Image processing method, device and system based on hybrid deep learning

The invention provides an image processing method, device and system based on hybrid deep learning. The method comprises the following steps: acquiring an image signal; carrying out convolutional neural network training on the image signal, and setting an optimization target feature as a rate distortion cost; extracting shallow network feature information of the image signals, classifying the image signals according to the shallow network feature information, and distinguishing highlighted image information or non-highlighted image information from the image signals; performing quantization transformation on the image signal by using DCT transformation; encoding the image signal after the DCT, and when the highlighted image information is encoded, adopting an intra-frame encoding mode for a head frame of the highlighted image information and adopting an inter-frame prediction encoding mode for other frames of the highlighted image information; performing corresponding DCT (Discrete Cosine Transform) inverse t