Enhancers are cis-regulatory DNA elements whose accurate Identification and strength prediction (strong vs. weak) is vital for understanding gene regulation yet remains challenging due to sequence variability and limited labels. We introduce DCA-Enhancer, a neural network that integrates fine-tuned DNABERT-2 embeddings, dual-scale convolutional filters, and a spatial attention mechanism to capture both short and long motifs and emphasize key regions. Evaluated on benchmark datasets with 5-fold cross-validation and independent tests against seven leading models, DCA-Enhancer achieved 92.1% accuracy (MCC 0.84) for enhancer Identification and 71.0% accuracy (MCC 0.43) for strength prediction (87.5% on independent test). These results demonstrate its robustness and superior performance in genomic sequence analysis.