The invention discloses a ceiling earthquake damage identification method based on a convolutional neural network, belongs to the field of computer vision, and aims to solve the problem of relativelylow efficiency of earthquake damage degree investigation by professionals on site at present. The identification method comprises the steps of 1, performing normalization processing on ceiling earthquake damage pictures; 2, performing rotation and mirroring processing on the target sample; 3, loading a pre-trained AlexNet model, modifying the last full connection layer of the model, and performingback propagation by adopting a cross entropy loss function and an Adam adaptive optimization function to obtain an initially-trained AlexNet model; 4, continuing to train the model, and adjusting thelearning rate, batchsize and the like of the neural network; and 5, evaluating the test sample by using the trained AlexNet model. According to the method, the use function of the post-earthquake building can be accurately, quickly and timely evaluated only by means of a certain hardware module and in combination with the trained model, and the model accuracy is high.