基于深度学习的验证码识别技术研究

针对基于机器学习的传统验证码识别受字符分割限制与人工操作过多等问题,基于深度学习Tensorflow学习框架将卷积神经网络应用到验证码的特性提取、分析、归类和识别中.将图片验证码作为整体输入,改进传统的LeNet-5网络结构,构建一种端到端的9层卷积神经网络,对验证码图像由低级到高级逐层提取图像特征,实现对图片验证码的识别.模型确定后采用控制变量法,针对每一迭代次数所处理的图片数量进行分析,对其准确率、损失值、训练时间进行综合测评,最终选取最优参数.实验结果显示,每批次处理128张图片,每迭代次数用时6秒,准确率的上限最高达到92%,损失值的下限最低达到0.0184.

Research of Captcha Recognition Based on Deep Learning

Aiming at the defects of traditional machine learning based captcha recognition,this paper deployed a deep learning Tensorflow framework based model,which improved the lenet-5 network,designed an end-to-end 9-layer convolutional neural network,and extracted image features of the captcha images layer by layer for identifying the captchas more better.In the new model,the control variable method was adopted to analyze the number of images processed in each iteration,and the accuracy,loss value and training time were comprehensively evaluated,and then the optimal parameters were finally selected.The experimental results showed that 128 images in each batch taking 6 seconds performed quite well.The upper limit of accuracy was up to 92%,and the lower limit of loss value is up to 0.0184.