TDNN时间延迟神经网络

《TDNN时间延迟神经网络》由会员分享，可在线阅读，更多相关《TDNN时间延迟神经网络（27页珍藏版）》请在装配图网上搜索。

1、1会计学TDNN时间延迟神经网络时间延迟神经网络Introduction In this paper, the author presents a novel concept for simultaneous shape estimation and motion analysis based on a TDNN(time-delay neural network) architecture with adaptable spatio-temporal receptive fields.Time delay network structure with adaptable spatio-tem

2、poral receptive fieldstThe input image sequence is of the dimension Sx(1) Sy(1) St(1)Time delay network structure with adaptable spatio-temporal receptive fieldsttBxytrsmnpBranchBranchsijttBxytBxytsijtvsijqkktkTime delay network structure with adaptable spatio-temporal receptive fieldsError function

3、 A k = c (current class c), A1?k = 0 k cExperiments The size of a single image is Sx(1) = 32 by Sy(1) = 16 pixels, one sequence consists of St(1) = 8 such images. There are two shape classes, and five speeds V0 = -4, -2, 0, 2, 4. The training sets consist of 5000 examples, 500 of each class.The netw

4、ork parameters are as follows:And the error rate on the test set is 1.1%.ExperimentsAn Adaptable Time-Delay Neural-Network Algorithm for Image Sequence AnalysisLiu Xiao 2014.8.18IEEE TRANSACTIONS ON NEURAL NETWORKS, 1999Introduction In this paper the author present an algorithm based on a time-delay

5、 neural network with spatio-temporal receptive fields and adaptable time delays(ATDNN) for image sequence analysis. The aim is to classify objects on temporal sequences of grayscale images and to estimate their motion behavior.The ATDNN algorithmThe ATDNN algorithmThe training algorithm As the adapt

6、ation procedure for and is based on a gradient descent method, next step is to define a network output k(, ) for real-valued and , which is achieved by bilinear interpolation. Error functionApplications A. Simple Synthetic Image Sequence The network parameters are Rx = Ry = 7, Rt = 5, NRF = 2, Dx =

7、Dy = 4, Rt = Rh = 2.ApplicationsApplications In the test, The error rate on the test set of 1.1% for above experiments. In this application, the error is 1.2% for training run 1, 1.1% for run 2, and 1.2% for run 3. Applications B. Recognition of Pedestrians In this section the purpose is to examine

8、a rather complex application: the recognition of pedestrians on image sequences based on the characteristic criss-cross motion of their legs. The aim is to distinguish between pedestrian and nonpedestrian patterns.ApplicationsApplications Rx = Ry = 9, Rt = 2, NRF = 2, Rh = 2, Dx = Dy = 5.Application

9、sIntroduction In this paper, the author presents a novel concept for simultaneous shape estimation and motion analysis based on a TDNN(time-delay neural network) architecture with adaptable spatio-temporal receptive fields.tBxytTime delay network structure with adaptable spatio-temporal receptive fieldsError function A k = c (current class c), A1?k = 0 k cThe network parameters are as follows:And the error rate on the test set is 1.1%.ExperimentsThe ATDNN algorithm