The invention discloses a structural surface crack detection method based on fusion of image features and Bayesian data, which comprises the following specific steps: A, a metal member surface video image is collected and a detection image library is built; B, image texture features are calculated by localized binarization; C, two-step support vector machine image crack scanning and collection arecarried out; and D, Bayesian data fusion and decision making are carried out. By adopting video image detection, many areas that are difficult to reach by human beings can be acquired; a computer isadopted to recognize the surface crack of a structural member, thereby greatly reducing the heavy degree of interpretation and improving the crack detection rate; the correlation with the surface light intensity of the member is good, and in comparison with the previous grayscale map or other methods, the texture recognition effects can be improved; timely early warning can be carried out on a crack disease, and a structural disease can be detected early; on the premise of keeping a high scanning speed, a radial basis function support vector machine nerve network is used to maintain a high accuracy rate; and the crack recognition accuracy is improved.