Image Matching Algorithm Based On Deep Learning

With the development of intelligent manufacturing,computer vision has become one of the most important technologies in many fields.As one of the branches of computer vision,image matching is the core of many image processing,the quality of matching can always determine the result of image processing system.There is a problem with the current image matching,the extraction of the algorithm features based on traditional pattern recognition is very complex,and the experienced people need to carry out the features.While the effect of the algorithm based on deep learning is good and doesn't need manual for feature extraction and selection,the current deep learning model is often difficult because of too many parameters and too much calculation result in too long processing time.The main work of this thesis is as follows:?1?The traditional feature extraction technology,some techniques and concepts are introduced,the purpose and significance of image matching technology is analyzed.?2?The traditional perceptive hash algorithm is introduced.Compared with the traditional image matching algorithm,the overall effect of the deep learning model is significantly improved.?3?Put forward a method of faster image annotation.This thesis put a new model which has fewer parameters by modifying the stride of the Convolution kernel and the number of outputs of each layer.The new model runs almost 6 times faster than VGG₁6model.?4?The influence mechanism of different activation functions on the prediction model was explored.By constructing new model,it's confirmed that for the two optimization model and the original VGG₁6 model,Relu?Rectified linear unit?is more suitable than the Sigmoid activation function.?5?In order to prevent the limitation of single training from.causing over-fitting problems,the effect of dropout value is researched.