Arithmetical Evaluation System based on improved-YOLOv5 and CRNN networks

For primary school teachers, revising arithmetical exercises is labor-intensive and time-consuming task. To reduce their burdens, we propose a lightweight arithmetic evaluation system that can automatically assess arithmetic exercises. The designed system divides into two branches: detection and recognition. In the detection branch, a coordinate attention module is added to the feature pyramid infrastructure of the YOLOv5 target detection network to improve the multi-scale target recognition ability. Then, the complete IoU (CIoU) loss function leads to faster convergence and better performance on the detection results. The experimental results show that the detection accuracy of the proposed method improves mAP@.5-95(%) by 5.4% compared to the original algorithm. Moreover, the number of parameters and FLOPS decrease to 1.23M and 3.8G. In the recognition branch, we employ the improved-CRNN network to recognize arithmetical exercises, achieving an accuracy being up to 97.1 %.