Research On Deep Learning Method In Improved CNN For Intent Recognition Of Power Lower-limb Prosthesis

For lower limb amputees caused by diseases,traffic accidents,work accidents,sports injuries,etc.,prostheses can replace some of the functions of the lost limbs,so that the amputees can restore some life self-care and work ability.The use of various sensor information fusion,microprocessor control and other technologies,intelligent prostheses that can exercise according to user intentions and have the ability to actively adapt to changes in external conditions,bring new possibilities for disabled people to walk naturally and smoothly.The intention of dynamic lower limb prosthesis movement is one of the important research topics in the research and development of intelligent lower limb prosthesis.In recent years,deep learning has received extensive attention from researchers,and has achieved excellent results in research fields such as target detection,image classification,and video analysis.Traditional intent recognition research is more based on machine learning classifiers,but the effectiveness of deep learning algorithms in motion intent research is not yet known.Introducing the advantages of deep learning classifier self-learning features to replace the manual feature extraction work in traditional intent recognition research;optimizing deep neural networks to make the algorithm more suitable for motion intent recognition research is the purpose of this research work.First,it aims to apply the deep learning technology of self-learning features to the recognition of motion intention of lower limb prosthesis,and expounds the feasibility of applying deep learning algorithm to motion intention recognition.Two types of typical deep learning networks were selected: CNN and LSTM.The feasibility of applying deep learning models to motion intention recognition was experimentally verified,and was verified by comparison with traditional machine learning algorithms.Performance advantages of deep learning algorithms.The experimental results prove that it is of great significance to apply deep learning technology to the recognition of lower limb prosthetic movement intention.Second,it aims to optimize the CNN so that the algorithm can be better applied to motion intention recognition.The algorithm framework is improved on the basis of the traditional CNN by adding standardization,improving the structure of the convolution layer,adjusting the parameters,and adding the Dropout layer,which is more suitable for the recognition of motion intention based on the short-term behavior sample data studied in this paper.The sliding window is performed in the intent recognition data set,the purpose is to expand the data of the time series samples,and the augmented target data set can make the training set more abundant and comprehensive,and improve the accuracy of recognition.The improved CNN is used to The augmented data set performs feature learning and classification,and the experimental results verify the effectiveness of the algorithm in this paper.This thesis aims to effectively apply deep learning algorithms to motion intent recognition,effectively make up for the shortcomings of traditional machine learning algorithms used for intent recognition to manually extract features,and at the same time combine the features of intent recognition short-term behavior data to improve the traditional Compared with the traditional CNN,the CNN has improved recognition rate and time complexity performance.