Research On Pattern Recognition Problems Based On Convolution Neural Networks

With the arrival of the era of artificial intelligence,more and more researchers have paid more attention to the field of computer vision.The rise of deep learning technology promotes the progress of the research work in this field to a great extent.Deep learning is a branch of machine learning,it has achieved significant progress in many fields,such as image detection,image recognition,scene analysis,driveless cars,speech recognition,etc.The different expressions of each person have an obvious distinction,but the same expression of different people also shows great differences,then we can see that the facial expression recognition is still a difficult problem to solve,the presentation of facial expression mainly rely on the movements of different components of the face to complete,this paper proposes a method of facial expression recognition based on multi-scale global and local facial features.Firstly,each facial expression image produces 10 slices(5global slices based on the original facial image and 5 local slices according to the five key points)which both have 3 different scales,then all slices are horizontally flipped,finally,each facial expression image has 60 slices.In order to get more distinguishing deep feature,this paper trained a better deep model by continuously adjusting the parameter combination of convolution neural network,finally the classification result is obtained by using LIBSVM classification algorithm.Compared with other methods,the method in this paper has achieved good recognition results on two publicly available datasets: CK+ and JAFFE.Traditional insect recognition methods mainly rely on the entomology experts who have the professional knowledge to identify,this method requires a lot of manpower,however,entomology expertswho have the professional knowledge are scarce,so the automatic recognition of insects species is becoming more and more important.At present,insect recognition is mainly based on insect morphology and insect sound system,because different insects species have obvious differences in the shape,texture and color features,and the sound of the same insect species have a strong similarity,but the different species have significant differences.This paper presents a new method of insect recognition which converts insect sound files into the spectrum image by using the Fast Fourier Transformation algorithm,and trains a good performance deep model by using these spectrum images and convolutional neural network,then extracts deep features corresponding to each insect sound file,finally conducts experiments with LIBSVM as the classification algorithm.Compared with other methods,the proposed method can effectively improve the insect recognition accuracy.To further improve the accuracy of insect recognition,this paper uses the batch normalization operation to ensure the consistency of the data distribution during training the network and reduce the influence of the data distribution in each batch.This paper mainly studies two pattern recognition problems based on deep learning technology which includes facial expression recognition and insect recognition,and effectively improves the accuracy of facial expression recognition and insect recognition,which lays the foundation for the future research work.