| Footprint is affected by foot bone,physiological state and other factors,and has a greater correlation with age.In the process of detecting criminal cases,the images of the footprints left on the scene can be used to analyze information such as the age of the criminals and the trajectory of the crime,which can accelerate the speed of case detection.However,in the field of age prediction of footprint images,traditional methods mostly rely on expert experience to manually extract the physical characteristics of the footprint to analyze the age information of the objects left by the footprint.This method is not only time-consuming but also very subjective.With the rapid development of deep learning,the neural network as the core of its technology has a strong learning ability,not only can improve the efficiency of feature extraction,but also can extract the deep semantic information of the footprint image.Aiming at the problems in the above-mentioned research on the age prediction of footprint images,combining deep learning and age prediction of footprint images,two age prediction algorithms for footprint images are given.The main research contents are as follows:(1)The optical shoe footprint dataset and the optical barefoot footprint dataset are constructed.The data sources are 1,035 pieces of optical shoe footprint data collected by relevant departments of the Ministry of Public Security of Henan Province and 2,073 optical barefoot footprint data collected by relevant departments of the Ministry of Public Security of Guangdong Province.In order to improve the signal-to-noise ratio of the footprint samples and improve the accuracy of age prediction,the collected footprint samples are de-scaled and denoised.Abnormal data is eliminated,so that the data samples are normally distributed.(2)A STA-Net-based age prediction algorithm for footprint images is given.The algorithm first uses the hollow convolution to expand the receptive field of VGG19,and adds the attention mechanism to enhance the significance information of the plantar pressure characteristics.Secondly,modify the traditional single-layer regression method of the fully connected layer of the regression network,split the fully connected layer into multiple modules and return to multiple ages at the same time.Finally,the multiple ages are weighted to get the predicted age.The algorithm is verified on the optical shoe footprint data set,and the results show that the algorithm has better performance.(3)An age prediction algorithm for footprint images based on two-point representation is given.First,divide the age into 11 age ranges.Secondly,the network model is used to predict the age range that the footprint image belongs to,and the weight of the age value of the end point of the range is calculated.Finally,the age of the object is predicted by age weighted summation.The algorithm was tested on the optical shoe footprint dataset and the optical barefoot footprint dataset,and achieved good prediction accuracy.Finally,a footprint image age prediction system was designed and implemented based on the algorithm. |
