In this paper, we propose a combined deep learning model of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) to predict PM2.5 concentration in Dalat, a famous tourism city in Vietnam. CNN is utilized to extract spatial features of environmental images. The remaining component, LSTM, is designed for the training of spatial and temporal features to predict PM2.5 over a specific period. In our model, training data consists of environmental images and PM2.5 concentrations collected from lifelog on different roads in Dalat City, Vietnam. Experimental results show that the proposed combined CNN and LSTM model using both spatial and temporal features achieves superior PM2.5 concentration prediction accuracy compared to the traditional LSTM approach.