Pollution Level Estimation Through Image Analysis

Abstract

Climate change is one of the hardest problems humanity will have to face in the next century. Data analysis and computer vision are two powerful tools that can help us perform tasks that would usually take more time and resources to finish. Therefore, monitoring air quality, especially in developing countries should be the first step to save the environment. Measurement of air quality is a task that, currently needs the help of specialized equipment and infrastructure. These equipments are either very costly or require skills to operate or both making it difficult to provide air quality information at remote locations or at desired spots even in cities. In this study, we have tried to measure the air quality through images which can be taken using a normal camera. For this purpose, we used deep learning techniques, where we trained ResNet18 using a public image database. Performance is evaluated by plotting confusion matrix. We also measure precision, recall, F1-score and accuracy. Results are analyzed by plotting ROC curve and precision-recall curve.