The invention discloses a deep learning model-oriented dynamic test method. The method comprises the following steps: S1, obtaining a picture data set and a deep learning model; S2, dividing the picture data set into a training set and a test set, and training the deep learning model by using the training sample to obtain a classification model; S3, randomly selecting pictures from the test set astest seed samples; S4, inputting the test seed samples into a classification model, if a classification result output by the classification model is inconsistent with a label of the test seed sample,taking the test seed samples as test samples, and if not, entering step S5; S5, calculating a gradient based on the cross entropy loss and the neuron coverage rate of the test seed samples in the classification model; S6, modifying the test seed samples according to a gradient rising algorithm; S7, circularly executing steps S4 to S6 until all the test seed samples become test samples and are output; and S8, evaluating the classification performance of the model by using the test samples.