Hardware perception mixing precision quantification method and system based on greedy search

The invention provides a hardware perception mixing precision quantification method and system based on greedy search, and the method comprises the steps: carrying out the same-bit-width high-precision quantification of all layers in a neural network, carrying out the training perception quantification, and obtaining a training model, reference reasoning precision and a total operand; performing single-layer low-precision post-training quantization on each layer in the neural network, and recording the reasoning precision corresponding to each layer and the corresponding total operand; calculating single-layer sensitivity according to the reference reasoning precision and the total operand as well as the reasoning precision and the corresponding total operand corresponding to each layer; and calculating a current total operand according to the single-layer sensitivity until a preset maximum bit operation number is reached, recording quantized layers and quantization precision, and determining a mixed precisio