The document presents a deep learning-based fire and smoke detection system. It proposes a convolutional neural network (CNN) architecture for feature extraction from images to classify scenes as containing fire, smoke, or neither. The CNN was trained on a dataset of over 5000 images across three classes and achieved 96.3% accuracy for fire/smoke detection over 100 epochs. Future work could focus on improving model accuracy through additional data and real-time recognition capabilities. The system shows potential for computer vision-based fire detection as an alternative to traditional sensor-based methods.