深度学习通过构建多个隐层的神经网络，进行数据训练。学习样本的特征是训练

Deep learning performs data training by building neural networks with multiple hidden layers. Learning the features of the samples is the essence of the training samples, thus eliminating the need for manual feature extraction that relies on experience in the past. What is different from the traditional machine learning algorithm is that it needs to perform artificial feature extraction first and then go through the sample data input into the model, and then update the weight parameters of the model through the algorithm. The model can get an accepted prediction result when a batch of data that meets the characteristics of the sample is input into the model. The deep learning algorithm does not need to manually extract features before going through the sample data input model. After the sample data is input into the algorithm, the algorithm model will extract basic features (image pixels) from it. As the model progresses, these basic features combine higher-level features, such as simple shapes. The combination is then transformed into more complex features, so that different categories of images can be distinguished. After going through steps such as updating model weight parameters in a machine learning algorithm, satisfactory results can be predicted. <br>I use deep learning to build a convolutional neural network of different depths to compare the handwritten digits, and compare the performance differences of using other methods to recognize characters such as SVM, and finally select the convolutional neural network with the best recognition effect to realize the recognition system.

Deep learning trains data by building neural networks with multiple hidden layers. The characteristics of the learning sample are the essence of the training sample, thus eliminating the previous process of manual feature extraction by relying on experience. Different from the traditional machine learning algorithm, it is necessary to carry out artificial feature extraction before going through the sample data input model, and then the algorithm to update the weight parameters of the model. The model can get an accepted prediction when a batch of data that meets the sample characteristics is entered into the model. The deep learning algorithm does not need to be extracted by artificial features before going through the sample data input model, from which the algorithm model extracts the basic features (image pixels) after the sample data is entered into the algorithm. As the model progresses, these basic features are combined with higher-level features, such as simple shapes. The combination then transforms into more complex features to make the different categories of images distinct. Further steps such as updating the model weight parameters in machine learning algorithms can be obtained with satisfactory results prediction.<br>I used deep learning to build a varied neural network of different depths to compare handwritten numbers, and compared the performance differences of characters such as SVM by other methods, and finally selected the best-recognized convolutional neural network to implement the recognition system.

Deep learning is to train data by constructing neural networks with multiple hidden layers. The feature of learning sample is the essence of training sample, so it can avoid the previous steps of artificial feature extraction depending on experience. Different from the traditional machine learning algorithm, we need to extract the artificial features first, then input the sample data into the model, and then update the weight parameters of the model through the algorithm. The model can obtain an acceptable prediction result when a batch of data that conforms to the characteristics of the sample is input into the model. However, the deep learning algorithm does not need to experience the sample data input model before the artificial feature extraction. After the sample data is input into the algorithm, the algorithm model will extract the basic features (image pixels) from it. With the development of the model, these basic features combine with higher-level features, such as simple shape. Then the combination is transformed into more complex features so that different types of images can be distinguished. After the steps of updating model weight parameters in machine learning algorithm, satisfactory result prediction can be achieved.<br>I use deep learning to build convolutional neural network of different depths to compare the handwritten numbers, and compare the performance differences of other methods to recognize characters such as SVM, and finally select the convolutional neural network with the best recognition effect to realize the recognition system.