Abstract: In order to solve the problems of automatic identification of cave target, a target recognition algorithm was designed based on various features. Firstly, discussing the model of typical cave target, summarizing its main features, and filtering the input images according to HOG feature, the images of the cave target were classified. Then a self-adaptive threshold algorithm Wiblack was proposed based on the gray feature of the cave target to extract the suspected target in the image. Finally, a mathematical model of the cave target was designed, and a target discrimination algorithm was developed based on shape similarity. The target discrimination and the target profile describing were completed with circular and ellipse similarity methods. The experimental results show that the proposed algorithm possesses better recognition accuracy, higher adaptability and robustness.