Finding location-specific aggregated air quality index with smartphone images using deep convolutional neural network

Abstract

Although smartphones have already become the de facto tool for environmental health research for their ubiquity and portability, utilizing them in finding location specific aggregated air quality index based on PM2.5 concentration is little ex plored in the literature to date. In this paper, therefore, we vigorously analyze the difficulties of predicting location-specific PM2.5 concentration from photos cap tured by smartphone cameras. Here, we particularly focus on Dhaka, the capital of Bangladesh, considering its very high level of air pollution exposure to a huge number of its dwellers. In our research, we develop a Deep Convolutional Neural Network (DCNN) and train it using more than a thousand outdoor photos cap tured and labeled by us. We capture the photos at various locations in Dhaka, Bangladesh, and label them based on PM2.5 concentration data extracted from the local US consulate as computed by the NowCast algorithm. During training with the dataset, our model learns a correlation index through supervised learning, which improves the model’s ability to act as a Picture-based Predictor of PM2.5 Concen tration (PPPC) making it capable of detecting comparable daily aggregated AQI index from a photo captured by a smartphone. Here, the computation necessary in our model is comparatively resource-efficient, as our model subsumes a much smaller number of parameters compared to most of the other alternatives. Moreover, our experimental results show that our model exhibits more robustness, for location specific PM2.5 prediction than existing state-of-the-art models such as ViT (Vision Transformer) and INN (Involutional Neural Network) as well as other popular mod els that are created based on CNN, such as VGG19, ResNet50, or MobileNetV2.