Page 21 - Kỷ yếu hội thảo khoa học lần thứ 12 - Công nghệ thông tin và Ứng dụng trong các lĩnh vực (CITA 2023)
P. 21
Van Vy and Hyungchul Yoon 5
well-suited for large-scale image and video processing tasks. Pooling is typically
applied after convolutional layers in convolutional neural networks to reduce the
dimensionality of the feature maps while retaining important information. It is used to
downsample the output of the previous layer by summarizing its inputs in a smaller,
more condensed form. During the training process, the weights of the convolution
kernel and bias are updated using backpropagation.
Fig. 3. The architecture of AECWT-3DR-Net
The AECWT-3DR-Net utilizes (5 5) pixel kernels with (1 1) strides in its
convolutional layer. Following each convolutional layer is a normalization layer, and
then a max pooling layer with (3 3) size and (1 1) strides. The branches are joined
together at the end of each branch, and the combined data passes through four
convolutional layers with (3 3) and (2 2) sizes, along with other layers. Finally, the
network utilizes global max pooling to synthesize precise information for estimating
x, y, and z coordinates. The hyperparameters used for tuning were epoch = 500,
learning rate = 0.01, batch size = 32, and the optimizer was Adam. Our experimental
programs were developed using the Python programming language on a computer
with an Intel(R) Core(TM) i7-10700 CPU, 2.90 2.90 GHz processor, 16.0 GB of
RAM, GPU NVIDIA GeForce RTX 3070.
ISBN: 978-604-80-8083-9 CITA 2023