Numerous ground-based and space-based high-precision photometric survey have brought about the era of big data in time-domain astronomy. An efficient classification of stellar variability is fundamental for the detection of exoplanets using the transit and microlensing methods, studies of stellar physics, high energy transients, and multi-messenger astronomy. However, the current classification pipelines for light curves are neither general nor highly automated and are thus not directly applicable in various time-domain projects. We will develop a general deep learning-based photometric classification method that can be applied to a variety of ground-based and space-based multi-timescale photometric time-domain surveys, enabling important discoveries in exoplanet detection, stellar physics and other time-domain science.