Application Of Machine Learning In Speech Recognition And Image Recognition

With the development of artificial intelligence,speech recognition and image recognition have been two important areas in pattern recognition.Speech recognition has good prospects in social production and life.Image recognition is an important branch of pattern recognition,and has been successfully applied in the field of computer vision,such as military,medical and industrial.Artificial neural networks(ANN),which belong to the field of machine learning,have an adaptive,self-learning and parallel distributed structure.Since the last century since 80 s,with the rapid development of computer science and technology,the research of artificial neural network has made great progress,there have been hundreds of artificial neural network,which is representative of the BP neural network and Kochelen neural network(SOM),convolutional neural network(CNN),has been widely applications in signal processing,pattern recognition,image processing,financial field,have a good performance in the field of speech recognition and image recognition.Support vector machine(SVM)is a kind of machine learning method based on statistical learning,which has very strong generalization and global optimality.This thesis is mainly used in different machine learning to classify the speech signal and image,mainly uses the BP neural network model and improved BP neural network model,SVM,improved particle swarm optimization algorithm(PSO)SVM model,PCA optimize SVM model,CNN model and improved CNN model.Different signal and picture set features are extracted and classified,The main contents of this paper include:1.The Mel cepstral coefficient method(MFCC)is applied to extract 4 different kinds of music signals,which can represent each signal's characteristics and discretize the continuous2.signals,thus reducing the complexity of the signals.3.The BP neural network model,the improved BP neural network model and the improved PSO SVM model are used to identify and classify the extracted feature signals,and the results are all over 85% accuracy.4.The combination of PCA and SVM is applied to the MNIST handwritten digital set,and the recognition classification is carried out.At the data level,the high dimension image data is reduced to reduce the dimension,and the data is compressed.This greatly improves the performance of the algorithm,the recognition accuracy is up to 98%,and the running time is shortened by about 90%.5.Using CNN to automatically learn the target of SAR image to extract features.Then the traditional CNN is improved,and the Softmax classifier is replaced by SVM classifier.The classification effect is further improved,and the recognition accuracy is as high as 99%.