Abstract: In order to recognize caving coal and rock traits in fully mechanized caving face, an identification method based on continuous wavelet transform and improved singular value decomposition (SVD) was proposed. The SVD method based on unilateral Jacobi is used to decompose wavelet coefficient matrix, so as to get singular value vectors corresponding to the column vector position of the wavelet coefficient matrix. The singular value vectors are used as input vector of neural network to identify two conditions of falling coal and falling rock. Field test results show that the singular value vectors acquired by the method based on continuous wavelet transform and SVD can be used to identify coal and rock, but the singular value vectors acquired by the method based on continuous wavelet transform and improved SVD has higher identification rate.