The invention provides a deep learning-based power grid equipment state monitoring device. The device comprises a video collection unit, a power grid equipment detection unit, a display unit and a storage unit. The industrial camera of the video collection unit is used for transmitting the collected video information to a video decoder through a network cable. Meanwhile, the collected video information is transmitted to the storage unit for data backup through the network cable. The video frames of a video are extracted, and then an image library exclusively used for identifying the state of power grid equipment is constructed. The power grid equipment detection unit is used for transmitting the state data of power grid equipment, stored in the storage unit, to a deep learning training module for the deep learning modeling operation. The deep learning training module is used for judging and classifying the state of power grid equipment according to a module established by the deep learning training model. Meanwhile, the judging and classifying result is displayed and stored. The device effectively relieves the manpower monitoring pressure, and achieves the real intelligent monitoring.