Enhancing Feature Extraction in DBNet: A Novel Approach with Squeeze-and-Excitation Block and CBAM Module

Accurately extracting feature information from street view images can improve the accuracy of digital detection. However, street view images vary in resolution, have complex backgrounds, and exhibit irregular arrangements of digits, posing significant challenges to object detection tasks. To better extract features from street view images, this paper proposes the CBAM-SE-DBNet model by incorporating two attention modules, namely the CBAM Module and the Squeeze-and-Excitation Block (SE Block), based on the DBNet architecture. The CBAM Module adaptively adjusts the channel and spatial attention of feature maps, aiding in the extraction of crucial information. On the other hand, the SE Block selectively enhances useful features by learning the importance weights for each channel. The combination of these two modules allows CBAM-SE-DBNet to effectively capture the key features of digits in street view images. The proposed approach follows a three-step process: Firstly, the CBAM Module is utilized to extract initial image features. Secondly, a ResNet network improved with SE Blocks is employed to further extract features and obtain deep features. Finally, the features are fed into the DBNet network to obtain the prediction results. The SVHN dataset is used as a practical example in this study, and comparative experiments are conducted against the original DBNet model and the Bidirectional LSTM and Deep Neural Network model. The experimental results demonstrate that the proposed model achieves better precision and score compared to other models. This research provides valuable insights for further advancements in street view digit detection and holds wide-ranging application prospects.