Security unit circuit for improving robustness of convolutional neural network hardware accelerator

The invention discloses a safety unit circuit for improving robustness of a convolutional neural network hardware accelerator, and belongs to the technical field of basic electronic circuits. The circuit comprises an input detection unit, a correction unit and a monitoring unit, wherein the input end of the input detection unit receives feature image pixel values and detects whether abnormal pixel values appear in the feature image pixel values; the correction unit selects and outputs a corrected pixel value according to a detection result of the input detection unit; the monitoring unit monitors the input detection unit, and if multiple abnormal pixel values occur, the reasoning process is terminated. According to the circuit provided by the invention, the bit flipping attack is detected and the abnormal pixel value is corrected in the data transmission process of forward reasoning of the neural network, and the circuit has the advantages of low complexity and low hardware resource consumption. 本发明公开一种提高卷积神经