Data enhancement system and method for SAR image DNN classification model back door defense

The invention discloses a data enhancement system and method for SAR image DNN classification model backdoor defense, and the method comprises the steps: generating an SAR image backdoor data set through employing a plurality of backdoor attack technologies, carrying out the variable-length coding of a data enhancement combination strategy, employing an offline automatic optimization technology, and carrying out the optimization of the backdoor data set. And obtaining an optimal data enhancement combination backdoor defense strategy giving consideration to both high defensive performance and low model performance loss, inputting the optimal data enhancement combination backdoor defense strategy into an online backdoor defensive performance evaluation module, and evaluating the test precision and attack success rate of the backdoor defense model after the optimal data enhancement combination strategy. Robust backdoor defense of the SAR image DNN classification model based on data enhancement automatic combinat