Classification of incident related images using machine learning

This project aimed to develop a machine learning framework for incident recognition in images using a subset of the Incidents1M dataset [1]. We used a minimum of 400 samples per category from the 12 incident categories. Preprocessing and data augmentation were performed before extracting descriptors from the images using ResNet, a popular deep learning architecture for image classification. K-nearest neighbors (KNN) was used for classification, and k-fold cross-validation was performed to assess the generalization capabilities of the model. The performance of the model was evaluated using metrics such as accuracy, f-score, and precision. The results demonstrate the effectiveness of the proposed framework for incident recognition in images and its potential relevance for relief organizations.