This study introduces EA-OBB, a lightweight rotated object detection framework designed for detecting one-dimensional (1D) and Quick Response barcodes. Built upon the YOLO11 architecture, EA-OBB integrates several innovative modules-KWConv, ORPNCSPELAN, and LADH-OBB-to enhance both accuracy and computational efficiency in rotated object detection. The KWConv module utilizes a dynamic convolution kernel mechanism to improve rotational barcode feature extraction. The ORPNCSPELAN module enhances computational efficiency through multi-path feature aggregation and online re-parameterization. The LADH-OBB module decouples classification and regression tasks, improving the precision of rotation angle regression. To further adapt to resource-constrained environments, this study incorporates the Taylor Pruning algorithm, significantly reducing model parameters and computational costs. Experimental results on the RotBar dataset demonstrate the superior performance of EA-OBB, achieving an optimal balance of precision, recall, and computational complexity compared to existing methods.
Luo et al. (Tue,) studied this question.