Abstract: In the process of selecting millimeter/submillimeter wave radio astronomical telescope sites, it is necessary to design an all-sky camera system for use in the field environment in order to fully understand the cloud amount information of candidate observatory sites. Therefore, according to the characteristics of submillimeter wave radio telescopes and the specific conditions of the field sites, this scheme innovatively use the planetary camera and embedded microcontroller to develop a full-time all-sky camera, which can operate in the field for a long time using solar energy, and the most important feature is that it can achieve unmanned and autonomous operation. In the data processing part, the deep learning neural network algorithm is also innovatively used to extract data feature values, establish machine learning model library, and automatically count cloud information of the site, which is more efficient and simpler than manual and general image processing algorithms. These studies provide important references for more comprehensive evaluation of millimeter/submillimeter wave radio observatory sites.