Artificial intelligence is often explained with abstract examples: cats and dogs, handwritten digits, or large public image datasets. These examples are useful, but they do not show how AI can be taught in a technical radar environment. In this short video, we demonstrate how FreeScopes AI turns real radar measurements into a practical machine-learning exercise: students collect and label data from three radar targets, train a neural network, and use it to classify the measured objects.
With FreeScopes AI, we take a different approach. Students work with radar and simulator data, collect measurements themselves, label the data, build a neural-network workflow visually, train a model, and then use the trained model to classify new radar measurements.
The video below shows a simple example from such a training workflow. Three radar targets are used: a big sphere, a small sphere, and a corner reflector. The task is to train a neural network to recognise which target produced a given radar measurement.
From radar measurements to labelled training data
The first step is not the neural network itself. The first step is data.
In the experiment shown in the video, radar measurements were collected with SkyRadar's NextGen 8 GHz Pulse Radar for three different objects:
-
Big sphere rotating in place
-
Corner reflector rotating in place
-
Small sphere rotating in place
Each measurement was stored as an .hdf5 file. The important point is that each file was also given a label. The big sphere measurements were labelled as “Big Sphere”, the corner reflector measurements as “Corner Reflector”, and the small sphere measurements as “Small sphere”.
This is called supervised learning. The model learns from examples where the correct answer is already known. It does not discover the meaning of the objects by itself. It learns the relationship between a radar measurement pattern and a label provided during training.
For students, this is important: AI does not start with magic. It starts with measured data, careful labelling, and a clear training objective.
The network shown in the video
The screenshot from the video shows a simple neural-network workflow built in FreeScopes AI.
The workflow is:
Radar training data → AI Input Layer → AI Dense Layer → AI Dense Layer → AI Model Compile → AI Model Train → AI Model Save → AI Trained Model
The first Dense Layer has 128 neurons and uses ReLU activation. This layer learns combinations of signal values that help distinguish the three targets. It is the main feature-learning stage in this simple network.
The second Dense Layer has 3 neurons and uses Softmax activation. This is the classification layer. It has 3 neurons because the task has 3 classes: big sphere, corner reflector, and small sphere.
Softmax converts the network output into class probabilities. In practical terms, the model can say: this measurement most likely belongs to the big sphere, the small sphere, or the corner reflector.
In the video, the training block shows that the model has completed 10 of 10 epochs. The displayed accuracy is very high and the loss is low. This means that, on the training data, the model learned to separate the three labelled object classes well.
Why this is useful in training
The purpose of FreeScopes AI is not only to show that a neural network can classify objects. The purpose is to make the full AI workflow understandable.
Students can see and practice the complete chain:
-
-
They collect radar data.
-
They label the data.
-
They define a neural-network architecture.
-
They train the model.
-
They observe accuracy and loss.
-
They save the trained model.
-
They load the model again for later use.
-
They can test whether the model recognises new measurements correctly.
-
This makes AI tangible. Instead of treating machine learning as a black box, students can see the stages of the process and understand how each decision affects the result.
What the Dense Layers do
A Dense Layer is a fully connected neural-network layer. Each neuron receives all outputs from the previous layer and learns how strongly each input should influence the result.
In this example, the first Dense Layer combines the radar measurement values into internal features. These features may represent characteristic differences in the radar response of the big sphere, the small sphere, and the corner reflector.
The final Dense Layer then maps these learned features to the three output classes.
This is a good baseline architecture for a simple classification task. It is easy to understand, fast to train, and well suited for first experiments in AI-based radar classification.
Why FreeScopes AI matters
FreeScopes AI is designed for education in radar, signal processing, and AI. It gives students a visual, block-based environment for building AI workflows without first requiring them to write code.
This is important in military academies, universities, and technical training environments. Many learners need to understand AI concepts, but they are not necessarily software developers. They need to understand what a model does, how training data is prepared, why labels matter, and how performance is evaluated.
FreeScopes AI supports this by combining radar data with visual machine-learning workflows.
Students can learn:
-
-
how neural networks are structured
-
how data labels define the learning task
-
how Dense Layers can classify feature patterns
-
how training accuracy and loss are interpreted
-
why models must be tested on data they have not simply memorised
-
how saved models can be reused
-
where simple models are sufficient and where more advanced models are needed
-
From simple classification to more advanced radar AI
The example in the video uses a simple Dense network. This is deliberate. A baseline model is often the best starting point because it shows the essential idea clearly.
For more complex radar patterns, FreeScopes AI can also support more advanced workflows, for example Conv1D-based neural-network processing. A Conv1D layer uses trainable filters that scan sequential radar or signal data for local patterns. This can be useful when the important feature is not just a global amplitude pattern, but a local structure in a range profile, signal trace, or radar representation.
This allows training to progress from simple classification toward more complex radar AI topics, such as anomaly detection, robust radar perception, and jamming-aware analysis where supported by the available data and module scope.
The learning value
The key learning value of the exercise is that students see the complete AI training process in a radar context.
They do not only hear that artificial intelligence can classify objects. They see how the training data is created. They see how labels are assigned. They build the network themselves. They train it. They observe the results. They save the trained model.
This makes the relationship between radar measurements and AI classification transparent.
The result is not just a trained model. The result is a better understanding of how machine learning can be applied to radar data, where it is useful, and why the quality of the data and labels matters as much as the neural network itself.
Conclusion
The video demonstrates a compact but powerful FreeScopes AI workflow: three physical radar targets, labelled radar measurements, a simple neural network, and a trained classifier.
This is the core idea behind FreeScopes AI. It turns machine learning from an abstract topic into a practical radar-training exercise. Students learn how AI models are built, trained, evaluated, saved and reused — using radar data they can understand and measurements they can reproduce in the lab.
Stay Tuned
Stay connected with our ongoing publications on Electronic Warfare and Radar Technology.
