Research On Recognition Algorithm Of Pressure Barefoot Footprint Based On Convolutional Neural Network

Footprint,as a biological feature of the human body,is one of the most valuable physical evidence at the crime scene,providing effective clues for investigating cases.Footprint is unique and stable under the influence of foot bones and behavior habits.Therefore,footprint identification plays an important role in public safety,identity identification,criminal investigation and other fields.However,for a long time,the traditional empirical manual analysis and judgment method has been used in footprint identification and recognition,which is mainly due to the lack of effective footprint feature extraction and recognition algorithms.With the increasing application of deep learning in the field of image processing and pattern recognition,it provides new ideas and directions for the study of footprints.In this dissertation,the pressure barefoot footprint image is taken as the research object,and the data set for footprint research is collected and constructed.Combined with the method of depth convolution network,the extraction and recognition algorithm of barefoot footprint features is studied.The main research contents and results are as follows:(1)A pressure footprint data set is collected and constructed.Using the pressure footprint acquisition instrument,the footprint acquisition process and specification were established.A total of 2000 pressure barefoot footprint images of 100 people are collected.In order to improve the quality of the footprint image,the collected footprint data are filtered,denoised and standardized to reduce noise interference.Finally,the pressure barefoot footprint data set is expanded by data enhancement method to increase data diversity.(2)A footprint recognition algorithm based on Spatial Aggregation Weighting Module(SAWM)is proposed.The algorithm uses the attention mechanism to extract the properties of local features and distinguish the nuances of different footprints,so as to recognize the footprints.The algorithm sequentially superimposes all channels in the feature map,focuses on the high-response area in the feature map,retains features above the threshold,multiplies the original feature map,extracts salient features,reduces feature dimensions and optimizes features.Then embedding the SAWM module in the VGG19 network can improve the performance of the network model and enable the network to focus on significant local area.This dissertation conducted a lot of experiments on the footprint data set to show the effectiveness of the algorithm.Finally,the algorithm in this dissertation compares with other attention mechanisms and traditional algorithms.The algorithm shows better performance on the task of footprint recognition and greatly improves the accuracy of footprint recognition.(3)A footprint recognition algorithm based on ResNet50 two-branch network is adopted.In order to further dig out the local feature information of the pressure barefoot footprint and improve the ability to distinguish features.The algorithm first uses an image reconstruction to destroy the original footprint image and reconstruct a new footprint image,increasing the diversity of data and highlighting the local information of the footprint.So that the network can easily learn the key features.Then it combines with the ResNet50 two-branch network to extract the features of the original footprint image and the reconstructed footprint image respectively.It locates the discriminative area and learn the discriminative features in the footprint image.Finally,the idea of discriminator is used to restrict two different features,so as to distinguish two images and eliminate the differences between them.A large number of experiments show that the algorithm further improves the accuracy of footprint recognition,which shows the effectiveness of the algorithm in the field of footprint recognition.