Neural network-based homework test paper detection segmentation method

Abstract

The invention discloses a homework test paper detection and segmentation method based on a neural network, and relates to the technical field of teaching. YOLOv7 and MobileNetV3 are combined, the YOLOv7 output structure is modified, any quadrilateral area can be detected, dependence on an accurate scanner is eliminated, the non-standardized image processing capacity is remarkably improved, meanwhile, a two-stage detection method is adopted, and the detection efficiency of the homework test paper is improved. According to the method, the overall detection precision is improved, the missing detection condition is reduced when a small target is processed, the MobileNetV3 is introduced as a backbone network, the model parameter quantity is reduced, the requirement for the image resolution is lowered, and a good detection effect can be obtained under a medium-low resolution image, text region detection is carried out based on the combination of YOLOv7 and MobileNetV3, image correction and rotation are carried out through OpenCV, and the detection accuracy is improved. A two-stage detection method is adopted to extract refined content information, and finally, a detection result is optimized through post-processing, so that the performance is excellent in a complex photographing scene, and the detection speed and precision are also remarkably improved.

Images