3-2-3. Inception-ResNet-v2:
Inception-ResNet-v2 [18] is a deep convolutional neural network
architecture . It combines the Inception architecture with residual
connections from ResNet to form a hybrid architecture that has both the
ability to learn hierarchical features and the ability to alleviate the
vanishing gradient problem in deep networks. The architecture has been
used for various computer vision tasks, including image classification
and object detection.
Simulation and results:
Data is gathered from the brain tumor MRI image database by Ahmad Hamada
[19]. This data set, which contains 3,000 actual brain scans
produced by radiologists using data from patients with malignancies, is
open to the public. There are lots of good data sources for machine
learning tournaments, such as Kaggle, which has a lot of datasets
available. Our data is split into two sections: training and validation.
The model includes 2550 images for training, 224 images for validation,
and 226 images for testing With a batch size of 64, we trained the
models for 16 epochs. On the Google Colab Pro platform, this experiment
was conducted using Python’s TensorFlow and Keras libraries. Our
suggested model was 100 percent accurate in our training data, and in
our validation dataset, it was 99 percent accurate.
We employed the transfer learning approach to compare our convolution
model to the pre-trained ResNet-v2-152, Inception-v3, and
Inception-Resnet-v2 models on a dataset. ResNet-v2-152 has a 100 percent
accuracy in training data and a 99 percent accuracy in validation data,
Inception-v3 has a 99 percent accuracy in training data and a 96 percent
accuracy in validation data, and Inception-Resnet-v2 has a 97 percent
accuracy in training data and a 99 percent accuracy in validation data.
In validation data, it achieved a 98 percent success rate. On Figure
3-8, we can see how accurate our suggested models were during the test
and validation phases during the iterative process.