Abstract: The real-time state tracking and position and attitude quantitative monitoring of hydraulic support side guard is one of the key technologies to realize safe efficient and intelligent mining in underground fully mechanized mining face. Aiming at the problems of low reliability and lack of quantitative monitoring ability of the existing hydraulic support upper guard plate condition monitoring methods, a non-contact upper guard plate motion condition monitoring method based on deep neural network is proposed in this paper. This method realizes the spatial position detection of the key points of the upper guard board through the convolution and deconvolution network, and then uses the feedforward neural network to convert the spatial motion trajectory of the key points of the upper guard board into the expansion angle of the upper guard board, so as to realize the quantification of the condition monitoring of the upper guard board. The results show that the average detection error of the key point spatial position of the upper guard board is less than 2 pixels, the average quantization error of the position and attitude angle of the upper guard board is less than 3 degrees, and the processing speed of the algorithm is more than 60 frames per second, which has good monitoring performance.