Particulate matter (PM) is one of the most common air pollutants and may cause many severe diseases. In particularly, PM with diameters less than 2.5 micrometers (PM2.5) is more harmful to human health than other air pollutants because it can penetrate deeply into the lungs and damage human respiratory system. An efficient PM2.5 monitoring is of great benefit for human health and air pollution control. In this paper, we described about our collected dataset, experiment conducted with the depth map, transmission map and RGB images before proposing an image-based method for PM2.5 analysis. The proposed method takes RGB images as the input and outputs the estimated concentration of PM2.5 based on the visual contents of input images using a deep convolutional neural network. The adopted network is a VGG-16 model pre-trained using ImageNet dataset and a transfer learning strategy is applied to fine tune the weights of the fully connected layers for PM2.5 analysis. We evaluated our method using one real-world datasets and the experimental results demonstrate that the proposed method can be used for image-based PM2.5 analysis and can potentially provide an efficient and affordable way for air pollution monitoring.