DenseNet Classifier: Detecting Regions of Interest in Synthetic Signals

This example demonstrates how to use DeepPeak’s DenseNet classifier to identify regions of interest (ROIs) in synthetic 1D signals containing Gaussian peaks.

We will: - Generate a dataset of noisy signals with random Gaussian peaks - Build and train a DenseNet classifier to detect ROIs - Visualize the training process and model predictions

Note

This example is fully reproducible and suitable for Sphinx-Gallery documentation.

Imports and reproducibility

import numpy as np

from DeepPeak.machine_learning.classifier import DenseNet
from DeepPeak.signals import Kernel, SignalDatasetGenerator

np.random.seed(42)

Generate synthetic dataset

NUM_PEAKS = 3
SEQUENCE_LENGTH = 200

generator = SignalDatasetGenerator(n_samples=100, sequence_length=SEQUENCE_LENGTH)

dataset = generator.generate(
    signal_type=Kernel.GAUSSIAN,
    n_peaks=(1, NUM_PEAKS),
    amplitude=(1, 20),
    position=(0.1, 0.9),
    width=(0.03, 0.05),
    noise_std=0.1,
    categorical_peak_count=False,
    compute_region_of_interest=True,
)

Visualize a few example signals and their regions of interest

dataset.plot(number_of_samples=3)

Build and summarize the DenseNet classifier

dense_net = DenseNet(
    sequence_length=SEQUENCE_LENGTH,
    filters=(32, 64, 128),
    dilation_rates=(1, 2, 4),
    kernel_size=3,
    optimizer="adam",
    loss="binary_crossentropy",
    metrics=["accuracy"],
)
dense_net.build()
dense_net.summary()

Model: "DenseNetDetector"
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ input (InputLayer)              │ (None, 200, 1)         │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv_0 (Conv1D)                 │ (None, 200, 32)        │           128 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv_1 (Conv1D)                 │ (None, 200, 64)        │         6,208 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv_2 (Conv1D)                 │ (None, 200, 128)       │        24,704 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ ROI (Conv1D)                    │ (None, 200, 1)         │           129 │
└─────────────────────────────────┴────────────────────────┴───────────────┘
 Total params: 31,169 (121.75 KB)
 Trainable params: 31,169 (121.75 KB)
 Non-trainable params: 0 (0.00 B)

Train the classifier

history = dense_net.fit(
    dataset.signals,
    dataset.region_of_interest,
    validation_split=0.2,
    epochs=20,
    batch_size=64,
)

Epoch 1/20

1/2 ━━━━━━━━━━━━━━━━━━━━ 1s 1s/step - accuracy: 0.9453 - loss: 0.6770
2/2 ━━━━━━━━━━━━━━━━━━━━ 1s 199ms/step - accuracy: 0.9474 - loss: 0.6745 - val_accuracy: 0.9613 - val_loss: 0.6526
Epoch 2/20

1/2 ━━━━━━━━━━━━━━━━━━━━ 0s 37ms/step - accuracy: 0.9502 - loss: 0.6451
2/2 ━━━━━━━━━━━━━━━━━━━━ 0s 48ms/step - accuracy: 0.9496 - loss: 0.6391 - val_accuracy: 0.9613 - val_loss: 0.6211
Epoch 3/20

1/2 ━━━━━━━━━━━━━━━━━━━━ 0s 35ms/step - accuracy: 0.9489 - loss: 0.6105
2/2 ━━━━━━━━━━━━━━━━━━━━ 0s 53ms/step - accuracy: 0.9496 - loss: 0.6096 - val_accuracy: 0.9613 - val_loss: 0.5923
Epoch 4/20

1/2 ━━━━━━━━━━━━━━━━━━━━ 0s 36ms/step - accuracy: 0.9519 - loss: 0.5785
2/2 ━━━━━━━━━━━━━━━━━━━━ 0s 48ms/step - accuracy: 0.9496 - loss: 0.5748 - val_accuracy: 0.9677 - val_loss: 0.5634
Epoch 5/20

1/2 ━━━━━━━━━━━━━━━━━━━━ 0s 36ms/step - accuracy: 0.9584 - loss: 0.5444
2/2 ━━━━━━━━━━━━━━━━━━━━ 0s 48ms/step - accuracy: 0.9596 - loss: 0.5410 - val_accuracy: 0.9613 - val_loss: 0.5250
Epoch 6/20

1/2 ━━━━━━━━━━━━━━━━━━━━ 0s 35ms/step - accuracy: 0.9486 - loss: 0.5045
2/2 ━━━━━━━━━━━━━━━━━━━━ 0s 48ms/step - accuracy: 0.9496 - loss: 0.5039 - val_accuracy: 0.9613 - val_loss: 0.4859
Epoch 7/20

1/2 ━━━━━━━━━━━━━━━━━━━━ 0s 36ms/step - accuracy: 0.9491 - loss: 0.4698
2/2 ━━━━━━━━━━━━━━━━━━━━ 0s 48ms/step - accuracy: 0.9510 - loss: 0.4663 - val_accuracy: 0.9790 - val_loss: 0.4432
Epoch 8/20

1/2 ━━━━━━━━━━━━━━━━━━━━ 0s 36ms/step - accuracy: 0.9741 - loss: 0.4254
2/2 ━━━━━━━━━━━━━━━━━━━━ 0s 53ms/step - accuracy: 0.9731 - loss: 0.4209 - val_accuracy: 0.9760 - val_loss: 0.3922
Epoch 9/20

1/2 ━━━━━━━━━━━━━━━━━━━━ 0s 36ms/step - accuracy: 0.9702 - loss: 0.3761
2/2 ━━━━━━━━━━━━━━━━━━━━ 0s 48ms/step - accuracy: 0.9664 - loss: 0.3731 - val_accuracy: 0.9783 - val_loss: 0.3395
Epoch 10/20

1/2 ━━━━━━━━━━━━━━━━━━━━ 0s 37ms/step - accuracy: 0.9715 - loss: 0.3277
2/2 ━━━━━━━━━━━━━━━━━━━━ 0s 53ms/step - accuracy: 0.9700 - loss: 0.3233 - val_accuracy: 0.9790 - val_loss: 0.2865
Epoch 11/20

1/2 ━━━━━━━━━━━━━━━━━━━━ 0s 35ms/step - accuracy: 0.9688 - loss: 0.2819
2/2 ━━━━━━━━━━━━━━━━━━━━ 0s 48ms/step - accuracy: 0.9706 - loss: 0.2741 - val_accuracy: 0.9772 - val_loss: 0.2355
Epoch 12/20

1/2 ━━━━━━━━━━━━━━━━━━━━ 0s 35ms/step - accuracy: 0.9697 - loss: 0.2328
2/2 ━━━━━━━━━━━━━━━━━━━━ 0s 48ms/step - accuracy: 0.9684 - loss: 0.2298 - val_accuracy: 0.9755 - val_loss: 0.1904
Epoch 13/20

1/2 ━━━━━━━━━━━━━━━━━━━━ 0s 35ms/step - accuracy: 0.9681 - loss: 0.1935
2/2 ━━━━━━━━━━━━━━━━━━━━ 0s 48ms/step - accuracy: 0.9681 - loss: 0.1894 - val_accuracy: 0.9790 - val_loss: 0.1514
Epoch 14/20

1/2 ━━━━━━━━━━━━━━━━━━━━ 0s 36ms/step - accuracy: 0.9702 - loss: 0.1559
2/2 ━━━━━━━━━━━━━━━━━━━━ 0s 48ms/step - accuracy: 0.9702 - loss: 0.1549 - val_accuracy: 0.9788 - val_loss: 0.1216
Epoch 15/20

1/2 ━━━━━━━━━━━━━━━━━━━━ 0s 35ms/step - accuracy: 0.9701 - loss: 0.1326
2/2 ━━━━━━━━━━━━━━━━━━━━ 0s 49ms/step - accuracy: 0.9700 - loss: 0.1288 - val_accuracy: 0.9778 - val_loss: 0.1002
Epoch 16/20

1/2 ━━━━━━━━━━━━━━━━━━━━ 0s 36ms/step - accuracy: 0.9681 - loss: 0.1152
2/2 ━━━━━━━━━━━━━━━━━━━━ 0s 47ms/step - accuracy: 0.9697 - loss: 0.1115 - val_accuracy: 0.9783 - val_loss: 0.0853
Epoch 17/20

1/2 ━━━━━━━━━━━━━━━━━━━━ 0s 36ms/step - accuracy: 0.9688 - loss: 0.0988
2/2 ━━━━━━━━━━━━━━━━━━━━ 0s 54ms/step - accuracy: 0.9692 - loss: 0.0978 - val_accuracy: 0.9772 - val_loss: 0.0784
Epoch 18/20

1/2 ━━━━━━━━━━━━━━━━━━━━ 0s 36ms/step - accuracy: 0.9698 - loss: 0.0934
2/2 ━━━━━━━━━━━━━━━━━━━━ 0s 48ms/step - accuracy: 0.9693 - loss: 0.0918 - val_accuracy: 0.9788 - val_loss: 0.0702
Epoch 19/20

1/2 ━━━━━━━━━━━━━━━━━━━━ 0s 35ms/step - accuracy: 0.9686 - loss: 0.0862
2/2 ━━━━━━━━━━━━━━━━━━━━ 0s 48ms/step - accuracy: 0.9677 - loss: 0.0857 - val_accuracy: 0.9785 - val_loss: 0.0666
Epoch 20/20

1/2 ━━━━━━━━━━━━━━━━━━━━ 0s 35ms/step - accuracy: 0.9683 - loss: 0.0798
2/2 ━━━━━━━━━━━━━━━━━━━━ 0s 48ms/step - accuracy: 0.9682 - loss: 0.0821 - val_accuracy: 0.9785 - val_loss: 0.0647

Plot training history

dense_net.plot_model_history()

Predict and visualize on a test signal

dense_net.plot_prediction(signal=dataset.signals[0:1, :], threshold=0.4)

<Figure size 1200x500 with 1 Axes>