Abstract: To address some practical engineering issues such as unmanned coal mining and automatic lifing of shearer drum, a coal-rock recognition method based on the wavelet packet singular value(WPSV) and BP-neural networks(BPNN) is proposed. The new method uses wavelet packet singular value to build a feature vector and combines it with BP-neural networks for the coal-rock recognition. First of all, the torque signal of shearer drum cutting coal is collected by sensors, and using wavelet packet to decompose the torque signal of shearer drum, obtain the wavelet packet decomposition coefficient and form the wavelet packet decomposition coefficient reconstruction signal matrix; Then, Obtaining the major WPSV and forming feature vector by using singular value decomposition(SVD) to the coefficient matrix; Finally, Inputing the feature vector into the BPNN classifer to identify the coal-rock interface automatically, and the results were compared with the ones by traditional method. Experimental results demonstrate that this new method has the higher accuracy, which can represent the working state of shearer drum effectively.