Abstract：Noise pollutions and randomness of asphalt pavement image are strong. The cracks cannot be better identified from a large number of similar noise cracks in the traditional filtering and edge detection of the pavement image. To solve the noise pollution problem, a new method of fracture recognition based on neural network is proposed by using BP neural network for learning and fault tolerance. Firstly, the asphalt pavement enhanced image is divided into 32 × 32 small square regions. Secondly, the image parameters of the small squares are extracted and its neighborhood squares are used for neural network training. Finally, the small trained images are divided into two kinds, on has cracks and another has no cracks, thus the initial extraction of asphalt pavement cracks is achieved. The experimental results show that the recognition rate of asphalt pavement cracks is above90%. It is a feasible method to meet the requirement of pavement crack recognition.

Keyword：asphalt pavement;homomorphic filtering;neural network;crack identification;