Monitoring air pollution by deep features and extreme learning machine |
| |
Authors: | Seyyed Amirhosein Rahimi |
| |
Affiliation: | Department of Computer Science, School of Mathematics, Statistics and Computer Science, College of Science, University of Tehran, Tehran, Iran |
| |
Abstract: | Smartphones have different kinds of applications that help to promote health and care of humans. This paper proposes a practical and low-cost method for predicting air pollution which is applicable to the smartphones based on an image taken by their camera. To find the best method, in the first approach, some convenionalconventional feature extraction methods including wavelet transform, scale-invariant feature transform and histogram of oriented gradients are implemented. Then, to reduce the dimension of the extracted feature vectors, principal component analysis is employed. For classification of the obtained reduced feature vectors, multilayer perceptron is employed. In the second approach, the performance of convolutional neural network (CNN) in classifying the sky images in terms of air quality is investigated. In CNN, the fully connected classifier can be replaced by other classifiers such as extreme learning machine (ELM). The results illustrate that if the deep features obtained by CNN are fed to the ELM, an accuracy of 66.92% in predicting the level of air quality is achieved, which is higher than the results of other previous and conventional methods. |
| |
Keywords: | Human health air quality air pollution convolutional neural network extreme learning machine |
|
|