论文部分内容阅读
在大规模高风险考试的阅卷过程中常会遇到一定数量的未作答空白题。这些空白题若能由计算机自动识别,将提高阅卷效率和降低阅卷成本。本研究尝试利用神经网络进行空白题识别技术的开发,并讨论该技术在CET主观题阅卷中的应用。该研究提取出图像像素灰度值矩阵行向量、列向量标准差的标准差作为识别空白题的特征参数,选取自学习能力较强的Elman模型,以训练速度快、准确度高的traindx函数为训练函数,以梯度下降的learngdm为学习函数,以非线性的tansig和logsig为隐藏层和输出层的传递函数,并通过对隐藏层神经元数目的调整来优化网络,使网络能以较少的运算消耗获得较好的识别效果。初步研究结果表明该技术可以较好地识别空白题,在保证识别正确率的同时节约人力等资源。
In a large-scale high-risk test marking process often encounter a certain number of unanswered blank questions. If these blank questions can be automatically identified by the computer, will improve the marking efficiency and reduce the marking costs. This study attempts to use neural networks to develop blank title recognition technology, and discusses the application of this technology in CET subjective question marking. In this study, the standard deviations of row vector and column vector standard deviation of the pixel gray value matrix were extracted as the characteristic parameters of the blank problem. The Elman model with strong learning ability was selected to train the traindx function with high training speed and accuracy As a training function, gradient learning learngdm as a learning function, nonlinear tansig and logsig as the transfer function of hidden layer and output layer, and by optimizing the number of hidden layer neurons to optimize the network, the network can be less The operation consumes a better recognition result. Preliminary results show that the technology can better identify blank questions, save resources while ensuring the correctness of recognition.