CMU dataset

Particulate pollution, PM2.5, is an important issue in Asian countries due to its health hazards. Knowing the measurement of PM2.5 concentration value is important to plan outdoor activities. Due to the less number of government run Air Quality Monitoring Stations, other options for getting location-specific PM2.5 concentration values are sought. This paper proposes to estimate the PM2.5 concentration value using photo image processing. This research aims to improve the efficacy and reduce the computational complexity of the PM2.5 concentration value estimation process. The proposed Efficient PM2.5 estimation framework uses EfficientNet-B1 and BiLSTM to estimate the PM2.5 concentration value. The Met-EfficientNet-B1-BiLSTM has been designed and implemented to incorporate the Meteorological features temperature, wind speed and humidity to further improve the estimation accuracy. The EfficientNet-B1 neural network is applied in the image feature vector extraction process. The BiLSTM is used for the regression of these image features with PM2.5 concentration values to get the estimated PM2.5 concentration values. The optimum EfficientNet variant for a smaller dataset of images required for PM2.5 concentration value estimation is figured out to be EfficientNet-B1 with 240 x 240 pixels resolution. A dataset containing HDR and non-HDR images created for this study is used for comparing the types of images that facilitate the feature extraction process and accuracy of PM2.5 concentration estimation. The proposed Efficient PM2.5 estimation framework, can reduce computational complexity and outperform the ResNet-18-LSTM by improving efficacy by 5.75\% in MAE and 11.43\% in SMAPE matrices. The proposed Efficient PM2.5 estimation framework demonstrates that the mobile image can be used for PM2.5 estimation.