A channel and spatial attentions mechanisms with CNN for enhanced license plate detection

Real-world License plate detections presents notable challenges arising due to some factors such as variable illumination, occlusions, complex backgrounds and plate distortions. Traditional approaches such as edge detection often generate numerous false positives under such conditions, while template matching methods typically struggle with adaptability to diverse plate styles. Although Convolutional Neural Networks (CNNs) have improved feature extraction capabilities, it may still misallocate attention to irrelevant regions, reducing detection accuracy. To overcome these limitations, this study proposes an refined deep learning architecture that integrates Convolutional Block Attention Modules (CBAM) into CNN-based feature extractors. By jointly exploiting both channel and spatial attention mechanisms, the model selectively highlights salient features and minimizes the influence of background noise, enhancing its focus on license plate regions across varied image environment. Experimental evaluation using Kaggle License Plate Dataset shows that the proposed CNN + CBAM framework achieves a precision of 92.41%, recall of 89.73%, F1-score of 91.05%, and mAP@0.5 of 88.9%. These results represents an improvement of 7.82% in F1-score and 10.25% in mAP compared to the baseline CNN, validating that the extracted crucial features substantially improve detection accuracy while minimizing false detections. The proposed approach also demonstrates a significantly reduced false detection rate of 5.6%, outperforming both traditional techniques and conventional CNNs based models.