The invention relates to the technical field of computer image processing and recognition method, and in particular to a method for identifying an open flame in a cable tunnel based on SSD. A way of training a SSD network model is selected, the characteristic that a SSD network selects different levels of feature images for multi-scale detection is fully used, and the defect of poor robustness ofdetecting a small target object by the existing method is made up. By the method, an open flame area in an image can be accurately detected and positioned, and accurate result detection is completed,the placing position of a camera and the fixed-point inspection position of an inspection robot are more free, and some complicated environmental conditions can be effectively handled; the method hasthe advantages of a convolutional neural network; in the process of detecting an open flame through shooting of the robot in the tunnel, interference of factors such a complicated background image anda relatively dim tunnel environment is reduced; the anti-interference ability is relatively strong; the accuracy of open fire detection in the tunnel is improved.