement of 3–4% over baseline models on CIFAR-10; and 85.0% accuracy with an F1 score of 85.2% for pneumonia and 86.0% for pulmonary nodules in the medical image dataset. These results demonstrate the model’s robustness, semantic sensitivity, and strong cross-domain generalization.