YaliWang2019 / MS_ForestFireDetection

This is an archive of my graduate project. This project analyzes the feasibility of deploying machine learning models on a drone to detect forest fires.

MS_ForestFireDetection

This is an archive of my graduate project. This project analyzes the feasibility of deploying machine learning models on a drone to detect forest fires.

Explanation for each folder:

D2Go:

The D2Go model is developed based on Faster_RCNN.

1. D2GO_RCNN_PoygonSize.ipynb: Script for detecting fires in an image and showing the bounding box sizes.
2. D2GO_RCNN_Video.ipynb: Script for detecting fires in the frames of a video.
3. forest-fires-t.json and forest-fires-v.json: Annotated COCO format JSON datasets for training the D2Go model.

PyTorch_CNN:

The convolutional neural network models developed using PyTorch framework.

1. datasheets: All the datasheets recorded during the tests.
2. models:

• pytorch_cnn_add1.pth: 4 convolutional2D layers + 4 maxpooling2D layers + 3 dense layers;
• pytorch_cnn_add2.pth: 5 convolutional2D layers + 5 maxpooling2D layers + 3 dense layers;
• pytorch_cnn_origin.pth: 3 convolutional2D layers + 3 maxpooling2D layers + 3 dense layers;
• pytorch_cnn_rm1.pth: 2 convolutional2D layers + 2 maxpooling2D layers + 3 dense layers;
• pytorch_cnn_rm2.pth: 1 convolutional2D layers + 1 maxpooling2D layers + 3 dense layers;
• pytorch_cnn_sig1.pth: 3 convolutional2D layers + 3 maxpooling2D layers + 1 dense layers;
• pytorch_cnn_add2.pth: 3 convolutional2D layers + 3 maxpooling2D layers + 2 dense layers;

3. scripts: The training scripts of the models mentioned above and their energy consumption measurement scripts.

TensorFlow_AE:

The Autoencoder models developed using TensorFlow framework.

1. datasheets: All the datasheets recorded during the tests.
2. models:

• ae_1En_1De.h5: 1 group of convolutional2D layer and maxpooling2D layer in Encoder + 1 group of convolutional2D layer and upsampling2D layer in Decoder;
• ae_2En_2De.h5: 2 groups of convolutional2D layer and maxpooling2D layer in Encoder + 2 groups of convolutional2D layer and upsampling2D layer in Decoder;
• ae_3En_3De.h5: 3 groups of convolutional2D layer and maxpooling2D layer in Encoder + 3 groups of convolutional2D layer and upsampling2D layer in Decoder;

3. scripts: The training script of the original model (ae_3En_3De.h5) and its energy consumption measurement script.

TensorFlow_CNN:

The convolutional neural network models developed using TensorFlow framework.

1. datasheets: All the datasheets recorded during the tests.
2. models:

• cnn_or.pth: 3 convolutional2D layers + 3 maxpooling2D layers + 3 dense layers;
• cnn_rm1ConvMaxp.pth: 2 convolutional2D layers + 2 maxpooling2D layers + 3 dense layers;
• cnn_rm2ConvMaxp.pth: 1 convolutional2D layers + 1 maxpooling2D layers + 3 dense layers;
• cnn_sig1_dense.pth: 3 convolutional2D layers + 3 maxpooling2D layers + 1 dense layers;
• cnn_sig2_dense.pth: 3 convolutional2D layers + 3 maxpooling2D layers + 2 dense layers;

3. scripts: The training scripts of the original model (cnn_or.pth) and its energy consumption measurement script.

TensorFlow_DBN:

The deep belief network models developed using TensorFlow models.

1. datasheets: All the datasheets recorded during the tests.
2. models: These four models can be accessed from this google drive link since each of them is larger than 100 MB which makes GitHub unhappy to upload them here.

• dbn_1RBM.joblib: 1 Restricted Boltzmann Machine layer;
• dbn_2RBM.joblib: 2 Restricted Boltzmann Machine layers;
• dbn_3RBM.joblib: 3 Restricted Boltzmann Machine layers;
• dbn_4RBM.joblib: 4 Restricted Boltzmann Machine layers;

3. scripts: The training scripts of the original model (dbn_2RBM.joblib) and its energy consumption measurement script.

TensorFlow_UNet:

The U-Net models developed using TensorFlow framework.

1. datasheets: All the datasheets recorded during the tests.
2. models:

• unet_1En_1De.h5: 1 group of convolutional2D layer and maxpooling2D layer in Encoder + 1 group of convolutional2D layer, upsampling2D layer, and concatenate layer in Decoder;
• unet_2En_2De.h5: 2 groups of convolutional2D layer and maxpooling2D layer in Encoder + 2 groups of convolutional2D layer, upsampling2D layer, and concatenate layer in Decoder;
• unet_3En_3De.h5: 3 groups of convolutional2D layer and maxpooling2D layer in Encoder + 3 groups of convolutional2D layer, upsampling2D layer, and concatenate layer in Decoder;

3. scripts: The training scripts of the original model (unet_3En_3De.h5) and its energy consumption measurement script.