摘要
为降低考试阅卷对专用识别设备与答题卡的要求、消除阅卷中的计分误差,依据传统阅卷环节设计了一种基于图像识别的阅卷平台。计分区域图像经图像预处理、倾斜校正、图像分割、卷积神经网络识别后获得试卷总分。与单一的卷积神经网络相比,该方法有效提高了手写数字的识别率及鲁棒性。在实际阅卷测试中,相比于传统阅卷,该平台优化了阅卷环节,降低了阅卷中的人为失误,减少了对人力、物力资源的浪费。最后,通过服务器搭建端对端的信息传递平台,完成了阅卷平台小型化、智能化的设计。
In order to reduce the requirements of the examination papers for identification devices and answer sheets,and to eliminate the accumulated error in the marking,a marking platform based on image recognition is designed.The scoring area is processed by the preliminary image processing,tilt correction,image segmentation and the convolutional neural network to obtain scores.Compared with a single convolutional neural network,the method effectively improves the precision rate of handwritten digits and ensures better robustness.In the actual test,compared with the traditional manual scoring,the platform optimizes the scoring process,reduces human errors and the waste of resources.In the end,the platform is built through the server,and the design of miniaturization and intelligence of the marking platform is completed.
出处
《科学技术创新》
2020年第3期75-77,共3页
Scientific and Technological Innovation
基金
国家级大学生创新训练计划项目资助(201810359016)
关键词
手写数字识别
卷积神经网络
阅卷平台
Handwriting numeral recognition
Convolutional neural network
Marking platform
