Streaming image semantic segmentation method, logical integrated circuit system and electronic device

A logical integrated circuit, comprising a convolution processing module and a deconvolution processing module electrically connected to the convolution processing module, a first direct memory access module (DMA) and a second DMA;

• wherein the convolution processing module and the deconvolution processing module are firstly electrically connected to the first DMA and the second DMA respectively, and then are finally connected to a memory through an interconnection;
• wherein the convolution processing module is configured to receive an image to be semantically segmented, and perform convolution operation processing on each and every feature image block of the image to generate a piece of feature data of the each and every feature image block of the image;
• the deconvolution processing module is configured to perform deconvolution operation processing on each piece of feature data to obtain a respective image block region with the same size as the each and every feature image block corresponding to the each piece of feature data;
• determine an approximation degree of the each piece of feature data and each and every preset semantic category of a plurality of preset semantic categories, and
• classify the each piece of feature data into a preset semantic category with a highest approximation degree with the each piece of feature data, wherein the plurality of preset semantic categories are in one-to-one correspondence with a plurality of filling colors;
• and
• fill each image block region corresponding to the each piece of feature data with a filling color associated with the preset semantic category into which the each piece of feature data is classified to achieve semantic segmentation of the image,
• wherein each of the convolution processing module and the deconvolution module is implemented by a circuit;
• the logical integrated circuit further comprises a streaming lake cache, wherein the streaming lake cache is electrically connected to the convolution processing module and the deconvolution processing module, and is used by the convolution processing module and the deconvolution processing module to transmit or receive data in a time-division multiplexing manner,
• wherein the streaming lake cache is implemented by a circuit.

An image semantic segmentation system, comprising a central processor, a memory electrically connected to the central processor, and a logical integrated circuit,

• wherein the logical integrated circuit comprises:
• a convolution processing module and a deconvolution processing module electrically connected to the convolution processing module, a first direct memory access module (DMA) and a second DMA;
• wherein the convolution processing module and the deconvolution processing module are firstly electrically connected to the first DMA and the second DMA respectively, and then are finally connected to the memory through an interconnection;
• wherein the convolution processing module is configured to receive an image to be semantically segmented, and perform convolution operation processing on each and every feature image block of the image to generate a piece of feature data of the each and every feature image block of the image;
• the deconvolution processing module is configured to perform deconvolution operation processing on each piece of feature data to obtain a respective image block region with the same size as the each and every feature image block corresponding to the each piece of feature data;
• determine an approximation degree of the each piece of feature data and each and every preset semantic category of a plurality of preset semantic categories, and
• classify the each piece of feature data into a preset semantic category with a highest approximation degree with the each piece of feature data, wherein the plurality of preset semantic categories are in one-to-one correspondence with a plurality of filling colors;
• and
• fill each image block region corresponding to the each piece of feature data with a filling color associated with the preset semantic category into which the each piece of feature data is classified to achieve semantic segmentation of the image,
• wherein each of the convolution processing module and the deconvolution module is implemented by a circuit
• wherein the convolution processing module is electrically connected to the central processor and the memory, the deconvolution processing module is electrically connected to the central processor and the memory, and in a case where the logical integrated circuit further comprises a streaming lake cache, the streaming lake cache is also electrically connected to the memory, and is used by the convolution processing module and the deconvolution processing module to transmit or receive data in a time-division multiplexing manner, wherein the streaming lake cache is implemented by a circuit.