Multi-task Digital Verification Code Recognition Based on Improved ALexNet

Verification code is a human-machine test method that is widely used in the Internet. It can essentially distinguish between human and network robots. However, with the development of technology, more and more problems has been found in the verification code. This paper takes the digital verification codes commonly used in university student management system as the research object. The verification code after scaling and graying is used as the data set. The eight-layer ALexNet convolutional neural network is constructed by using the Google deep learning framework TensorFlow to train the verification code. The data set is tested to obtain a network model that can identify different complexity digital verification codes. The recognition rate of digital verification codes commonly used in college student management systems can reach 99%.