Title : Deep Convolutional Neural Network (CNN) for threedimensional (3-D) objects classification using phase-only digital holographic information
Abstract:
A deep CNN-based binary classification of three-dimensional (3-D) objects for phase-only digital holographic information has been presented. The 3-D objects considered for the binary classification task are ‘triangle-square’, ‘circle-square’, ‘square-triangle’, and ‘triangle-circle’. The 3-D object ‘triangle-square’ is considered for the TRUE class and the remaining 3-D objects ‘circle-square’, ‘square-circle’, and ‘triangle-circle’ are considered for the FALSE class. The 3-D object volume ‘triangle-square’ was constructed in such a way that the feature triangle was considered in the first plane and the feature square was considered in the second plane. Each plane is separated by various distances , and respectively. The remaining three 3-D objects were constructed similarly except that the different features were considered in the first and second planes respectively. The digital holograms of 3-D objects have been formed using the two-step phase-shifting digital holography (PSDH) technique and computationally post-processed to obtain phase-only digital holographic data. The phase-only image dataset was prepared by performing a rotation of on each phase image. Then the training of the deep CNN was performed on a phase-only image dataset consisting of 2880 images to produce the results. The results such as the loss and accuracy curves, confusion matrix, Receiver Operating Characteristic (ROC), and precision-recall characteristic are shown for the confirmation of the work. The classification of phase images implies the classification of 3-D objects using deep CNN
