Application Research Of Offline Handwritten Digit Recognition Based On Artificial Intelligence

In recent years,handwritten numeral recognition has become an important branch of optical character recognition.At present,the application of printed and online handwritten numeral recognition systems is very mature,but offline handwritten numeral recognition is difficult to achieve a practical numeral recognition system because of its different writing methods.If the offline handwritten digital system can achieve high precision and high efficiency in mass production,this can greatly liberate people’s labor,not only reduce the possible errors in manual entry,but also reduce a lot of time consumption,responding to the pace of human progress towards an intelligent society.On the issue of how to improve the recognition rate of handwritten digits,this paper mainly focuses on the performance of classifiers,involving machine learning,artificial neural networks,in-depth learning and other fields.The innovations of the content are as follows:1)In the aspect of feature extraction,this paper chooses the strategy of parallel fusion of13-point grid feature and wavelet coefficient feature as the method of feature extraction.Because the 13-point grid divides the image into four parts horizontally and two parts vertically,the grid operation can extract the global contour features of the image,and the wavelet transform can characterize the time domain and frequency domain information of the image at the same time,with the characteristics of multi-resolution,so the wavelet coefficient features can extract more details of local features in the image.Features can be extracted better.2)Facing the problem of kernel function selection and parameter optimization in Support Vector Machine(SVM)and Kernel Extreme Learning Machine(KELM),the intelligent optimization algorithm is used to optimize the parameters in SVM and KELM respectively,and the best combination of parameters is obtained within the given range of values,which eliminates the need to repeatedly adjust the values of the parameters to be optimized in the experiment and saves a lot of time at the same time.3)To address the problems of non-convergence and overfitting in the Le Net5 model of convolutional neural network(CNN),a Le Net5+ model is proposed based on the improved Le Net5 model,which is improved in terms of convolutional kernel processing,pooling layer selection,activation function,etc.The improved Le Net5+ model effectively improves the overfitting phenomenon of the network.4)To address the shortcomings such as computational overflow of classifiers in convolutional neural networks(CNN),CNN is combined with SVM to build CNN+SVM hybrid network model,which enhances the classification performance in the hybrid model.To sum up,handwritten digit recognition is a cross field of modern pattern recognition and intelligent application,which not only has important theoretical value,but also has broad application value.