This study aims to develop a convolutional neural network (CNN) algorithm for fire disaster management in commercial complexes in Nigeria, addressing the limitations of traditional fire detection systems.
Secondary data were utilized due to the absence of observational data, with images collected from Google Images. The dataset comprised two classes: fire and non-fire images. Google Colaboratory was employed for algorithm development, leveraging the advantages of its cloud-based environment and integration with deep learning frameworks like TensorFlow. Dataset augmentation and pre-processing were performed to mitigate overfitting and enhance model generalization. The CNN architecture consisted of multiple convolutional and fully connected layers, designed to maximize feature extraction from input images.
The CNN model demonstrated a good performance, achieving an accuracy of 73% on a separate test dataset. Metrics like precision, F1-score and recall indicated satisfactory performance in classifying fire images. The model exhibited potential in accurately detecting fire instances in commercial complex environments.
This research contributes to the advancement of the fire disaster management narrative in the Nigerian local context by proposing a computer vision-based approach using region-based convolutional neural network for fire disaster management in commercial complexes in Nigeria. The dataset augmentation technique used showcases the rigour and reflects the inadequacy of such datasets in Nigeria. Future implementation of this algorithm will involve significant investment in the creation of an image or video database for the algorithm to work on real-life scenarios. Its implementation could enhance safety measures in commercial complexes, mitigating economic losses and safeguarding lives and properties.
