System and method for compressing image based on flash in-memory computing array

A method for compressing an image based on a FLASH in-memory computing array, comprising:

• writing a weight matrix of a convolutional neural network for encoding and a weight matrix of a convolutional neural network for decoding into a FLASH in-memory computing array;
• and inputting an original image;
• encoding the original image by using the convolutional neural network for encoding based on the FLASH in-memory computing array, to obtain a feature image;
• quantifying the feature image by using a quantization module, to obtain a quantized image;
• decoding the quantized image by using the convolutional neural network for decoding based on the FLASH in-memory computing array, to obtain a compressed image;
• wherein the method further comprises:
• initializing the network, and constructing a convolutional neural network model for encoding and a convolutional neural network model for decoding;
• forward propagating the convolutional neural network model for encoding and the convolutional neural network model for decoding by using training data, and computing an network error;
• back propagating the convolutional neural network model for encoding and the convolutional neural network model for decoding, and updating weight values of the convolutional neural network model for encoding and the convolutional neural network model for decoding;
• completing the training and ending the training steps in response to the trained model meets an image compression requirement;
• returning to the forward propagation step and continuing the training in response to the trained model not meeting the image compression requirement.