Deep Network Designer app 

Deep Learning Toolbox is a platform for developing and deploying deep neural networks, including algorithms, pre-trained models, and apps.

We can create, evaluate, and train networks graphically with the Deep Network Designer App. The Experiment Manager tool assists us in managing many deep learning experiments, keeping track of training parameters, analyzing outcomes, and compare code from several experiments. Layer activations may be seen, and training progress can be viewed graphically. This blog demonstrates how to use Deep Network Designer to classify a new collection of pictures using various network blocks.

Follow the steps below to do classification analysis in the Deep Network learner app.

• In the apps tab in the Machine Learning and Deep learning section, click on the Deep Network designer Section.
• Creation of Layers
• Importing of Data
• Training of Model
• Evaluate the performance of the model.
• Exporting a model.

Information about the dataset.

The soil type's data set is made up of 156 images. Each image has different pixels values and comes with a label that indicates which class it represents. The data set is divided into five categories.

Deep Network Learner App

Deep Network designer

We can create a network from scratch or any pre-trained network and network from the workspace for making a deep neural network. We would generate a network from scratch, so select a blank network.

Below we can see the default window after clicking on the blank network.

• Section Layers: To create a deep neural network.
• Properties of Layers Section: We can change the properties of layers as per the requirements.
• Different Options to Train Network Section: We can create a network, import data, and train the network.

Preparation of Network

We can add layers and change the properties as per our requirements. Below, we all can see the prepared network.

Import Data

Click Import Data > Import Image Data on the Data tab to import the data into Deep Network Designer. Import Image Data's dialogue box appears—select Folder from the Data source drop-down menu. Select the extracted Data folder from the Browse menu. You may also separate the validation data from within the app using the dialogue box. Seventy percent of the data should be used for training and 30 percent for validation. Define the augmentation operations that will be applied to the training images. Apply a random rescaling from the range [1,2] for this example.

Import the data into Deep Network Designer by clicking the Import button.

In the Data tab of Deep Network Designer, you can see how the training and validation data are distributed visually. There are five classes in this data collection, as we can see. Random observations from each class are also available to view.

Train Network

Click Train on the Training tab to train the network with the default settings. Click Program Options and select the parameters to train with if you want more control over the training. Large data sets benefit from the default training options. Use reduced mini-batch sizes and validation frequencies for small data sets. See trainingOptions for additional details on training options selection.

lose and then Train to train the network with the selected training options.

Deep Network Designer allows you to visualize and track your training progress. You can then change the training options and, if necessary, retrain the network.

Conclusion

• Preparing deep networks, data augmentation, specifying validation schemes, training models, and evaluating outcomes will all be straightforward and quick with this app.
• Results are optimum, and retraining is possible with few clicks. We can export the model to the workspace or produce the MATLAB code to reuse it with new data or do a programmatic classification.

Note:

1. We can adjust other parameters of the created convolutional neural network.
2. Because of the differences in initial conditions and sampling, training numerous times will yield different results.