Communication Signal Classification Method Based on Convolutional Neural Network

This paper proposes a multi-scale feature fusion framework for CNNs to coordinate local time-frequency features with global semantic features in communication signal classification. The framework integrates Inception-v3 to address the fixed receptive field of traditional CNNs, uses a Feature Pyramid Network (FPN) to align shallow high-frequency details with deep semantic features, and incorporates the SE (Squeeze-and-Excitation) attention module for adaptive feature optimization. Experimental results show that the model achieves accuracy of 85.3% and 92.3% in low signal-to-noise ratio and multi-interference scenarios, respectively, outperforming ResNet18 by 12.7%, with memory usage under 4 GB.