Method for rapid recognition of radar targets based on ASIC machine learning processor

The invention discloses a method for rapid recognition of radar targets based on an ASIC machine learning processor, and mainly solves the problems that the existing method uses a neural network modelto identify a radar target, resulting in a slow speed and a low energy efficiency ratio. The solution comprises the following steps of obtaining a VGG16 recognition model, quantizing the model parameters by int8, and evenly splitting the structure into N parts; importing the split model from the main processor into the coprocessor; performing distance Fourier transform and short-time Fourier transform on the millimeter wave radar data to get a micro-Doppler map of the target; importing the micro-Doppler map from the main processor into the coprocessor; using the coprocessor to perform end-to-end calculations according to the split model to obtain the probability that the targets belong to different classes, and copying the result into the main processor; and outputting the result with thehighest recognition proba