Fig. 5
DNN classifier architecture. A conceptual representation of the proposed neural network architecture for processing sequential speckle images using 2D ConvLSTM layers. The input to the network consists of stacks of sequential speckle signal, preserving the spatial and temporal information. Each stack is fed into a ConvLSTM cell, which integrates convolutional operations with LSTM-based temporal processing to retain the two-dimensional spatial structure of the input. The ConvLSTM layer outputs feature maps that are subsequently processed by fully connected (dense) layers for the classification task.
