Fire Classifier (CNNv2)

This model uses an architecture that is nearly identical to that of the previous release, except we now use more pooling layers and adjusted the numbers of filters and nodes to make it more memory efficient.

The entire model code is at the bottom of the notebook here. It accomplished a validation accuracy of about 83% (better than the previous CNN model).

Fire Classifier (CNNv1)

• This model is the first iteration of using convolutional neural networks to solve this problem.
• This is the first model which has been trained on a balanced dataset of fire/normal images.
• The entire model code is at the bottom of the notebook here.

Fire Classifier Model (MLPv2)

• This model fixes an error in the last layer, replacing the incorrectly used softmax activation function with sigmoid
• Reduces memory size by saving only the model weights and biases (aka "parameters") to an H5 file, and the information about the model architecture (i.e. the "layers" of the neural network) to JSON

Fire Classifier Model (MLPv1)

This is a MLP model trained on the image dataset here.

The notebook is here.