In order to solve the issue how Duct Cleaning Robot senses the unknown duct environment, an environmental feature extraction method based on visual information is proposed. This method detects the edge of environmental images using the wavelet modulus maximum(wmm), and then extracts the geometric features of those images by Hough transformation. Finally, feed forward Back-Propagation Neural Network(BPNN) is used to environmental status identification. The relevant experimental results are given to prove the effectiveness of the proposed method.